├── Music_genre_classification.ipynb
├── README.md
├── audiofile_read.py
├── filelist_GTZAN_mp3_wclasses.txt
├── keras.json
├── requirements.txt
├── rp_extract.py
└── wavio.py


/README.md:
--------------------------------------------------------------------------------
  1 | # Tutorial: Deep Learning on Music Information Retrieval
  2 | 
  3 | (c) 2017 by Thomas Lidy, TU Wien - http://ifs.tuwien.ac.at/~lidy
  4 | 
  5 | This is a set of tutorials showing how to use <b>Deep learning algorithms</b> for music analysis and retrieval problems. More specifically, we use Convolutional Neural Networks to classify (categorize) music into genres.
  6 | 
  7 | It uses Python 2.7 as the programming language with the popular [Keras](https://keras.io/) and [Theano](http://deeplearning.net/software/theano/) Deep Learning libraries underneath.
  8 | 
  9 | # Tutorials
 10 | 
 11 | For the tutorials, we use iPython / Jupyter notebook, which allows to program and execute Python code interactively in the browser.
 12 | 
 13 | ### Viewing Only
 14 | 
 15 | If you do not want to install anything, you can simply view the tutorials' content in your browser, by clicking on
 16 | the tutorial's filenames listed below in the GIT file listing (above, resp. on https://github.com/tuwien-musicir/DL_MIR_Tutorial ).
 17 | 
 18 | The tutorial will open in your browser for viewing.
 19 | 
 20 | ### Interactive Coding
 21 | 
 22 | If you want to follow the Tutorials by actually executing the code on your computer, please [install first the pre-requisites](#installation-of-pre-requisites) as described below.
 23 | 
 24 | After that, to run the tutorials go into the `DL_MIR_Tutorial` folder and start from the command line:
 25 | 
 26 | `ipython notebook` or `jupyter notebook`
 27 | 
 28 | Your web browser will open showing a list of files. Start the tutorials one after another by clicking on the following:
 29 | 
 30 | <b>Music_genre_classification.ipynb</b><br/>
 31 |    This tutorial shows how music is categorized into 1 of 10 music genres using the GTZAN music collection (see below).
 32 |    It includes audio and data preprocessing for Deep Learning and creating and training different architectures and parameters of a Convolutional Neural Network. It also includes techniques such as Batch Normalization, ReLU Activation and Dropout.
 33 | 
 34 | 
 35 | # Installation of Pre-requisites
 36 | 
 37 | ## Install Python 2.7
 38 | 
 39 | Note: On most Mac and Linux systems Python is already pre-installed. Check with `python --version` on the command line whether you have Python 2.7.x installed.
 40 | 
 41 | Otherwise install Python 2.7 from https://www.python.org/download/releases/2.7/
 42 | 
 43 | ## Install Python libraries:
 44 | 
 45 | ### Mac, Linux or Windows
 46 | 
 47 | (on Windows leave out `sudo`)
 48 | 
 49 | ```
 50 | sudo pip install ipython
 51 | ```
 52 | 
 53 | Try if you can open 
 54 | ```
 55 | ipython notebook
 56 | ```
 57 | on the command line. Otherwise try to install:
 58 | ```
 59 | sudo pip install jupyter
 60 | ```
 61 | 
 62 | Then download or clone the Tutorials from this GIT repository:
 63 | 
 64 | ```
 65 | git clone https://github.com/tuwien-musicir/DL_MIR_Tutorial.git
 66 | ```
 67 | or download https://github.com/tuwien-musicir/DL_MIR_Tutorial/archive/master.zip <br/>
 68 | unzip it and rename the folder to `DL_MIR_Tutorial`.
 69 | 
 70 | Install the remaining Python libraries needed:
 71 | 
 72 | Either by:
 73 | 
 74 | ```
 75 | sudo pip install Keras==1.2.1 Theano==0.8.2 scikit-learn>=0.17 pandas librosa
 76 | ```
 77 | 
 78 | or, if you downloaded or cloned this repository, by:
 79 | 
 80 | ```
 81 | cd DL_MIR_Tutorial
 82 | sudo pip install -r requirements.txt
 83 | ```
 84 | 
 85 | ### Install MP3 Decoder
 86 | 
 87 | If you want to use audio formats other than .wav files (e.g. .mp3, .flac, .au, .mp4), you have to install FFMPEG on you computer:
 88 | 
 89 | - Linux: install `ffmpeg`, via `sudo apt-get install ffmpeg`)
 90 |   - for Ubuntu 14.04: see http://fcorti.com/2014/04/22/ffmpeg-ubuntu-14-04-lts
 91 | - Mac: download FFMPeg for Mac: http://ffmpegmac.net or if you use brew, execute: `brew install ffmpeg`
 92 | - Windows: download FFMpeg.exe from https://github.com/tuwien-musicir/rp_extract/blob/master/bin/external/win/ffmpeg.exe
 93 | 
 94 | Make sure that the exectuable is in a PATH found by the system.
 95 | 
 96 | ## Configure Keras to use Theano
 97 | 
 98 | Since we use Theano as the Deep Learning computation backend, but Keras is configured to use TensorFlow by default, we have to change this in the `keras.json` configuration file, which is in the `.keras` folder of the user's HOME directory.
 99 | 
100 | Copy the `keras.json` included in the `DL_MIR_Tutorial` to one of the following target directories (you can overwrite an existing file):
101 | 
102 | * Windows: `C:\Users\<user>\.keras\`
103 | * Mac: `/Users/<user>/.keras`
104 | * Linux: `/home/<user>/.keras`
105 | 
106 | An alternantive is to change these 2 lines in your `keras.json` file to the following:
107 | ```
108 | {
109 |     "image_dim_ordering": "th",
110 |     "backend": "theano"
111 | }
112 | ```
113 | 
114 | See https://keras.io/backend/ for details or http://ankivil.com/installing-keras-theano-and-dependencies-on-windows-10/ for a step by step guide.
115 | 
116 | ### Optional for GPU computation
117 | 
118 | If you want to train your neural networks on your GPU, also install the following (not needed for the tutorials):
119 | 
120 | * [NVidia drivers](http://www.nvidia.com/Download/index.aspx?lang=en-us)
121 | * [CUDA](https://developer.nvidia.com/cuda-downloads)
122 | * [cuDNN](https://developer.nvidia.com/cudnn) (optional, for further speedup)
123 | 
124 | To permanently configure Keras/Theano to use the GPU place a file `.theanorc` in your home directory with the following content:
125 | 
126 | ```
127 | [global]
128 | device = gpu
129 | floatX = float32
130 | mode=FAST_RUN
131 | ```
132 | 
133 | ### Check if installed correctly
134 | 
135 | To check whether Python, Keras and Theano were installed correctly, do:
136 | 
137 | `
138 | python test_keras.py
139 | `
140 | 
141 | If everything is installed correctly, it should print `Using Theano backend.`<br/>
142 | If the GPU is configured correctly, it should also print `Using gpu device 0: GeForce GTX 980 Ti` or similar.
143 | 
144 | 
145 | # Source Credits
146 | 
147 | ## Python libraries
148 | 
149 | The following helper Python libraries are used in these tutorials:
150 | 
151 | * `audiofile_read.py` and `rp_extract.py`: by Thomas Lidy and Alexander Schindler, taken from the [RP_extract](https://github.com/tuwien-musicir/rp_extract) git repository
152 | * `wavio.py`: by Warren Weckesser
153 | 
154 | ## Data Sources
155 | 
156 | The data sets we use in the tutorials are from the following sources:
157 | 
158 | * GTZAN music genre data set:
159 | by George Tzanetakis
160 | 1000 audio files with 30 sec. each, across 10 music genres, 100 audio files each
161 | 
162 | * GTZAN music speech data set: (currently not used)
163 | by George Tzanetakis
164 | Collected for the purposes of music/speech discrimination. 128 tracks, each 30 seconds long. Each class (music or speech) has 64 examples in 22050Hz Mono 16-bit WAV audio format.
165 | 
166 | both data sets available from:
167 | http://marsyasweb.appspot.com/download/data_sets/
168 | 


--------------------------------------------------------------------------------
/audiofile_read.py:
--------------------------------------------------------------------------------
  1 | # 2015-04 by Thomas Lidy
  2 | 
  3 | # MP3 READ: mini function to decode mp3 using external program
  4 | # as there is no Python library for it, we need to use external tools (mpg123, lame, ffmpeg)
  5 | 
  6 | import os         # for calling external program for mp3 decoding
  7 | import subprocess # for subprocess calls
  8 | import tempfile
  9 | import uuid
 10 | 
 11 | # Reading WAV files
 12 | # from scipy.io import wavfile
 13 | # scipy.io.wavfile does not support 24 bit Wav files
 14 | # therefore we switch to wavio by Warren Weckesser - https://github.com/WarrenWeckesser/wavio - BSD 3-Clause License
 15 | import wavio
 16 | 
 17 | 
 18 | 
 19 | class DecoderException(Exception):
 20 |     
 21 |     def __init__(self, message, command=[], orig_error=None):
 22 | 
 23 |         # Call the base class constructor with the parameters it needs
 24 |         super(DecoderException, self).__init__(message)
 25 |         self.command        = command
 26 |         self.original_error = orig_error
 27 | 
 28 | 
 29 | 
 30 | def get_audioformat_info(input_file, split_list = True):
 31 |     import magic    # pip install python-magic
 32 |     from string import split
 33 |     info = magic.from_file(input_file)
 34 |     if split_list:
 35 |         info = split(info, ', ')
 36 |     return info
 37 | 
 38 | 
 39 | 
 40 | # Normalize integer WAV data to float in range (-1,1)
 41 | # Note that this works fine with Wav files read with Wavio
 42 | # when using scipy.io.wavfile to read Wav files, use divisor = np.iinfo(wavedata.dtype).max + 1
 43 | # but in this case it will not work with 24 bit files due to scipy scaling 24 bit up to 32bit
 44 | def normalize_wav(wavedata,samplewidth):
 45 | 
 46 |     # samplewidth in byte (i.e.: 1 = 8bit, 2 = 16bit, 3 = 24bit, 4 = 32bit)
 47 |     divisor  = 2**(8*samplewidth)/2
 48 |     wavedata = wavedata / float(divisor)
 49 |     return (wavedata)
 50 | 
 51 | 
 52 | 
 53 | def wav_read(filename,normalize=True,verbose=True,auto_resample=True):
 54 |     '''read WAV files
 55 | 
 56 |     :param filename: input filename to read from
 57 |     :param normalize: normalize the read values (usually signed integers) to range (-1,1)
 58 |     :param verbose: output some information during reading
 59 |     :param auto_resample: auto-resampling: if sample rate is different than 11, 22 or 44 kHz it will resample to 44 khZ
 60 |     :return: tuple of 3 elements: samplereate (e.g. 44100), samplewith (e.g. 2 for 16 bit) and wavedata (simple array for mono, 2-dim. array for stereo)
 61 |     '''
 62 | 
 63 |     # check if file exists
 64 |     if not os.path.exists(filename):
 65 |         raise NameError("File does not exist: " + filename)
 66 | 
 67 |     samplerate, samplewidth, wavedata = wavio.readwav(filename)
 68 | 
 69 |     if auto_resample and samplerate != 11025 and samplerate != 22050 and samplerate != 44100:
 70 |         #print original file info
 71 |         if verbose:
 72 |             print samplerate, "Hz,", wavedata.shape[1], "channel(s),", wavedata.shape[0], "samples"
 73 | 
 74 |         to_samplerate = 22050 if samplerate < 22050 else 44100
 75 |         filename2 = resample(filename, to_samplerate, normalize=True, verbose=verbose)
 76 |         samplerate, samplewidth, wavedata = wavio.readwav(filename2)
 77 |         os.remove(filename2) # delete temp file
 78 | 
 79 |     if (normalize):
 80 |         wavedata = normalize_wav(wavedata,samplewidth)
 81 | 
 82 |     return (samplerate, samplewidth, wavedata)
 83 | 
 84 | 
 85 | def get_temp_filename(suffix=None):
 86 |     
 87 |     temp_dir      = tempfile.gettempdir()
 88 |     rand_filename = str(uuid.uuid4())
 89 | 
 90 |     if suffix != None:
 91 |         rand_filename = "%s%s" % (rand_filename, suffix)
 92 |         
 93 |     return os.path.join(temp_dir, rand_filename)
 94 | 
 95 | 
 96 | def resample(filename, to_samplerate=44100, normalize=True, verbose=True):
 97 | 
 98 |     tempfile = get_temp_filename(suffix='.wav')
 99 | 
100 |     try:
101 |         cmd = ['ffmpeg','-v','1','-y','-i', filename, '-ar', str(to_samplerate), tempfile]
102 |         if verbose:
103 |             print "Resampling to", to_samplerate, "..."
104 |             #print " ".join(cmd)
105 | 
106 |         return_code = subprocess.call(cmd)  # subprocess.call takes a list of command + arguments
107 | 
108 |         if return_code != 0:
109 |             raise DecoderException("Problem appeared during resampling.", command=cmd)
110 |         #if verbose: print 'Resampled with:', " ".join(cmd)
111 | 
112 |     except OSError as e:
113 |         if os.path.exists(tempfile):
114 |             os.remove(tempfile)
115 | 
116 |         if e.errno == 2:     # probably ffmpeg binary not found
117 |             try:
118 |                 subprocess.call(cmd[0])  # check if we can just call the binary
119 |             except OSError as e:
120 |                 raise DecoderException("Decoder not found. Please install " + cmd[0], command=cmd, orig_error=e)
121 | 
122 |         raise DecoderException("Unknown problem appeared during resampling.", command=cmd, orig_error=e)
123 | 
124 |     return tempfile
125 | 
126 | 
127 | 
128 | def mp3_decode(in_filename, out_filename=None, verbose=True):
129 |     ''' mp3_decode
130 | 
131 |     decoding of MP3 files
132 | 
133 |     now handled by decode function (for parameters see there)
134 |     kept for code compatibility
135 |     '''
136 |     return decode(in_filename, out_filename, verbose)
137 | 
138 | 
139 | def decode(in_filename, out_filename=None, verbose=True, no_extension_check=False, force_mono=False, force_resampling=None):
140 |     ''' calls external decoder to convert an MP3, FLAC, AIF(F) or M4A file to a WAV file
141 | 
142 |     One of the following decoder programs must be installed on the system:
143 | 
144 |     ffmpeg: for mp3, flac, aif(f), or m4a
145 |     mpg123: for mp3
146 |     lame: for mp3
147 | 
148 |     (consider adding their path  using os.environ['PATH'] += os.pathsep + path )
149 | 
150 |     in_filename: input audio file name to process
151 |     out_filename: output filename after conversion; if omitted, the input filename is used, replacing the extension by .wav
152 |     verbose: print decoding command line information or not
153 |     no_extension_check: does not check file format extension. means that *first* specified decoder is called on ANY files type
154 |     force_mono: force mono output when decoding (works with FFMPEG only!)
155 |     force_resampling: force a target sampling rate (provided in Hz) when decoding (works with FFMPEG only!)
156 |     returns: decoder command used (without parameters)
157 |     '''
158 | 
159 |     basename, ext = os.path.splitext(in_filename)
160 |     ext = ext.lower()
161 | 
162 |     if out_filename == None:
163 |         out_filename = basename + '.wav'
164 | 
165 |     # check a number of external MP3 decoder tools whether they are available on the system
166 | 
167 |     # for subprocess.call, we prepare the commands and the arguments as a list
168 |     # cmd_list is a list of commands with their arguments, which will be iterated over to try to find each tool
169 |     # cmd_types is a list of file types supported by each command/tool
170 | 
171 |     cmd1 = ['ffmpeg','-v','1','-y','-i', in_filename] # -v adjusts log level, -y option overwrites output file, because it has been created already by tempfile before when passed
172 |     if force_resampling: cmd1.extend(['-ar',str(force_resampling)])  # add option to resample to targate Hz
173 |     if force_mono: cmd1.extend(['-ac','1'])  # add option to force to mono (1 output channel)
174 |     cmd1.append(out_filename)
175 |     cmd1_types = ['.mp3','.m4a','.aif','.aiff','.flac']
176 | 
177 |     cmd2 = ['mpg123','-q', '-w', out_filename, in_filename]
178 |     cmd2_types = ['.mp3']
179 | 
180 |     cmd3 = ['lame','--quiet','--decode', in_filename, out_filename]
181 |     cmd3_types = ['.mp3']
182 | 
183 |     cmd_list = [cmd1,cmd2,cmd3]
184 |     cmd_types = [cmd1_types,cmd2_types,cmd3_types]
185 | 
186 |     success = False
187 | 
188 |     for cmd, types in zip(cmd_list,cmd_types):
189 | 
190 |         # we decode only if the current command supports the file type that we are having (except no_extension_check is True)
191 |         if ext in types or no_extension_check:
192 |             try:
193 |                 return_code = subprocess.call(cmd)  # subprocess.call takes a list of command + arguments
194 | 
195 |                 if return_code != 0:
196 |                     raise DecoderException("Problem appeared during executing decoder. Return_code: " + str(return_code), command=cmd)
197 |                 if verbose: print 'Decoded', ext, 'with:', " ".join(cmd)
198 |                 success = True
199 | 
200 |             except OSError as e:
201 |                 if e.errno != 2: #  2 = No such file or directory (i.e. decoder not found, which we want to catch at the end below)
202 |                     raise DecoderException("Problem appeared during decoding: " + str(e), command=cmd, orig_error=e)
203 | 
204 |         if success:
205 |             break  # no need to loop further
206 | 
207 |     if not success:
208 |         commands = ", ".join( c[0] for c in cmd_list)
209 |         raise OSError("No appropriate decoder found for" + ext + " file. Check if any of these programs is on your system path: " + commands + \
210 |                        ". Otherwise install one of these and/or add them to the path using os.environ['PATH'] += os.pathsep + path.")
211 | 
212 |     return cmd[0]
213 | 
214 | def get_supported_audio_formats():
215 |     # TODO: update this list here every time a new format is added; to avoid this, make a more elegant solution getting the list of formats from where the commands are defined above
216 |     return ('.wav','.mp3','.m4a','.aif','.aiff','.flac')
217 | 
218 | 
219 | # testing decoding to memory instead of file: did NOT bring any speedup!
220 | # Also note:  sample rate and number of channels not returned with this method. can be derived with
221 | # ffprobe -v quiet -show_streams -of json <input_file>
222 | # which already converts plain text to json, but then the json needs to be parsed.
223 | def decode_to_memory(in_filename, verbose=True):
224 |     cmd1 = ['ffmpeg','-v','1','-y','-i', in_filename, "-f", "f32le", "pipe:1"]    # -v adjusts log level, -y option overwrites output file, because it has been created already by tempfile above
225 |     # "pipe:1" sends output to std_out (probably Linux only)
226 |     # original:  call = [cmd, "-v", "quiet", "-i", infile, "-f", "f32le", "-ar", str(sample_rate), "-ac", "1", "pipe:1"]
227 |     # for Windows: \\.\pipe\from_ffmpeg  # http://stackoverflow.com/questions/32157774/ffmpeg-output-pipeing-to-named-windows-pipe
228 |     cmd1_types = ('.mp3','.aif','.aiff','.m4a')
229 | 
230 |     ext = ''
231 |     if verbose: print 'Decoding', ext, 'with:', " ".join(cmd1)
232 | 
233 |     import numpy as np
234 |     decoded_wav = subprocess.check_output(cmd1)
235 |     wavedata = np.frombuffer(decoded_wav, dtype=np.float32)
236 |     return wavedata
237 | 
238 | 
239 | def mp3_read(filename,normalize=True,verbose=True):
240 |     ''' mp3_read:
241 |     call mp3_decode and read from wav file ,then delete wav file
242 |     returns samplereate (e.g. 44100), samplewith (e.g. 2 for 16 bit) and wavedata (simple array for mono, 2-dim. array for stereo)
243 |     '''
244 | 
245 |     try:
246 |         tempfile = get_temp_filename(suffix='.wav')
247 |         decode(filename,tempfile,verbose)
248 |         samplerate, samplewidth, wavedata = wav_read(tempfile,normalize,verbose)
249 | 
250 |     finally: # delete temp file
251 | 
252 |         if os.path.exists(tempfile):
253 |             os.remove(tempfile)
254 | 
255 |     return (samplerate, samplewidth, wavedata)
256 | 
257 | 
258 | 
259 | def audiofile_read(filename,normalize=True,verbose=True,include_decoder=False,no_extension_check=False,force_resampling=None):
260 |     ''' audiofile_read
261 | 
262 |     generic function capable of reading WAV, MP3 and AIF(F) files
263 | 
264 |     :param filename: file name path to audio file
265 |     :param normalize: normalize to (-1,1) if True (default), or keep original values (16 bit, 24 bit or 32 bit)
266 |     :param verbose: whether to print a message while decoding files or not
267 |     :param include_decoder: includes a 4th return value: string which decoder has been used to decode the audio file
268 |     :param no_extension_check: does not check file format via extension. means that decoder is called on ALL files.
269 |     :param force_resampling: force a target sampling rate (provided in Hz) when decoding (works with FFMPEG only!)
270 |     :return: a tuple with 3 or 4 entries: samplerate in Hz (e.g. 44100), samplewidth in bytes (e.g. 2 for 16 bit),
271 |             wavedata (simple array for mono, 2-dim. array for stereo), and optionally a decoder string
272 | 
273 |     Example:
274 |     >>> samplerate, samplewidth, wavedata = audiofile_read("music/BoxCat_Games_-_10_-_Epic_Song.mp3",verbose=False)
275 |     >>> print samplerate, "Hz,", samplewidth*8, "bit,", wavedata.shape[1], "channels,", wavedata.shape[0], "samples"
276 |     44100 Hz, 16 bit, 2 channels, 2421504 samples
277 | 
278 |     '''
279 | 
280 |     # check if file exists or has 0 bytes
281 |     if not os.path.exists(filename):
282 |         raise NameError("File does not exist: " + filename)
283 |     if os.path.getsize(filename) == 0:
284 |         raise ValueError("File has 0 bytes: " + filename)
285 | 
286 |     basename, ext = os.path.splitext(filename)
287 |     ext = ext.lower()
288 | 
289 |     if ext == '.wav' and not no_extension_check==True:
290 |         samplerate, samplewidth, wavedata = wav_read(filename,normalize,verbose)
291 |         decoder = 'wavio.py' # for log file
292 |     else:
293 |         try: # try to decode
294 |             tempfile = get_temp_filename(suffix='.wav')
295 |             decoder = decode(filename,tempfile,verbose,no_extension_check,force_resampling=force_resampling)
296 |             samplerate, samplewidth, wavedata = wav_read(tempfile,normalize,verbose)
297 | 
298 |         finally: # delete temp file in any case
299 |             if os.path.exists(tempfile):
300 |                 os.remove(tempfile)
301 | 
302 |     if include_decoder:
303 |         return samplerate, samplewidth, wavedata, decoder
304 |     else:
305 |         return samplerate, samplewidth, wavedata
306 | 
307 | 
308 | # function to self test audiofile_read if working properly
309 | def self_test():
310 |     import doctest
311 |     #doctest.testmod()
312 |     doctest.run_docstring_examples(audiofile_read, globals())
313 | 
314 | 
315 | # main routine: to test if decoding works properly
316 | 
317 | if __name__ == '__main__':
318 | 
319 |     # to run self test:
320 |     #self_test()
321 |     #exit()
322 |     # (no output means that everything went fine)
323 | 
324 |     import sys
325 | 
326 |     # if your MP3 decoder is not on the system PATH, add it like this:
327 |     # path = '/path/to/ffmpeg/'
328 |     # os.environ['PATH'] += os.pathsep + path
329 |     
330 |     # test audio file: "Epic Song" by "BoxCat Game" (included in repository)
331 |     # Epic Song by BoxCat Games is licensed under a Creative Commons Attribution License.
332 |     # http://freemusicarchive.org/music/BoxCat_Games/Nameless_the_Hackers_RPG_Soundtrack/BoxCat_Games_-_Nameless-_the_Hackers_RPG_Soundtrack_-_10_Epic_Song
333 |     if len(sys.argv) > 1:
334 |         file = sys.argv[1]
335 |     else:
336 |         file = "music/BoxCat_Games_-_10_-_Epic_Song.mp3"
337 | 
338 |     #print get_audioformat_info(file)
339 | 
340 |     # import time
341 |     # start = time.time()
342 |     samplerate, samplewidth, wavedata = audiofile_read(file)
343 |     # print time.time() - start
344 |     # print wavedata.shape
345 |     #
346 |     # start = time.time()
347 |     # wavedata2 = decode_to_memory(file)
348 |     # print time.time() - start
349 |     # print wavedata2.shape
350 |     #
351 |     # print "EQUAL" if wavedata == wavedata2 else "NOT EQUAL"
352 | 
353 |     print "Successfully read audio file:"
354 |     print samplerate, "Hz,", samplewidth*8, "bit,", wavedata.shape[1], "channels,", wavedata.shape[0], "samples"
355 | 


--------------------------------------------------------------------------------
/filelist_GTZAN_mp3_wclasses.txt:
--------------------------------------------------------------------------------
   1 | ./rock/rock.00053.mp3,rock
   2 | ./rock/rock.00051.mp3,rock
   3 | ./rock/rock.00076.mp3,rock
   4 | ./rock/rock.00084.mp3,rock
   5 | ./rock/rock.00052.mp3,rock
   6 | ./rock/rock.00057.mp3,rock
   7 | ./rock/rock.00028.mp3,rock
   8 | ./rock/rock.00035.mp3,rock
   9 | ./rock/rock.00095.mp3,rock
  10 | ./rock/rock.00088.mp3,rock
  11 | ./rock/rock.00048.mp3,rock
  12 | ./rock/rock.00009.mp3,rock
  13 | ./rock/rock.00036.mp3,rock
  14 | ./rock/rock.00032.mp3,rock
  15 | ./rock/rock.00019.mp3,rock
  16 | ./rock/rock.00042.mp3,rock
  17 | ./rock/rock.00003.mp3,rock
  18 | ./rock/rock.00061.mp3,rock
  19 | ./rock/rock.00013.mp3,rock
  20 | ./rock/rock.00038.mp3,rock
  21 | ./rock/rock.00054.mp3,rock
  22 | ./rock/rock.00014.mp3,rock
  23 | ./rock/rock.00004.mp3,rock
  24 | ./rock/rock.00091.mp3,rock
  25 | ./rock/rock.00025.mp3,rock
  26 | ./rock/rock.00086.mp3,rock
  27 | ./rock/rock.00090.mp3,rock
  28 | ./rock/rock.00046.mp3,rock
  29 | ./rock/rock.00075.mp3,rock
  30 | ./rock/rock.00026.mp3,rock
  31 | ./rock/rock.00049.mp3,rock
  32 | ./rock/rock.00093.mp3,rock
  33 | ./rock/rock.00001.mp3,rock
  34 | ./rock/rock.00099.mp3,rock
  35 | ./rock/rock.00010.mp3,rock
  36 | ./rock/rock.00081.mp3,rock
  37 | ./rock/rock.00089.mp3,rock
  38 | ./rock/rock.00016.mp3,rock
  39 | ./rock/rock.00012.mp3,rock
  40 | ./rock/rock.00069.mp3,rock
  41 | ./rock/rock.00083.mp3,rock
  42 | ./rock/rock.00043.mp3,rock
  43 | ./rock/rock.00062.mp3,rock
  44 | ./rock/rock.00073.mp3,rock
  45 | ./rock/rock.00060.mp3,rock
  46 | ./rock/rock.00056.mp3,rock
  47 | ./rock/rock.00023.mp3,rock
  48 | ./rock/rock.00077.mp3,rock
  49 | ./rock/rock.00041.mp3,rock
  50 | ./rock/rock.00015.mp3,rock
  51 | ./rock/rock.00033.mp3,rock
  52 | ./rock/rock.00027.mp3,rock
  53 | ./rock/rock.00058.mp3,rock
  54 | ./rock/rock.00094.mp3,rock
  55 | ./rock/rock.00020.mp3,rock
  56 | ./rock/rock.00017.mp3,rock
  57 | ./rock/rock.00005.mp3,rock
  58 | ./rock/rock.00031.mp3,rock
  59 | ./rock/rock.00050.mp3,rock
  60 | ./rock/rock.00024.mp3,rock
  61 | ./rock/rock.00021.mp3,rock
  62 | ./rock/rock.00018.mp3,rock
  63 | ./rock/rock.00065.mp3,rock
  64 | ./rock/rock.00087.mp3,rock
  65 | ./rock/rock.00097.mp3,rock
  66 | ./rock/rock.00000.mp3,rock
  67 | ./rock/rock.00085.mp3,rock
  68 | ./rock/rock.00082.mp3,rock
  69 | ./rock/rock.00007.mp3,rock
  70 | ./rock/rock.00002.mp3,rock
  71 | ./rock/rock.00037.mp3,rock
  72 | ./rock/rock.00059.mp3,rock
  73 | ./rock/rock.00080.mp3,rock
  74 | ./rock/rock.00039.mp3,rock
  75 | ./rock/rock.00072.mp3,rock
  76 | ./rock/rock.00022.mp3,rock
  77 | ./rock/rock.00079.mp3,rock
  78 | ./rock/rock.00047.mp3,rock
  79 | ./rock/rock.00006.mp3,rock
  80 | ./rock/rock.00008.mp3,rock
  81 | ./rock/rock.00040.mp3,rock
  82 | ./rock/rock.00096.mp3,rock
  83 | ./rock/rock.00045.mp3,rock
  84 | ./rock/rock.00029.mp3,rock
  85 | ./rock/rock.00030.mp3,rock
  86 | ./rock/rock.00064.mp3,rock
  87 | ./rock/rock.00063.mp3,rock
  88 | ./rock/rock.00074.mp3,rock
  89 | ./rock/rock.00055.mp3,rock
  90 | ./rock/rock.00068.mp3,rock
  91 | ./rock/rock.00067.mp3,rock
  92 | ./rock/rock.00044.mp3,rock
  93 | ./rock/rock.00070.mp3,rock
  94 | ./rock/rock.00066.mp3,rock
  95 | ./rock/rock.00092.mp3,rock
  96 | ./rock/rock.00098.mp3,rock
  97 | ./rock/rock.00078.mp3,rock
  98 | ./rock/rock.00034.mp3,rock
  99 | ./rock/rock.00071.mp3,rock
 100 | ./rock/rock.00011.mp3,rock
 101 | ./hiphop/hiphop.00096.mp3,hiphop
 102 | ./hiphop/hiphop.00023.mp3,hiphop
 103 | ./hiphop/hiphop.00097.mp3,hiphop
 104 | ./hiphop/hiphop.00090.mp3,hiphop
 105 | ./hiphop/hiphop.00053.mp3,hiphop
 106 | ./hiphop/hiphop.00085.mp3,hiphop
 107 | ./hiphop/hiphop.00073.mp3,hiphop
 108 | ./hiphop/hiphop.00007.mp3,hiphop
 109 | ./hiphop/hiphop.00036.mp3,hiphop
 110 | ./hiphop/hiphop.00010.mp3,hiphop
 111 | ./hiphop/hiphop.00071.mp3,hiphop
 112 | ./hiphop/hiphop.00047.mp3,hiphop
 113 | ./hiphop/hiphop.00035.mp3,hiphop
 114 | ./hiphop/hiphop.00066.mp3,hiphop
 115 | ./hiphop/hiphop.00052.mp3,hiphop
 116 | ./hiphop/hiphop.00095.mp3,hiphop
 117 | ./hiphop/hiphop.00046.mp3,hiphop
 118 | ./hiphop/hiphop.00089.mp3,hiphop
 119 | ./hiphop/hiphop.00029.mp3,hiphop
 120 | ./hiphop/hiphop.00025.mp3,hiphop
 121 | ./hiphop/hiphop.00088.mp3,hiphop
 122 | ./hiphop/hiphop.00100.mp3,hiphop
 123 | ./hiphop/hiphop.00058.mp3,hiphop
 124 | ./hiphop/hiphop.00050.mp3,hiphop
 125 | ./hiphop/hiphop.00051.mp3,hiphop
 126 | ./hiphop/hiphop.00057.mp3,hiphop
 127 | ./hiphop/hiphop.00080.mp3,hiphop
 128 | ./hiphop/hiphop.00032.mp3,hiphop
 129 | ./hiphop/hiphop.00013.mp3,hiphop
 130 | ./hiphop/hiphop.00011.mp3,hiphop
 131 | ./hiphop/hiphop.00092.mp3,hiphop
 132 | ./hiphop/hiphop.00028.mp3,hiphop
 133 | ./hiphop/hiphop.00000.mp3,hiphop
 134 | ./hiphop/hiphop.00079.mp3,hiphop
 135 | ./hiphop/hiphop.00001.mp3,hiphop
 136 | ./hiphop/hiphop.00072.mp3,hiphop
 137 | ./hiphop/hiphop.00049.mp3,hiphop
 138 | ./hiphop/hiphop.00087.mp3,hiphop
 139 | ./hiphop/hiphop.00041.mp3,hiphop
 140 | ./hiphop/hiphop.00083.mp3,hiphop
 141 | ./hiphop/hiphop.00059.mp3,hiphop
 142 | ./hiphop/hiphop.00063.mp3,hiphop
 143 | ./hiphop/hiphop.00048.mp3,hiphop
 144 | ./hiphop/hiphop.00039.mp3,hiphop
 145 | ./hiphop/hiphop.00009.mp3,hiphop
 146 | ./hiphop/hiphop.00008.mp3,hiphop
 147 | ./hiphop/hiphop.00101.mp3,hiphop
 148 | ./hiphop/hiphop.00068.mp3,hiphop
 149 | ./hiphop/hiphop.00031.mp3,hiphop
 150 | ./hiphop/hiphop.00061.mp3,hiphop
 151 | ./hiphop/hiphop.00024.mp3,hiphop
 152 | ./hiphop/hiphop.00081.mp3,hiphop
 153 | ./hiphop/hiphop.00054.mp3,hiphop
 154 | ./hiphop/hiphop.00094.mp3,hiphop
 155 | ./hiphop/hiphop.00075.mp3,hiphop
 156 | ./hiphop/hiphop.00042.mp3,hiphop
 157 | ./hiphop/hiphop.00078.mp3,hiphop
 158 | ./hiphop/hiphop.00027.mp3,hiphop
 159 | ./hiphop/hiphop.00026.mp3,hiphop
 160 | ./hiphop/hiphop.00074.mp3,hiphop
 161 | ./hiphop/hiphop.00004.mp3,hiphop
 162 | ./hiphop/hiphop.00060.mp3,hiphop
 163 | ./hiphop/hiphop.00043.mp3,hiphop
 164 | ./hiphop/hiphop.00014.mp3,hiphop
 165 | ./hiphop/hiphop.00018.mp3,hiphop
 166 | ./hiphop/hiphop.00062.mp3,hiphop
 167 | ./hiphop/hiphop.00076.mp3,hiphop
 168 | ./hiphop/hiphop.00020.mp3,hiphop
 169 | ./hiphop/hiphop.00064.mp3,hiphop
 170 | ./hiphop/hiphop.00055.mp3,hiphop
 171 | ./hiphop/hiphop.00017.mp3,hiphop
 172 | ./hiphop/hiphop.00070.mp3,hiphop
 173 | ./hiphop/hiphop.00033.mp3,hiphop
 174 | ./hiphop/hiphop.00040.mp3,hiphop
 175 | ./hiphop/hiphop.00012.mp3,hiphop
 176 | ./hiphop/hiphop.00002.mp3,hiphop
 177 | ./hiphop/hiphop.00015.mp3,hiphop
 178 | ./hiphop/hiphop.00077.mp3,hiphop
 179 | ./hiphop/hiphop.00044.mp3,hiphop
 180 | ./hiphop/hiphop.00056.mp3,hiphop
 181 | ./hiphop/hiphop.00038.mp3,hiphop
 182 | ./hiphop/hiphop.00086.mp3,hiphop
 183 | ./hiphop/hiphop.00091.mp3,hiphop
 184 | ./hiphop/hiphop.00037.mp3,hiphop
 185 | ./hiphop/hiphop.00016.mp3,hiphop
 186 | ./hiphop/hiphop.00067.mp3,hiphop
 187 | ./hiphop/hiphop.00082.mp3,hiphop
 188 | ./hiphop/hiphop.00069.mp3,hiphop
 189 | ./hiphop/hiphop.00005.mp3,hiphop
 190 | ./hiphop/hiphop.00034.mp3,hiphop
 191 | ./hiphop/hiphop.00065.mp3,hiphop
 192 | ./hiphop/hiphop.00084.mp3,hiphop
 193 | ./hiphop/hiphop.00021.mp3,hiphop
 194 | ./hiphop/hiphop.00098.mp3,hiphop
 195 | ./hiphop/hiphop.00045.mp3,hiphop
 196 | ./hiphop/hiphop.00093.mp3,hiphop
 197 | ./hiphop/hiphop.00099.mp3,hiphop
 198 | ./hiphop/hiphop.00006.mp3,hiphop
 199 | ./hiphop/hiphop.00003.mp3,hiphop
 200 | ./hiphop/hiphop.00022.mp3,hiphop
 201 | ./disco/disco.00047.mp3,disco
 202 | ./disco/disco.00066.mp3,disco
 203 | ./disco/disco.00027.mp3,disco
 204 | ./disco/disco.00046.mp3,disco
 205 | ./disco/disco.00067.mp3,disco
 206 | ./disco/disco.00052.mp3,disco
 207 | ./disco/disco.00086.mp3,disco
 208 | ./disco/disco.00026.mp3,disco
 209 | ./disco/disco.00060.mp3,disco
 210 | ./disco/disco.00096.mp3,disco
 211 | ./disco/disco.00076.mp3,disco
 212 | ./disco/disco.00058.mp3,disco
 213 | ./disco/disco.00097.mp3,disco
 214 | ./disco/disco.00002.mp3,disco
 215 | ./disco/disco.00032.mp3,disco
 216 | ./disco/disco.00024.mp3,disco
 217 | ./disco/disco.00091.mp3,disco
 218 | ./disco/disco.00006.mp3,disco
 219 | ./disco/disco.00078.mp3,disco
 220 | ./disco/disco.00092.mp3,disco
 221 | ./disco/disco.00020.mp3,disco
 222 | ./disco/disco.00073.mp3,disco
 223 | ./disco/disco.00057.mp3,disco
 224 | ./disco/disco.00072.mp3,disco
 225 | ./disco/disco.00011.mp3,disco
 226 | ./disco/disco.00065.mp3,disco
 227 | ./disco/disco.00068.mp3,disco
 228 | ./disco/disco.00033.mp3,disco
 229 | ./disco/disco.00083.mp3,disco
 230 | ./disco/disco.00085.mp3,disco
 231 | ./disco/disco.00093.mp3,disco
 232 | ./disco/disco.00036.mp3,disco
 233 | ./disco/disco.00037.mp3,disco
 234 | ./disco/disco.00071.mp3,disco
 235 | ./disco/disco.00088.mp3,disco
 236 | ./disco/disco.00090.mp3,disco
 237 | ./disco/disco.00008.mp3,disco
 238 | ./disco/disco.00098.mp3,disco
 239 | ./disco/disco.00022.mp3,disco
 240 | ./disco/disco.00001.mp3,disco
 241 | ./disco/disco.00061.mp3,disco
 242 | ./disco/disco.00041.mp3,disco
 243 | ./disco/disco.00082.mp3,disco
 244 | ./disco/disco.00074.mp3,disco
 245 | ./disco/disco.00042.mp3,disco
 246 | ./disco/disco.00012.mp3,disco
 247 | ./disco/disco.00000.mp3,disco
 248 | ./disco/disco.00080.mp3,disco
 249 | ./disco/disco.00095.mp3,disco
 250 | ./disco/disco.00051.mp3,disco
 251 | ./disco/disco.00040.mp3,disco
 252 | ./disco/disco.00069.mp3,disco
 253 | ./disco/disco.00028.mp3,disco
 254 | ./disco/disco.00016.mp3,disco
 255 | ./disco/disco.00062.mp3,disco
 256 | ./disco/disco.00014.mp3,disco
 257 | ./disco/disco.00007.mp3,disco
 258 | ./disco/disco.00087.mp3,disco
 259 | ./disco/disco.00049.mp3,disco
 260 | ./disco/disco.00029.mp3,disco
 261 | ./disco/disco.00018.mp3,disco
 262 | ./disco/disco.00081.mp3,disco
 263 | ./disco/disco.00039.mp3,disco
 264 | ./disco/disco.00025.mp3,disco
 265 | ./disco/disco.00050.mp3,disco
 266 | ./disco/disco.00059.mp3,disco
 267 | ./disco/disco.00015.mp3,disco
 268 | ./disco/disco.00004.mp3,disco
 269 | ./disco/disco.00070.mp3,disco
 270 | ./disco/disco.00079.mp3,disco
 271 | ./disco/disco.00043.mp3,disco
 272 | ./disco/disco.00094.mp3,disco
 273 | ./disco/disco.00064.mp3,disco
 274 | ./disco/disco.00099.mp3,disco
 275 | ./disco/disco.00034.mp3,disco
 276 | ./disco/disco.00075.mp3,disco
 277 | ./disco/disco.00013.mp3,disco
 278 | ./disco/disco.00077.mp3,disco
 279 | ./disco/disco.00045.mp3,disco
 280 | ./disco/disco.00023.mp3,disco
 281 | ./disco/disco.00009.mp3,disco
 282 | ./disco/disco.00035.mp3,disco
 283 | ./disco/disco.00031.mp3,disco
 284 | ./disco/disco.00003.mp3,disco
 285 | ./disco/disco.00089.mp3,disco
 286 | ./disco/disco.00010.mp3,disco
 287 | ./disco/disco.00063.mp3,disco
 288 | ./disco/disco.00019.mp3,disco
 289 | ./disco/disco.00054.mp3,disco
 290 | ./disco/disco.00030.mp3,disco
 291 | ./disco/disco.00048.mp3,disco
 292 | ./disco/disco.00055.mp3,disco
 293 | ./disco/disco.00044.mp3,disco
 294 | ./disco/disco.00053.mp3,disco
 295 | ./disco/disco.00017.mp3,disco
 296 | ./disco/disco.00056.mp3,disco
 297 | ./disco/disco.00084.mp3,disco
 298 | ./disco/disco.00038.mp3,disco
 299 | ./disco/disco.00021.mp3,disco
 300 | ./disco/disco.00005.mp3,disco
 301 | ./metal/metal.00058.mp3,metal
 302 | ./metal/metal.00053.mp3,metal
 303 | ./metal/metal.00059.mp3,metal
 304 | ./metal/metal.00043.mp3,metal
 305 | ./metal/metal.00086.mp3,metal
 306 | ./metal/metal.00006.mp3,metal
 307 | ./metal/metal.00035.mp3,metal
 308 | ./metal/metal.00028.mp3,metal
 309 | ./metal/metal.00048.mp3,metal
 310 | ./metal/metal.00020.mp3,metal
 311 | ./metal/metal.00052.mp3,metal
 312 | ./metal/metal.00071.mp3,metal
 313 | ./metal/metal.00080.mp3,metal
 314 | ./metal/metal.00091.mp3,metal
 315 | ./metal/metal.00065.mp3,metal
 316 | ./metal/metal.00025.mp3,metal
 317 | ./metal/metal.00083.mp3,metal
 318 | ./metal/metal.00013.mp3,metal
 319 | ./metal/metal.00030.mp3,metal
 320 | ./metal/metal.00019.mp3,metal
 321 | ./metal/metal.00072.mp3,metal
 322 | ./metal/metal.00050.mp3,metal
 323 | ./metal/metal.00024.mp3,metal
 324 | ./metal/metal.00060.mp3,metal
 325 | ./metal/metal.00045.mp3,metal
 326 | ./metal/metal.00026.mp3,metal
 327 | ./metal/metal.00081.mp3,metal
 328 | ./metal/metal.00005.mp3,metal
 329 | ./metal/metal.00015.mp3,metal
 330 | ./metal/metal.00051.mp3,metal
 331 | ./metal/metal.00097.mp3,metal
 332 | ./metal/metal.00016.mp3,metal
 333 | ./metal/metal.00036.mp3,metal
 334 | ./metal/metal.00057.mp3,metal
 335 | ./metal/metal.00040.mp3,metal
 336 | ./metal/metal.00011.mp3,metal
 337 | ./metal/metal.00032.mp3,metal
 338 | ./metal/metal.00087.mp3,metal
 339 | ./metal/metal.00061.mp3,metal
 340 | ./metal/metal.00085.mp3,metal
 341 | ./metal/metal.00004.mp3,metal
 342 | ./metal/metal.00009.mp3,metal
 343 | ./metal/metal.00038.mp3,metal
 344 | ./metal/metal.00075.mp3,metal
 345 | ./metal/metal.00039.mp3,metal
 346 | ./metal/metal.00031.mp3,metal
 347 | ./metal/metal.00084.mp3,metal
 348 | ./metal/metal.00070.mp3,metal
 349 | ./metal/metal.00055.mp3,metal
 350 | ./metal/metal.00062.mp3,metal
 351 | ./metal/metal.00041.mp3,metal
 352 | ./metal/metal.00023.mp3,metal
 353 | ./metal/metal.00090.mp3,metal
 354 | ./metal/metal.00069.mp3,metal
 355 | ./metal/metal.00079.mp3,metal
 356 | ./metal/metal.00001.mp3,metal
 357 | ./metal/metal.00088.mp3,metal
 358 | ./metal/metal.00018.mp3,metal
 359 | ./metal/metal.00063.mp3,metal
 360 | ./metal/metal.00049.mp3,metal
 361 | ./metal/metal.00096.mp3,metal
 362 | ./metal/metal.00056.mp3,metal
 363 | ./metal/metal.00077.mp3,metal
 364 | ./metal/metal.00002.mp3,metal
 365 | ./metal/metal.00010.mp3,metal
 366 | ./metal/metal.00003.mp3,metal
 367 | ./metal/metal.00037.mp3,metal
 368 | ./metal/metal.00017.mp3,metal
 369 | ./metal/metal.00073.mp3,metal
 370 | ./metal/metal.00078.mp3,metal
 371 | ./metal/metal.00008.mp3,metal
 372 | ./metal/metal.00076.mp3,metal
 373 | ./metal/metal.00022.mp3,metal
 374 | ./metal/metal.00092.mp3,metal
 375 | ./metal/metal.00042.mp3,metal
 376 | ./metal/metal.00033.mp3,metal
 377 | ./metal/metal.00074.mp3,metal
 378 | ./metal/metal.00094.mp3,metal
 379 | ./metal/metal.00027.mp3,metal
 380 | ./metal/metal.00054.mp3,metal
 381 | ./metal/metal.00064.mp3,metal
 382 | ./metal/metal.00068.mp3,metal
 383 | ./metal/metal.00089.mp3,metal
 384 | ./metal/metal.00044.mp3,metal
 385 | ./metal/metal.00047.mp3,metal
 386 | ./metal/metal.00066.mp3,metal
 387 | ./metal/metal.00000.mp3,metal
 388 | ./metal/metal.00067.mp3,metal
 389 | ./metal/metal.00099.mp3,metal
 390 | ./metal/metal.00093.mp3,metal
 391 | ./metal/metal.00046.mp3,metal
 392 | ./metal/metal.00082.mp3,metal
 393 | ./metal/metal.00012.mp3,metal
 394 | ./metal/metal.00014.mp3,metal
 395 | ./metal/metal.00034.mp3,metal
 396 | ./metal/metal.00021.mp3,metal
 397 | ./metal/metal.00007.mp3,metal
 398 | ./metal/metal.00098.mp3,metal
 399 | ./metal/metal.00029.mp3,metal
 400 | ./metal/metal.00100.mp3,metal
 401 | ./jazz/jazz.00026.mp3,jazz
 402 | ./jazz/jazz.00075.mp3,jazz
 403 | ./jazz/jazz.00030.mp3,jazz
 404 | ./jazz/jazz.00080.mp3,jazz
 405 | ./jazz/jazz.00042.mp3,jazz
 406 | ./jazz/jazz.00073.mp3,jazz
 407 | ./jazz/jazz.00014.mp3,jazz
 408 | ./jazz/jazz.00015.mp3,jazz
 409 | ./jazz/jazz.00007.mp3,jazz
 410 | ./jazz/jazz.00059.mp3,jazz
 411 | ./jazz/jazz.00021.mp3,jazz
 412 | ./jazz/jazz.00003.mp3,jazz
 413 | ./jazz/jazz.00091.mp3,jazz
 414 | ./jazz/jazz.00052.mp3,jazz
 415 | ./jazz/jazz.00041.mp3,jazz
 416 | ./jazz/jazz.00049.mp3,jazz
 417 | ./jazz/jazz.00088.mp3,jazz
 418 | ./jazz/jazz.00022.mp3,jazz
 419 | ./jazz/jazz.00004.mp3,jazz
 420 | ./jazz/jazz.00067.mp3,jazz
 421 | ./jazz/jazz.00043.mp3,jazz
 422 | ./jazz/jazz.00082.mp3,jazz
 423 | ./jazz/jazz.00083.mp3,jazz
 424 | ./jazz/jazz.00074.mp3,jazz
 425 | ./jazz/jazz.00077.mp3,jazz
 426 | ./jazz/jazz.00016.mp3,jazz
 427 | ./jazz/jazz.00098.mp3,jazz
 428 | ./jazz/jazz.00000.mp3,jazz
 429 | ./jazz/jazz.00096.mp3,jazz
 430 | ./jazz/jazz.00097.mp3,jazz
 431 | ./jazz/jazz.00090.mp3,jazz
 432 | ./jazz/jazz.00034.mp3,jazz
 433 | ./jazz/jazz.00028.mp3,jazz
 434 | ./jazz/jazz.00010.mp3,jazz
 435 | ./jazz/jazz.00070.mp3,jazz
 436 | ./jazz/jazz.00093.mp3,jazz
 437 | ./jazz/jazz.00008.mp3,jazz
 438 | ./jazz/jazz.00055.mp3,jazz
 439 | ./jazz/jazz.00094.mp3,jazz
 440 | ./jazz/jazz.00009.mp3,jazz
 441 | ./jazz/jazz.00085.mp3,jazz
 442 | ./jazz/jazz.00063.mp3,jazz
 443 | ./jazz/jazz.00011.mp3,jazz
 444 | ./jazz/jazz.00086.mp3,jazz
 445 | ./jazz/jazz.00047.mp3,jazz
 446 | ./jazz/jazz.00061.mp3,jazz
 447 | ./jazz/jazz.00056.mp3,jazz
 448 | ./jazz/jazz.00078.mp3,jazz
 449 | ./jazz/jazz.00054.mp3,jazz
 450 | ./jazz/jazz.00072.mp3,jazz
 451 | ./jazz/jazz.00020.mp3,jazz
 452 | ./jazz/jazz.00039.mp3,jazz
 453 | ./jazz/jazz.00019.mp3,jazz
 454 | ./jazz/jazz.00071.mp3,jazz
 455 | ./jazz/jazz.00065.mp3,jazz
 456 | ./jazz/jazz.00081.mp3,jazz
 457 | ./jazz/jazz.00035.mp3,jazz
 458 | ./jazz/jazz.00057.mp3,jazz
 459 | ./jazz/jazz.00023.mp3,jazz
 460 | ./jazz/jazz.00001.mp3,jazz
 461 | ./jazz/jazz.00068.mp3,jazz
 462 | ./jazz/jazz.00048.mp3,jazz
 463 | ./jazz/jazz.00002.mp3,jazz
 464 | ./jazz/jazz.00087.mp3,jazz
 465 | ./jazz/jazz.00053.mp3,jazz
 466 | ./jazz/jazz.00069.mp3,jazz
 467 | ./jazz/jazz.00095.mp3,jazz
 468 | ./jazz/jazz.00060.mp3,jazz
 469 | ./jazz/jazz.00089.mp3,jazz
 470 | ./jazz/jazz.00084.mp3,jazz
 471 | ./jazz/jazz.00058.mp3,jazz
 472 | ./jazz/jazz.00064.mp3,jazz
 473 | ./jazz/jazz.00006.mp3,jazz
 474 | ./jazz/jazz.00099.mp3,jazz
 475 | ./jazz/jazz.00018.mp3,jazz
 476 | ./jazz/jazz.00031.mp3,jazz
 477 | ./jazz/jazz.00044.mp3,jazz
 478 | ./jazz/jazz.00032.mp3,jazz
 479 | ./jazz/jazz.00038.mp3,jazz
 480 | ./jazz/jazz.00033.mp3,jazz
 481 | ./jazz/jazz.00051.mp3,jazz
 482 | ./jazz/jazz.00005.mp3,jazz
 483 | ./jazz/jazz.00012.mp3,jazz
 484 | ./jazz/jazz.00062.mp3,jazz
 485 | ./jazz/jazz.00029.mp3,jazz
 486 | ./jazz/jazz.00025.mp3,jazz
 487 | ./jazz/jazz.00024.mp3,jazz
 488 | ./jazz/jazz.00027.mp3,jazz
 489 | ./jazz/jazz.00092.mp3,jazz
 490 | ./jazz/jazz.00076.mp3,jazz
 491 | ./jazz/jazz.00046.mp3,jazz
 492 | ./jazz/jazz.00037.mp3,jazz
 493 | ./jazz/jazz.00036.mp3,jazz
 494 | ./jazz/jazz.00013.mp3,jazz
 495 | ./jazz/jazz.00066.mp3,jazz
 496 | ./jazz/jazz.00079.mp3,jazz
 497 | ./jazz/jazz.00017.mp3,jazz
 498 | ./jazz/jazz.00040.mp3,jazz
 499 | ./jazz/jazz.00045.mp3,jazz
 500 | ./jazz/jazz.00050.mp3,jazz
 501 | ./classical/classical.00070.mp3,classical
 502 | ./classical/classical.00006.mp3,classical
 503 | ./classical/classical.00045.mp3,classical
 504 | ./classical/classical.00065.mp3,classical
 505 | ./classical/classical.00077.mp3,classical
 506 | ./classical/classical.00044.mp3,classical
 507 | ./classical/classical.00082.mp3,classical
 508 | ./classical/classical.00053.mp3,classical
 509 | ./classical/classical.00021.mp3,classical
 510 | ./classical/classical.00092.mp3,classical
 511 | ./classical/classical.00078.mp3,classical
 512 | ./classical/classical.00054.mp3,classical
 513 | ./classical/classical.00074.mp3,classical
 514 | ./classical/classical.00000.mp3,classical
 515 | ./classical/classical.00089.mp3,classical
 516 | ./classical/classical.00024.mp3,classical
 517 | ./classical/classical.00020.mp3,classical
 518 | ./classical/classical.00017.mp3,classical
 519 | ./classical/classical.00069.mp3,classical
 520 | ./classical/classical.00026.mp3,classical
 521 | ./classical/classical.00018.mp3,classical
 522 | ./classical/classical.00029.mp3,classical
 523 | ./classical/classical.00052.mp3,classical
 524 | ./classical/classical.00094.mp3,classical
 525 | ./classical/classical.00073.mp3,classical
 526 | ./classical/classical.00013.mp3,classical
 527 | ./classical/classical.00022.mp3,classical
 528 | ./classical/classical.00056.mp3,classical
 529 | ./classical/classical.00075.mp3,classical
 530 | ./classical/classical.00007.mp3,classical
 531 | ./classical/classical.00050.mp3,classical
 532 | ./classical/classical.00047.mp3,classical
 533 | ./classical/classical.00025.mp3,classical
 534 | ./classical/classical.00083.mp3,classical
 535 | ./classical/classical.00060.mp3,classical
 536 | ./classical/classical.00014.mp3,classical
 537 | ./classical/classical.00057.mp3,classical
 538 | ./classical/classical.00058.mp3,classical
 539 | ./classical/classical.00027.mp3,classical
 540 | ./classical/classical.00091.mp3,classical
 541 | ./classical/classical.00086.mp3,classical
 542 | ./classical/classical.00068.mp3,classical
 543 | ./classical/classical.00031.mp3,classical
 544 | ./classical/classical.00038.mp3,classical
 545 | ./classical/classical.00071.mp3,classical
 546 | ./classical/classical.00098.mp3,classical
 547 | ./classical/classical.00011.mp3,classical
 548 | ./classical/classical.00061.mp3,classical
 549 | ./classical/classical.00002.mp3,classical
 550 | ./classical/classical.00030.mp3,classical
 551 | ./classical/classical.00093.mp3,classical
 552 | ./classical/classical.00087.mp3,classical
 553 | ./classical/classical.00096.mp3,classical
 554 | ./classical/classical.00037.mp3,classical
 555 | ./classical/classical.00036.mp3,classical
 556 | ./classical/classical.00095.mp3,classical
 557 | ./classical/classical.00028.mp3,classical
 558 | ./classical/classical.00019.mp3,classical
 559 | ./classical/classical.00012.mp3,classical
 560 | ./classical/classical.00035.mp3,classical
 561 | ./classical/classical.00016.mp3,classical
 562 | ./classical/classical.00010.mp3,classical
 563 | ./classical/classical.00040.mp3,classical
 564 | ./classical/classical.00043.mp3,classical
 565 | ./classical/classical.00062.mp3,classical
 566 | ./classical/classical.00059.mp3,classical
 567 | ./classical/classical.00079.mp3,classical
 568 | ./classical/classical.00088.mp3,classical
 569 | ./classical/classical.00097.mp3,classical
 570 | ./classical/classical.00051.mp3,classical
 571 | ./classical/classical.00041.mp3,classical
 572 | ./classical/classical.00085.mp3,classical
 573 | ./classical/classical.00033.mp3,classical
 574 | ./classical/classical.00004.mp3,classical
 575 | ./classical/classical.00048.mp3,classical
 576 | ./classical/classical.00049.mp3,classical
 577 | ./classical/classical.00084.mp3,classical
 578 | ./classical/classical.00023.mp3,classical
 579 | ./classical/classical.00055.mp3,classical
 580 | ./classical/classical.00003.mp3,classical
 581 | ./classical/classical.00081.mp3,classical
 582 | ./classical/classical.00080.mp3,classical
 583 | ./classical/classical.00001.mp3,classical
 584 | ./classical/classical.00032.mp3,classical
 585 | ./classical/classical.00064.mp3,classical
 586 | ./classical/classical.00063.mp3,classical
 587 | ./classical/classical.00090.mp3,classical
 588 | ./classical/classical.00009.mp3,classical
 589 | ./classical/classical.00099.mp3,classical
 590 | ./classical/classical.00066.mp3,classical
 591 | ./classical/classical.00072.mp3,classical
 592 | ./classical/classical.00039.mp3,classical
 593 | ./classical/classical.00034.mp3,classical
 594 | ./classical/classical.00046.mp3,classical
 595 | ./classical/classical.00008.mp3,classical
 596 | ./classical/classical.00005.mp3,classical
 597 | ./classical/classical.00076.mp3,classical
 598 | ./classical/classical.00042.mp3,classical
 599 | ./classical/classical.00015.mp3,classical
 600 | ./classical/classical.00067.mp3,classical
 601 | ./reggae/reggae.00073.mp3,reggae
 602 | ./reggae/reggae.00044.mp3,reggae
 603 | ./reggae/reggae.00014.mp3,reggae
 604 | ./reggae/reggae.00030.mp3,reggae
 605 | ./reggae/reggae.00068.mp3,reggae
 606 | ./reggae/reggae.00027.mp3,reggae
 607 | ./reggae/reggae.00034.mp3,reggae
 608 | ./reggae/reggae.00066.mp3,reggae
 609 | ./reggae/reggae.00038.mp3,reggae
 610 | ./reggae/reggae.00007.mp3,reggae
 611 | ./reggae/reggae.00063.mp3,reggae
 612 | ./reggae/reggae.00074.mp3,reggae
 613 | ./reggae/reggae.00090.mp3,reggae
 614 | ./reggae/reggae.00062.mp3,reggae
 615 | ./reggae/reggae.00086.mp3,reggae
 616 | ./reggae/reggae.00021.mp3,reggae
 617 | ./reggae/reggae.00033.mp3,reggae
 618 | ./reggae/reggae.00058.mp3,reggae
 619 | ./reggae/reggae.00029.mp3,reggae
 620 | ./reggae/reggae.00067.mp3,reggae
 621 | ./reggae/reggae.00023.mp3,reggae
 622 | ./reggae/reggae.00022.mp3,reggae
 623 | ./reggae/reggae.00028.mp3,reggae
 624 | ./reggae/reggae.00025.mp3,reggae
 625 | ./reggae/reggae.00024.mp3,reggae
 626 | ./reggae/reggae.00011.mp3,reggae
 627 | ./reggae/reggae.00078.mp3,reggae
 628 | ./reggae/reggae.00080.mp3,reggae
 629 | ./reggae/reggae.00057.mp3,reggae
 630 | ./reggae/reggae.00082.mp3,reggae
 631 | ./reggae/reggae.00016.mp3,reggae
 632 | ./reggae/reggae.00079.mp3,reggae
 633 | ./reggae/reggae.00039.mp3,reggae
 634 | ./reggae/reggae.00008.mp3,reggae
 635 | ./reggae/reggae.00012.mp3,reggae
 636 | ./reggae/reggae.00056.mp3,reggae
 637 | ./reggae/reggae.00059.mp3,reggae
 638 | ./reggae/reggae.00083.mp3,reggae
 639 | ./reggae/reggae.00010.mp3,reggae
 640 | ./reggae/reggae.00052.mp3,reggae
 641 | ./reggae/reggae.00095.mp3,reggae
 642 | ./reggae/reggae.00003.mp3,reggae
 643 | ./reggae/reggae.00001.mp3,reggae
 644 | ./reggae/reggae.00075.mp3,reggae
 645 | ./reggae/reggae.00065.mp3,reggae
 646 | ./reggae/reggae.00060.mp3,reggae
 647 | ./reggae/reggae.00081.mp3,reggae
 648 | ./reggae/reggae.00048.mp3,reggae
 649 | ./reggae/reggae.00019.mp3,reggae
 650 | ./reggae/reggae.00045.mp3,reggae
 651 | ./reggae/reggae.00091.mp3,reggae
 652 | ./reggae/reggae.00097.mp3,reggae
 653 | ./reggae/reggae.00037.mp3,reggae
 654 | ./reggae/reggae.00018.mp3,reggae
 655 | ./reggae/reggae.00049.mp3,reggae
 656 | ./reggae/reggae.00031.mp3,reggae
 657 | ./reggae/reggae.00017.mp3,reggae
 658 | ./reggae/reggae.00042.mp3,reggae
 659 | ./reggae/reggae.00004.mp3,reggae
 660 | ./reggae/reggae.00088.mp3,reggae
 661 | ./reggae/reggae.00085.mp3,reggae
 662 | ./reggae/reggae.00092.mp3,reggae
 663 | ./reggae/reggae.00070.mp3,reggae
 664 | ./reggae/reggae.00093.mp3,reggae
 665 | ./reggae/reggae.00084.mp3,reggae
 666 | ./reggae/reggae.00064.mp3,reggae
 667 | ./reggae/reggae.00050.mp3,reggae
 668 | ./reggae/reggae.00032.mp3,reggae
 669 | ./reggae/reggae.00046.mp3,reggae
 670 | ./reggae/reggae.00035.mp3,reggae
 671 | ./reggae/reggae.00009.mp3,reggae
 672 | ./reggae/reggae.00076.mp3,reggae
 673 | ./reggae/reggae.00089.mp3,reggae
 674 | ./reggae/reggae.00047.mp3,reggae
 675 | ./reggae/reggae.00043.mp3,reggae
 676 | ./reggae/reggae.00005.mp3,reggae
 677 | ./reggae/reggae.00071.mp3,reggae
 678 | ./reggae/reggae.00000.mp3,reggae
 679 | ./reggae/reggae.00040.mp3,reggae
 680 | ./reggae/reggae.00072.mp3,reggae
 681 | ./reggae/reggae.00013.mp3,reggae
 682 | ./reggae/reggae.00015.mp3,reggae
 683 | ./reggae/reggae.00077.mp3,reggae
 684 | ./reggae/reggae.00099.mp3,reggae
 685 | ./reggae/reggae.00096.mp3,reggae
 686 | ./reggae/reggae.00006.mp3,reggae
 687 | ./reggae/reggae.00061.mp3,reggae
 688 | ./reggae/reggae.00002.mp3,reggae
 689 | ./reggae/reggae.00094.mp3,reggae
 690 | ./reggae/reggae.00051.mp3,reggae
 691 | ./reggae/reggae.00055.mp3,reggae
 692 | ./reggae/reggae.00026.mp3,reggae
 693 | ./reggae/reggae.00036.mp3,reggae
 694 | ./reggae/reggae.00087.mp3,reggae
 695 | ./reggae/reggae.00020.mp3,reggae
 696 | ./reggae/reggae.00053.mp3,reggae
 697 | ./reggae/reggae.00069.mp3,reggae
 698 | ./reggae/reggae.00054.mp3,reggae
 699 | ./reggae/reggae.00041.mp3,reggae
 700 | ./reggae/reggae.00098.mp3,reggae
 701 | ./pop/pop.00066.mp3,pop
 702 | ./pop/pop.00002.mp3,pop
 703 | ./pop/pop.00082.mp3,pop
 704 | ./pop/pop.00093.mp3,pop
 705 | ./pop/pop.00088.mp3,pop
 706 | ./pop/pop.00025.mp3,pop
 707 | ./pop/pop.00083.mp3,pop
 708 | ./pop/pop.00061.mp3,pop
 709 | ./pop/pop.00076.mp3,pop
 710 | ./pop/pop.00069.mp3,pop
 711 | ./pop/pop.00065.mp3,pop
 712 | ./pop/pop.00053.mp3,pop
 713 | ./pop/pop.00050.mp3,pop
 714 | ./pop/pop.00058.mp3,pop
 715 | ./pop/pop.00018.mp3,pop
 716 | ./pop/pop.00052.mp3,pop
 717 | ./pop/pop.00008.mp3,pop
 718 | ./pop/pop.00094.mp3,pop
 719 | ./pop/pop.00022.mp3,pop
 720 | ./pop/pop.00028.mp3,pop
 721 | ./pop/pop.00091.mp3,pop
 722 | ./pop/pop.00090.mp3,pop
 723 | ./pop/pop.00057.mp3,pop
 724 | ./pop/pop.00005.mp3,pop
 725 | ./pop/pop.00027.mp3,pop
 726 | ./pop/pop.00011.mp3,pop
 727 | ./pop/pop.00086.mp3,pop
 728 | ./pop/pop.00074.mp3,pop
 729 | ./pop/pop.00084.mp3,pop
 730 | ./pop/pop.00099.mp3,pop
 731 | ./pop/pop.00044.mp3,pop
 732 | ./pop/pop.00024.mp3,pop
 733 | ./pop/pop.00075.mp3,pop
 734 | ./pop/pop.00095.mp3,pop
 735 | ./pop/pop.00039.mp3,pop
 736 | ./pop/pop.00045.mp3,pop
 737 | ./pop/pop.00009.mp3,pop
 738 | ./pop/pop.00034.mp3,pop
 739 | ./pop/pop.00081.mp3,pop
 740 | ./pop/pop.00042.mp3,pop
 741 | ./pop/pop.00001.mp3,pop
 742 | ./pop/pop.00004.mp3,pop
 743 | ./pop/pop.00020.mp3,pop
 744 | ./pop/pop.00068.mp3,pop
 745 | ./pop/pop.00041.mp3,pop
 746 | ./pop/pop.00089.mp3,pop
 747 | ./pop/pop.00019.mp3,pop
 748 | ./pop/pop.00012.mp3,pop
 749 | ./pop/pop.00078.mp3,pop
 750 | ./pop/pop.00051.mp3,pop
 751 | ./pop/pop.00029.mp3,pop
 752 | ./pop/pop.00033.mp3,pop
 753 | ./pop/pop.00067.mp3,pop
 754 | ./pop/pop.00055.mp3,pop
 755 | ./pop/pop.00060.mp3,pop
 756 | ./pop/pop.00097.mp3,pop
 757 | ./pop/pop.00030.mp3,pop
 758 | ./pop/pop.00015.mp3,pop
 759 | ./pop/pop.00026.mp3,pop
 760 | ./pop/pop.00092.mp3,pop
 761 | ./pop/pop.00071.mp3,pop
 762 | ./pop/pop.00096.mp3,pop
 763 | ./pop/pop.00072.mp3,pop
 764 | ./pop/pop.00064.mp3,pop
 765 | ./pop/pop.00035.mp3,pop
 766 | ./pop/pop.00007.mp3,pop
 767 | ./pop/pop.00062.mp3,pop
 768 | ./pop/pop.00000.mp3,pop
 769 | ./pop/pop.00032.mp3,pop
 770 | ./pop/pop.00037.mp3,pop
 771 | ./pop/pop.00087.mp3,pop
 772 | ./pop/pop.00063.mp3,pop
 773 | ./pop/pop.00036.mp3,pop
 774 | ./pop/pop.00046.mp3,pop
 775 | ./pop/pop.00023.mp3,pop
 776 | ./pop/pop.00049.mp3,pop
 777 | ./pop/pop.00070.mp3,pop
 778 | ./pop/pop.00043.mp3,pop
 779 | ./pop/pop.00038.mp3,pop
 780 | ./pop/pop.00080.mp3,pop
 781 | ./pop/pop.00014.mp3,pop
 782 | ./pop/pop.00040.mp3,pop
 783 | ./pop/pop.00098.mp3,pop
 784 | ./pop/pop.00013.mp3,pop
 785 | ./pop/pop.00010.mp3,pop
 786 | ./pop/pop.00047.mp3,pop
 787 | ./pop/pop.00021.mp3,pop
 788 | ./pop/pop.00048.mp3,pop
 789 | ./pop/pop.00054.mp3,pop
 790 | ./pop/pop.00085.mp3,pop
 791 | ./pop/pop.00017.mp3,pop
 792 | ./pop/pop.00031.mp3,pop
 793 | ./pop/pop.00056.mp3,pop
 794 | ./pop/pop.00006.mp3,pop
 795 | ./pop/pop.00059.mp3,pop
 796 | ./pop/pop.00077.mp3,pop
 797 | ./pop/pop.00016.mp3,pop
 798 | ./pop/pop.00079.mp3,pop
 799 | ./pop/pop.00073.mp3,pop
 800 | ./pop/pop.00003.mp3,pop
 801 | ./blues/blues.00053.mp3,blues
 802 | ./blues/blues.00039.mp3,blues
 803 | ./blues/blues.00066.mp3,blues
 804 | ./blues/blues.00011.mp3,blues
 805 | ./blues/blues.00005.mp3,blues
 806 | ./blues/blues.00000.mp3,blues
 807 | ./blues/blues.00029.mp3,blues
 808 | ./blues/blues.00086.mp3,blues
 809 | ./blues/blues.00052.mp3,blues
 810 | ./blues/blues.00051.mp3,blues
 811 | ./blues/blues.00087.mp3,blues
 812 | ./blues/blues.00034.mp3,blues
 813 | ./blues/blues.00037.mp3,blues
 814 | ./blues/blues.00045.mp3,blues
 815 | ./blues/blues.00082.mp3,blues
 816 | ./blues/blues.00028.mp3,blues
 817 | ./blues/blues.00054.mp3,blues
 818 | ./blues/blues.00032.mp3,blues
 819 | ./blues/blues.00060.mp3,blues
 820 | ./blues/blues.00081.mp3,blues
 821 | ./blues/blues.00092.mp3,blues
 822 | ./blues/blues.00030.mp3,blues
 823 | ./blues/blues.00017.mp3,blues
 824 | ./blues/blues.00036.mp3,blues
 825 | ./blues/blues.00001.mp3,blues
 826 | ./blues/blues.00004.mp3,blues
 827 | ./blues/blues.00091.mp3,blues
 828 | ./blues/blues.00093.mp3,blues
 829 | ./blues/blues.00077.mp3,blues
 830 | ./blues/blues.00064.mp3,blues
 831 | ./blues/blues.00020.mp3,blues
 832 | ./blues/blues.00084.mp3,blues
 833 | ./blues/blues.00043.mp3,blues
 834 | ./blues/blues.00035.mp3,blues
 835 | ./blues/blues.00042.mp3,blues
 836 | ./blues/blues.00022.mp3,blues
 837 | ./blues/blues.00023.mp3,blues
 838 | ./blues/blues.00059.mp3,blues
 839 | ./blues/blues.00015.mp3,blues
 840 | ./blues/blues.00072.mp3,blues
 841 | ./blues/blues.00048.mp3,blues
 842 | ./blues/blues.00047.mp3,blues
 843 | ./blues/blues.00044.mp3,blues
 844 | ./blues/blues.00078.mp3,blues
 845 | ./blues/blues.00065.mp3,blues
 846 | ./blues/blues.00056.mp3,blues
 847 | ./blues/blues.00003.mp3,blues
 848 | ./blues/blues.00019.mp3,blues
 849 | ./blues/blues.00016.mp3,blues
 850 | ./blues/blues.00097.mp3,blues
 851 | ./blues/blues.00062.mp3,blues
 852 | ./blues/blues.00040.mp3,blues
 853 | ./blues/blues.00069.mp3,blues
 854 | ./blues/blues.00006.mp3,blues
 855 | ./blues/blues.00098.mp3,blues
 856 | ./blues/blues.00057.mp3,blues
 857 | ./blues/blues.00095.mp3,blues
 858 | ./blues/blues.00099.mp3,blues
 859 | ./blues/blues.00008.mp3,blues
 860 | ./blues/blues.00013.mp3,blues
 861 | ./blues/blues.00074.mp3,blues
 862 | ./blues/blues.00096.mp3,blues
 863 | ./blues/blues.00070.mp3,blues
 864 | ./blues/blues.00088.mp3,blues
 865 | ./blues/blues.00010.mp3,blues
 866 | ./blues/blues.00038.mp3,blues
 867 | ./blues/blues.00046.mp3,blues
 868 | ./blues/blues.00073.mp3,blues
 869 | ./blues/blues.00083.mp3,blues
 870 | ./blues/blues.00007.mp3,blues
 871 | ./blues/blues.00041.mp3,blues
 872 | ./blues/blues.00049.mp3,blues
 873 | ./blues/blues.00009.mp3,blues
 874 | ./blues/blues.00090.mp3,blues
 875 | ./blues/blues.00012.mp3,blues
 876 | ./blues/blues.00085.mp3,blues
 877 | ./blues/blues.00089.mp3,blues
 878 | ./blues/blues.00050.mp3,blues
 879 | ./blues/blues.00076.mp3,blues
 880 | ./blues/blues.00080.mp3,blues
 881 | ./blues/blues.00021.mp3,blues
 882 | ./blues/blues.00061.mp3,blues
 883 | ./blues/blues.00067.mp3,blues
 884 | ./blues/blues.00063.mp3,blues
 885 | ./blues/blues.00024.mp3,blues
 886 | ./blues/blues.00031.mp3,blues
 887 | ./blues/blues.00071.mp3,blues
 888 | ./blues/blues.00027.mp3,blues
 889 | ./blues/blues.00002.mp3,blues
 890 | ./blues/blues.00014.mp3,blues
 891 | ./blues/blues.00058.mp3,blues
 892 | ./blues/blues.00025.mp3,blues
 893 | ./blues/blues.00079.mp3,blues
 894 | ./blues/blues.00094.mp3,blues
 895 | ./blues/blues.00068.mp3,blues
 896 | ./blues/blues.00026.mp3,blues
 897 | ./blues/blues.00018.mp3,blues
 898 | ./blues/blues.00055.mp3,blues
 899 | ./blues/blues.00075.mp3,blues
 900 | ./blues/blues.00033.mp3,blues
 901 | ./country/country.00021.mp3,country
 902 | ./country/country.00057.mp3,country
 903 | ./country/country.00060.mp3,country
 904 | ./country/country.00000.mp3,country
 905 | ./country/country.00006.mp3,country
 906 | ./country/country.00076.mp3,country
 907 | ./country/country.00061.mp3,country
 908 | ./country/country.00085.mp3,country
 909 | ./country/country.00091.mp3,country
 910 | ./country/country.00009.mp3,country
 911 | ./country/country.00087.mp3,country
 912 | ./country/country.00016.mp3,country
 913 | ./country/country.00005.mp3,country
 914 | ./country/country.00064.mp3,country
 915 | ./country/country.00088.mp3,country
 916 | ./country/country.00046.mp3,country
 917 | ./country/country.00097.mp3,country
 918 | ./country/country.00036.mp3,country
 919 | ./country/country.00096.mp3,country
 920 | ./country/country.00074.mp3,country
 921 | ./country/country.00070.mp3,country
 922 | ./country/country.00043.mp3,country
 923 | ./country/country.00079.mp3,country
 924 | ./country/country.00083.mp3,country
 925 | ./country/country.00038.mp3,country
 926 | ./country/country.00054.mp3,country
 927 | ./country/country.00062.mp3,country
 928 | ./country/country.00026.mp3,country
 929 | ./country/country.00086.mp3,country
 930 | ./country/country.00053.mp3,country
 931 | ./country/country.00014.mp3,country
 932 | ./country/country.00045.mp3,country
 933 | ./country/country.00001.mp3,country
 934 | ./country/country.00024.mp3,country
 935 | ./country/country.00004.mp3,country
 936 | ./country/country.00056.mp3,country
 937 | ./country/country.00002.mp3,country
 938 | ./country/country.00013.mp3,country
 939 | ./country/country.00041.mp3,country
 940 | ./country/country.00029.mp3,country
 941 | ./country/country.00037.mp3,country
 942 | ./country/country.00023.mp3,country
 943 | ./country/country.00050.mp3,country
 944 | ./country/country.00098.mp3,country
 945 | ./country/country.00081.mp3,country
 946 | ./country/country.00008.mp3,country
 947 | ./country/country.00073.mp3,country
 948 | ./country/country.00055.mp3,country
 949 | ./country/country.00092.mp3,country
 950 | ./country/country.00030.mp3,country
 951 | ./country/country.00031.mp3,country
 952 | ./country/country.00080.mp3,country
 953 | ./country/country.00007.mp3,country
 954 | ./country/country.00022.mp3,country
 955 | ./country/country.00044.mp3,country
 956 | ./country/country.00049.mp3,country
 957 | ./country/country.00048.mp3,country
 958 | ./country/country.00063.mp3,country
 959 | ./country/country.00019.mp3,country
 960 | ./country/country.00090.mp3,country
 961 | ./country/country.00072.mp3,country
 962 | ./country/country.00003.mp3,country
 963 | ./country/country.00011.mp3,country
 964 | ./country/country.00035.mp3,country
 965 | ./country/country.00066.mp3,country
 966 | ./country/country.00069.mp3,country
 967 | ./country/country.00071.mp3,country
 968 | ./country/country.00010.mp3,country
 969 | ./country/country.00047.mp3,country
 970 | ./country/country.00051.mp3,country
 971 | ./country/country.00012.mp3,country
 972 | ./country/country.00089.mp3,country
 973 | ./country/country.00099.mp3,country
 974 | ./country/country.00068.mp3,country
 975 | ./country/country.00032.mp3,country
 976 | ./country/country.00082.mp3,country
 977 | ./country/country.00020.mp3,country
 978 | ./country/country.00094.mp3,country
 979 | ./country/country.00017.mp3,country
 980 | ./country/country.00078.mp3,country
 981 | ./country/country.00052.mp3,country
 982 | ./country/country.00034.mp3,country
 983 | ./country/country.00028.mp3,country
 984 | ./country/country.00093.mp3,country
 985 | ./country/country.00040.mp3,country
 986 | ./country/country.00065.mp3,country
 987 | ./country/country.00059.mp3,country
 988 | ./country/country.00067.mp3,country
 989 | ./country/country.00027.mp3,country
 990 | ./country/country.00039.mp3,country
 991 | ./country/country.00058.mp3,country
 992 | ./country/country.00033.mp3,country
 993 | ./country/country.00077.mp3,country
 994 | ./country/country.00025.mp3,country
 995 | ./country/country.00018.mp3,country
 996 | ./country/country.00042.mp3,country
 997 | ./country/country.00075.mp3,country
 998 | ./country/country.00095.mp3,country
 999 | ./country/country.00015.mp3,country
1000 | ./country/country.00084.mp3,country
1001 | 


--------------------------------------------------------------------------------
/keras.json:
--------------------------------------------------------------------------------
1 | {
2 |     "image_dim_ordering": "th",
3 |     "epsilon": 1e-07,
4 |     "floatx": "float32",
5 |     "backend": "theano"
6 | }


--------------------------------------------------------------------------------
/requirements.txt:
--------------------------------------------------------------------------------
1 | Keras>=1.2.1,<=1.3
2 | Theano==0.8.2
3 | scikit-learn>=0.17
4 | pandas
5 | librosa
6 | matplotlib


--------------------------------------------------------------------------------
/rp_extract.py:
--------------------------------------------------------------------------------
  1 | '''
  2 | 
  3 | RP_extract: Rhythm Patterns Audio Feature Extractor
  4 | 
  5 | @author: 2014-2015 Alexander Schindler, Thomas Lidy
  6 | 
  7 | 
  8 | Re-implementation by Alexander Schindler of RP_extract for Matlab
  9 | Matlab version originally by Thomas Lidy, based on Musik Analysis Toolbox by Elias Pampalk
 10 | ( see http://ifs.tuwien.ac.at/mir/downloads.html )
 11 | 
 12 | Main function is rp_extract. See function definition and description for more information,
 13 | or example usage in main function.
 14 | 
 15 | Note: All required functions are provided by the two main scientific libraries numpy and scipy.
 16 | 
 17 | Note: In case you alter the code to use transform2mel, librosa needs to be installed: pip install librosa
 18 | '''
 19 | 
 20 | 
 21 | import numpy as np
 22 | 
 23 | from scipy import stats
 24 | from scipy.fftpack import fft
 25 | #from scipy.fftpack import rfft #  	Discrete Fourier transform of a real sequence.
 26 | from scipy import interpolate
 27 | 
 28 | # suppress numpy warnings (divide by 0 etc.)
 29 | np.set_printoptions(suppress=True)
 30 | 
 31 | # required for debugging
 32 | np.set_printoptions(precision=8,
 33 |                     threshold=10,
 34 |                     suppress=True,
 35 |                     linewidth=200,
 36 |                     edgeitems=10)
 37 | 
 38 | 
 39 | # INITIALIZATION: Constants & Mappings
 40 | 
 41 | # Bark Scale
 42 | bark = [100, 200, 300, 400, 510, 630, 770, 920, 1080, 1270, 1480, 1720, 2000, 2320, 2700, 3150, 3700, 4400, 5300, 6400, 7700, 9500, 12000, 15500]
 43 | n_bark_bands = len(bark)
 44 | 
 45 | # copy the bark vector (using [:]) and add a 0 in front (to make calculations below easier)
 46 | barks = bark[:]
 47 | barks.insert(0,0)
 48 | 
 49 | # Phone Scale
 50 | phon = [3, 20, 40, 60, 80, 100, 101]
 51 | 
 52 | # copy the bark vector (using [:]) and add a 0 in front (to make calculations below easier)
 53 | phons     = phon[:]
 54 | phons.insert(0,0)
 55 | phons     = np.asarray(phons)
 56 | 
 57 | 
 58 | # Loudness Curves
 59 | 
 60 | eq_loudness = np.array([[55,  40, 32, 24, 19, 14, 10,  6,  4,  3,  2,   2, 0,-2,-5,-4, 0,  5, 10, 14, 25, 35], 
 61 |                         [66,  52, 43, 37, 32, 27, 23, 21, 20, 20, 20,  20,19,16,13,13,18, 22, 25, 30, 40, 50], 
 62 |                         [76,  64, 57, 51, 47, 43, 41, 41, 40, 40, 40,39.5,38,35,33,33,35, 41, 46, 50, 60, 70], 
 63 |                         [89,  79, 74, 70, 66, 63, 61, 60, 60, 60, 60,  59,56,53,52,53,56, 61, 65, 70, 80, 90], 
 64 |                         [103, 96, 92, 88, 85, 83, 81, 80, 80, 80, 80,  79,76,72,70,70,75, 79, 83, 87, 95,105], 
 65 |                         [118,110,107,105,103,102,101,100,100,100,100,  99,97,94,90,90,95,100,103,105,108,115]])
 66 | 
 67 | loudn_freq = np.array([31.62, 50, 70.7, 100, 141.4, 200, 316.2, 500, 707.1, 1000, 1414, 1682, 2000, 2515, 3162, 3976, 5000, 7071, 10000, 11890, 14140, 15500])
 68 | 
 69 | # We have the loudness values for the frequencies in loudn_freq
 70 | # now we calculate in loudn_bark a matrix of loudness sensation values for the bark bands margins
 71 | 
 72 | i = 0
 73 | j = 0
 74 | 
 75 | loudn_bark = np.zeros((eq_loudness.shape[0], len(bark)))
 76 | 
 77 | for bsi in bark:
 78 | 
 79 |     while j < len(loudn_freq) and bsi > loudn_freq[j]:
 80 |         j += 1
 81 |     
 82 |     j -= 1
 83 |     
 84 |     if np.where(loudn_freq == bsi)[0].size != 0: # loudness value for this frequency already exists
 85 |         loudn_bark[:,i] = eq_loudness[:,np.where(loudn_freq == bsi)][:,0,0]
 86 |     else:
 87 |         w1 = 1 / np.abs(loudn_freq[j] - bsi)
 88 |         w2 = 1 / np.abs(loudn_freq[j + 1] - bsi)
 89 |         loudn_bark[:,i] = (eq_loudness[:,j]*w1 + eq_loudness[:,j+1]*w2) / (w1 + w2)
 90 |     
 91 |     i += 1
 92 | 
 93 | 
 94 | 
 95 | # SPECTRAL MASKING Spreading Function
 96 | # CONST_spread contains matrix of spectral frequency masking factors
 97 | 
 98 | CONST_spread = np.zeros((n_bark_bands,n_bark_bands))
 99 | 
100 | for i in range(n_bark_bands):
101 |     CONST_spread[i,:] = 10**((15.81+7.5*((i-np.arange(n_bark_bands))+0.474)-17.5*(1+((i-np.arange(n_bark_bands))+0.474)**2)**0.5)/10)
102 | 
103 | 
104 | 
105 | # UTILITY FUNCTIONS
106 | 
107 | 
108 | def nextpow2(num):
109 |     '''NextPow2
110 | 
111 |     find the next highest number to the power of 2 to a given number
112 |     and return the exponent to 2
113 |     (analogously to Matlab's nextpow2() function)
114 |     '''
115 | 
116 |     n = 2
117 |     i = 1
118 |     while n < num:
119 |         n *= 2 
120 |         i += 1
121 |     return i
122 | 
123 | 
124 | 
125 | # FFT FUNCTIONS
126 | 
127 | def periodogram(x,win,Fs=None,nfft=1024):
128 |     ''' Periodogram
129 | 
130 |     Periodogram power spectral density estimate
131 |     Note: this function was written with 1:1 Matlab compatibility in mind.
132 | 
133 |     The number of points, nfft, in the discrete Fourier transform (DFT) is the maximum of 256 or the next power of two greater than the signal length.
134 | 
135 |     :param x: time series data (e.g. audio signal), ideally length matches nfft
136 |     :param win: window function to be applied (e.g. Hanning window). in this case win expects already data points of the window to be provided.
137 |     :param Fs: sampling frequency (unused)
138 |     :param nfft: number of bins for FFT (ideally matches length of x)
139 |     :return: Periodogram power spectrum (np.array)
140 |     '''
141 | 
142 | 
143 |     #if Fs == None:
144 |     #    Fs = 2 * np.pi         # commented out because unused
145 |    
146 |     U  = np.dot(win.conj().transpose(), win) # compensates for the power of the window.
147 |     Xx = fft((x * win),nfft) # verified
148 |     P  = Xx*np.conjugate(Xx)/U
149 |     
150 |     # Compute the 1-sided or 2-sided PSD [Power/freq] or mean-square [Power].
151 |     # Also, compute the corresponding freq vector & freq units.
152 |     
153 |     # Generate the one-sided spectrum [Power] if so wanted
154 |     if nfft % 2 != 0:
155 |         select = np.arange((nfft+1)/2)  # ODD
156 |         P = P[select,:] # Take only [0,pi] or [0,pi)
157 |         P[1:-1] = P[1:-1] * 2 # Only DC is a unique point and doesn't get doubled
158 |     else:
159 |         #select = np.arange(nfft/2+1);    # EVEN
160 |         #P = P[select,:]         # Take only [0,pi] or [0,pi) # TODO: why commented out?
161 |         P[1:-2] = P[1:-2] * 2
162 | 
163 |     P = P / (2 * np.pi)
164 | 
165 |     return P
166 | 
167 | 
168 | 
169 | 
170 | def calc_spectrogram(wavsegment,fft_window_size,fft_overlap = 0.5,real_values=True):
171 | 
172 |     ''' Calc_Spectrogram
173 | 
174 |     calculate spectrogram using periodogram function (which performs FFT) to convert wave signal data
175 |     from time to frequency domain (applying a Hanning window and (by default) 50 % window overlap)
176 | 
177 |     :param wavsegment: audio wave file data for a segment to be analyzed (mono (i.e. 1-dimensional vector) only
178 |     :param fft_window_size: windows size to apply FFT to
179 |     :param fft_overlap: overlap to apply during FFT analysis in % fraction (e.g. default = 0.5, means 50% overlap)
180 |     :param real_values: if True, return real values by taking abs(spectrogram), if False return complex values
181 |     :return: spectrogram matrix as numpy array (fft_window_size, n_frames)
182 |     '''
183 | 
184 |     # hop_size (increment step in samples, determined by fft_window_size and fft_overlap)
185 |     hop_size = int(fft_window_size*(1-fft_overlap))
186 | 
187 |     # this would compute the segment length, but it's pre-defined above ...
188 |     # segment_size = fft_window_size + (frames-1) * hop_size
189 |     # ... therefore we convert the formula to give the number of frames needed to iterate over the segment:
190 |     n_frames = (wavsegment.shape[0] - fft_window_size) / hop_size + 1
191 |     # n_frames_old = wavsegment.shape[0] / fft_window_size * 2 - 1  # number of iterations with 50% overlap
192 | 
193 |     # TODO: provide this as parameter for better caching?
194 |     han_window = np.hanning(fft_window_size) # verified
195 | 
196 |     # initialize result matrix for spectrogram
197 |     spectrogram  = np.zeros((fft_window_size, n_frames), dtype=np.complex128)
198 | 
199 |     # start index for frame-wise iteration
200 |     ix = 0
201 | 
202 |     for i in range(n_frames): # stepping through the wave segment, building spectrum for each window
203 |         spectrogram[:,i] = periodogram(wavsegment[ix:ix+fft_window_size], win=han_window,nfft=fft_window_size)
204 |         ix = ix + hop_size
205 | 
206 |         # NOTE: tested scipy periodogram BUT it delivers totally different values AND takes 2x the time of our periodogram function (0.13 sec vs. 0.06 sec)
207 |         # from scipy.signal import periodogram # move on top
208 |         #f,  spec = periodogram(x=wavsegment[idx],fs=samplerate,window='hann',nfft=fft_window_size,scaling='spectrum',return_onesided=True)
209 | 
210 |     if real_values: spectrogram = np.abs(spectrogram)
211 | 
212 |     return (spectrogram)
213 | 
214 | 
215 | # FEATURE FUNCTIONS
216 | 
217 | def calc_statistical_features(matrix):
218 | 
219 |     result = np.zeros((matrix.shape[0],7))
220 |     
221 |     result[:,0] = np.mean(matrix, axis=1)
222 |     result[:,1] = np.var(matrix, axis=1, dtype=np.float64) # the values for variance differ between MATLAB and Numpy!
223 |     result[:,2] = stats.skew(matrix, axis=1)
224 |     result[:,3] = stats.kurtosis(matrix, axis=1, fisher=False) # Matlab calculates Pearson's Kurtosis
225 |     result[:,4] = np.median(matrix, axis=1)
226 |     result[:,5] = np.min(matrix, axis=1)
227 |     result[:,6] = np.max(matrix, axis=1)
228 |     
229 |     result[np.where(np.isnan(result))] = 0
230 |     
231 |     return result
232 | 
233 | 
234 | # PSYCHO-ACOUSTIC TRANSFORMS as individual functions
235 | 
236 | 
237 | # Transform 2 Mel Scale: NOT USED by rp_extract, but included for testing purposes or for import into other programs
238 | 
239 | def transform2mel(spectrogram,samplerate,fft_window_size,n_mel_bands = 80,freq_min = 0,freq_max = None):
240 |     '''Transform to Mel
241 | 
242 |     convert a spectrogram to a Mel scale spectrogram by grouping original frequency bins
243 |     to Mel frequency bands (using Mel filter from Librosa)
244 | 
245 |     Parameters
246 |     spectrogram: input spectrogram
247 |     samplerate: samplerate of audio signal
248 |     fft_window_size: number of time window / frequency bins in the FFT analysis
249 |     n_mel_bands: number of desired Mel bands, typically 20, 40, 80 (max. 128 which is default when 'None' is provided)
250 |     freq_min: minimum frequency (Mel filters will be applied >= this frequency, but still return n_meld_bands number of bands)
251 |     freq_max: cut-off frequency (Mel filters will be applied <= this frequency, but still return n_meld_bands number of bands)
252 | 
253 |     Returns:
254 |     mel_spectrogram: Mel spectrogram: np.array of shape(n_mel_bands,frames) maintaining the number of frames in the original spectrogram
255 |     '''
256 | 
257 |     from librosa.filters import mel
258 | 
259 |     # Syntax: librosa.filters.mel(sr, n_fft, n_mels=128, fmin=0.0, fmax=None, htk=False)
260 |     mel_basis = mel(samplerate,fft_window_size, n_mels=n_mel_bands,fmin=freq_min,fmax=freq_max)
261 | 
262 |     freq_bin_max = mel_basis.shape[1] # will be fft_window_size / 2 + 1
263 | 
264 |     # IMPLEMENTATION WITH FOR LOOP
265 |     # initialize Mel Spectrogram matrix
266 |     #n_mel_bands = mel_basis.shape[0]  # get the number of bands from result in case 'None' was specified as parameter
267 |     #mel_spectrogram = np.empty((n_mel_bands, frames))
268 | 
269 |     #for i in range(frames): # stepping through the wave segment, building spectrum for each window
270 |     #    mel_spectrogram[:,i] = np.dot(mel_basis,spectrogram[0:freq_bin_max,i])
271 | 
272 |     # IMPLEMENTATION WITH DOT PRODUCT (15% faster)
273 |     # multiply the mel filter of each band with the spectogram frame (dot product executes it on all frames)
274 |     # filter will be adapted in a way so that frequencies beyond freq_max will be discarded
275 |     mel_spectrogram = np.dot(mel_basis,spectrogram[0:freq_bin_max,:])
276 |     return (mel_spectrogram)
277 | 
278 | 
279 | 
280 | 
281 | # Bark Transform: Convert Spectrogram to Bark Scale
282 | # matrix: Spectrogram values as returned from periodogram function
283 | # freq_axis: array of frequency values along the frequency axis
284 | # max_bands: limit number of Bark bands (1...24) (counting from lowest band)
285 | def transform2bark(matrix, freq_axis, max_bands=None):
286 | 
287 |     # barks and n_bark_bands have been initialized globally above
288 | 
289 |     if max_bands == None:
290 |         max_band = n_bark_bands
291 |     else:
292 |         max_band = min(n_bark_bands,max_bands)
293 | 
294 |     matrix_out = np.zeros((max_band,matrix.shape[1]),dtype=matrix.dtype)
295 | 
296 |     for b in range(max_band-1):
297 |         # TODO: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 1024 but corresponding boolean dimension is 513
298 |         matrix_out[b] = np.sum(matrix[((freq_axis >= barks[b]) & (freq_axis < barks[b+1]))], axis=0)
299 | 
300 |     return(matrix_out)
301 | 
302 | # Spectral Masking (assumes values are arranged in <=24 Bark bands)
303 | def do_spectral_masking(matrix):
304 | 
305 |     n_bands = matrix.shape[0]
306 | 
307 |     # CONST_spread has been initialized globally above
308 |     spread = CONST_spread[0:n_bands,0:n_bands] # not sure if column limitation is right here; was originally written for n_bark_bands = 24 only
309 |     matrix = np.dot(spread, matrix)
310 |     return(matrix)
311 | 
312 | # Map to Decibel Scale
313 | def transform2db(matrix):
314 |     '''Map to Decibel Scale'''
315 |     matrix[np.where(matrix < 1)] = 1
316 |     matrix = 10 * np.log10(matrix)
317 |     return(matrix)
318 | 
319 | # Transform to Phon (assumes matrix is in dB scale)
320 | def transform2phon(matrix):
321 | 
322 |     old_npsetting = np.seterr(invalid='ignore') # avoid 'RuntimeWarning: invalid value encountered in divide' at ifac division below
323 | 
324 |     # number of bark bands, matrix length in time dim
325 |     n_bands = matrix.shape[0]
326 |     t       = matrix.shape[1]
327 | 
328 |     # DB-TO-PHON BARK-SCALE-LIMIT TABLE
329 |     # introducing 1 level more with level(1) being infinite
330 |     # to avoid (levels - 1) producing errors like division by 0
331 | 
332 |     #%%table_dim = size(CONST_loudn_bark,2);
333 |     table_dim = n_bands; # OK
334 |     cbv       = np.concatenate((np.tile(np.inf,(table_dim,1)), loudn_bark[:,0:n_bands].transpose()),1) # OK
335 | 
336 |     # init lowest level = 2
337 |     levels = np.tile(2,(n_bands,t)) # OK
338 | 
339 |     for lev in range(1,6): # OK
340 |         db_thislev = np.tile(np.asarray([cbv[:,lev]]).transpose(),(1,t))
341 |         levels[np.where(matrix > db_thislev)] = lev + 2
342 | 
343 |     # the matrix 'levels' stores the correct Phon level for each data point
344 |     cbv_ind_hi = np.ravel_multi_index(dims=(table_dim,7), multi_index=np.array([np.tile(np.array([range(0,table_dim)]).transpose(),(1,t)), levels-1]), order='F')
345 |     cbv_ind_lo = np.ravel_multi_index(dims=(table_dim,7), multi_index=np.array([np.tile(np.array([range(0,table_dim)]).transpose(),(1,t)), levels-2]), order='F')
346 | 
347 |     # interpolation factor % OPT: pre-calc diff
348 |     ifac = (matrix[:,0:t] - cbv.transpose().ravel()[cbv_ind_lo]) / (cbv.transpose().ravel()[cbv_ind_hi] - cbv.transpose().ravel()[cbv_ind_lo])
349 | 
350 |     ifac[np.where(levels==2)] = 1 # keeps the upper phon value;
351 |     ifac[np.where(levels==8)] = 1 # keeps the upper phon value;
352 | 
353 |     # phons has been initialized globally above
354 | 
355 |     matrix[:,0:t] = phons.transpose().ravel()[levels - 2] + (ifac * (phons.transpose().ravel()[levels - 1] - phons.transpose().ravel()[levels - 2])) # OPT: pre-calc diff
356 | 
357 |     np.seterr(invalid=old_npsetting['invalid']) # restore RuntimeWarning setting for np division error
358 | 
359 |     return(matrix)
360 | 
361 | 
362 | # Transform to Sone scale (assumes matrix is in Phon scale)
363 | def transform2sone(matrix):
364 |     idx     = np.where(matrix >= 40)
365 |     not_idx = np.where(matrix < 40)
366 | 
367 |     matrix[idx]     =  2**((matrix[idx]-40)/10)    #
368 |     matrix[not_idx] =  (matrix[not_idx]/40)**2.642 # max => 438.53
369 |     return(matrix)
370 | 
371 | 
372 | # MAIN Rhythm Pattern Extraction Function
373 | 
374 | def rp_extract( wavedata,                          # pcm (wav) signal data normalized to (-1,1)
375 |                 samplerate,                    # signal sampling rate
376 | 
377 |                 # which features to extract
378 |                 extract_rp   = False,          # extract Rhythm Patterns features
379 |                 extract_ssd  = False,          # extract Statistical Spectrum Descriptor
380 |                 extract_tssd = False,          # extract temporal Statistical Spectrum Descriptor
381 |                 extract_rh   = False,          # extract Rhythm Histogram features
382 |                 extract_rh2  = False,          # extract Rhythm Histogram features including Fluctuation Strength Weighting
383 |                 extract_trh  = False,          # extract temporal Rhythm Histogram features
384 |                 extract_mvd  = False,          # extract modulation variance descriptor
385 | 
386 |                 # processing options
387 |                 skip_leadin_fadeout =  1,      # >=0  how many sample windows to skip at the beginning and the end
388 |                 step_width          =  1,      # >=1  each step_width'th sample window is analyzed
389 |                 n_bark_bands        = 24,      # 2..24 number of desired Bark bands (from low frequencies to high) (e.g. 15 or 20 or 24 for 11, 22 and 44 kHz audio respectively) (1 delivers undefined output)
390 |                 mod_ampl_limit      = 60,      # 2..257 number of modulation frequencies on x-axis
391 |                 
392 |                 # enable/disable parts of feature extraction
393 |                 transform_bark                 = True,  # [S2] transform to Bark scale
394 |                 spectral_masking               = True,  # [S3] compute Spectral Masking
395 |                 transform_db                   = True,  # [S4] transfrom to dB: advisable only to turn off when [S5] and [S6] are turned off too
396 |                 transform_phon                 = True,  # [S5] transform to Phon: if disabled, Sone_transform will be disabled too
397 |                 transform_sone                 = True,  # [S6] transform to Sone scale (only applies if transform_phon = True)
398 |                 fluctuation_strength_weighting = True,  # [R2] apply Fluctuation Strength weighting curve
399 |                 #blurring                       = True  # [R3] Gradient+Gauss filter # TODO: not yet implemented
400 | 
401 |                 return_segment_features = False,     # this will return features per each analyzed segment instead of aggregated ones
402 |                 verbose = True                       # print messages whats going on
403 |                 ):
404 | 
405 |     '''Rhythm Pattern Feature Extraction
406 | 
407 |     performs segment-wise audio feature extraction from provided audio wave (PCM) data
408 |     and extracts the following features:
409 | 
410 |         Rhythm Pattern
411 |         Statistical Spectrum Descriptor
412 |         Statistical Histogram
413 |         temporal Statistical Spectrum Descriptor
414 |         Rhythm Histogram
415 |         temporal Rhythm Histogram features
416 |         Modulation Variance Descriptor
417 | 
418 |     Examples:
419 |     >>> from audiofile_read import *
420 |     >>> samplerate, samplewidth, wavedata = audiofile_read("music/BoxCat_Games_-_10_-_Epic_Song.mp3") #doctest: +ELLIPSIS
421 |     Decoded .mp3 with: mpg123 -q -w /....wav music/BoxCat_Games_-_10_-_Epic_Song.mp3
422 |     >>> feat = rp_extract(wavedata, samplerate, extract_rp=True, extract_ssd=True, extract_rh=True)
423 |     Analyzing 7 segments
424 |     >>> for k in feat.keys():
425 |     ...     print k.upper() +  ":", feat[k].shape[0], "dimensions"
426 |     SSD: 168 dimensions
427 |     RH: 60 dimensions
428 |     RP: 1440 dimensions
429 |     >>> print feat["rp"]
430 |     [ 0.01599218  0.01979605  0.01564305  0.01674175  0.00959912  0.00931604  0.00937831  0.00709122  0.00929631  0.00754473 ...,  0.02998088  0.03602739  0.03633861  0.03664331  0.02589753  0.02110256
431 |       0.01457744  0.01221825  0.0073788   0.00164668]
432 |     >>> print feat["rh"]
433 |     [  7.11614842  12.58303013   6.96717295   5.24244146   6.49677561   4.21249659  12.43844045   4.19672357   5.30714983   6.1674115  ...,   1.55870044   2.69988854   2.75075831   3.67269877  13.0351257
434 |       11.7871738    3.76106713   2.45225195   2.20457928   2.06494926]
435 |     >>> print feat["ssd"]
436 |     [  3.7783279    5.84444695   5.58439197   4.87849697   4.14983056   4.09638223   4.04971225   3.96152261   3.65551062   3.2857232  ...,  14.45953191  14.6088727   14.03351539  12.84783095  10.81735946
437 |        9.04121124   7.13804008   5.6633501    3.09678286   0.52076428]
438 | 
439 |     '''
440 | 
441 | 
442 |     # PARAMETER INITIALIZATION
443 |     # non-exhibited parameters
444 |     include_DC = False
445 |     FLATTEN_ORDER = 'F' # order how matrices are flattened to vector: 'F' for Matlab/Fortran, 'C' for C order (IMPORTANT TO USE THE SAME WHEN reading+reshaping the features)
446 | 
447 |     # segment_size should always be ~6 sec, fft_window_size should always be ~ 23ms
448 | 
449 |     if (samplerate == 11025):
450 |         segment_size    = 2**16
451 |         fft_window_size = 256
452 |     elif (samplerate == 22050):
453 |         segment_size    = 2**17
454 |         fft_window_size = 512
455 |     elif (samplerate == 44100):
456 |         segment_size    = 2**18
457 |         fft_window_size = 1024
458 |     else:
459 |         # throw error not supported
460 |         raise ValueError('A sample rate of ' + str(samplerate) + " is not supported (only 11, 22 and 44 kHz).")
461 |     
462 |     # calculate frequency values on y-axis (for Bark scale calculation):
463 |     # freq_axis = float(samplerate)/fft_window_size * np.arange(0,(fft_window_size/2) + 1)
464 |     # linear space from 0 to samplerate/2 in (fft_window_size/2+1) steps
465 |     freq_axis = np.linspace(0, float(samplerate)/2, int(fft_window_size//2) + 1, endpoint=True)
466 | 
467 | 
468 |     # CONVERT STEREO TO MONO: Average the channels
469 |     if wavedata.ndim > 1:                    # if we have more than 1 dimension
470 |         if wavedata.shape[1] == 1:           # check if 2nd dimension is just 1
471 |             wavedata = wavedata[:,0]         # then we take first and only channel
472 |         else:
473 |             wavedata = np.mean(wavedata, 1)  # otherwise we average the signals over the channels
474 | 
475 | 
476 |     # SEGMENT INITIALIZATION
477 |     # find positions of wave segments
478 |     
479 |     skip_seg = skip_leadin_fadeout
480 |     seg_pos  = np.array([1, segment_size]) # array with 2 entries: start and end position of selected segment
481 | 
482 |     seg_pos_list = []  # list to store all the individual segment positions (only when return_segment_features == True)
483 | 
484 |     # if file is too small, don't skip leadin/fadeout and set step_width to 1
485 |     if ((skip_leadin_fadeout > 0) or (step_width > 1)):
486 | 
487 |         duration =  wavedata.shape[0]/samplerate
488 | 
489 |         if (duration < 45):
490 |             step_width = 1
491 |             skip_seg   = 0
492 |             # TODO: do this as a warning?
493 |             if verbose: print "Duration < 45 seconds: setting step_width to 1 and skip_leadin_fadeout to 0."
494 | 
495 |         else:
496 |             # advance by number of skip_seg segments (i.e. skip lead_in)
497 |             seg_pos = seg_pos + segment_size * skip_seg
498 |     
499 |     # calculate number of segments
500 |     n_segments = int(np.floor( (np.floor( (wavedata.shape[0] - (skip_seg*2*segment_size)) / segment_size ) - 1 ) / step_width ) + 1)
501 |     if verbose: print "Analyzing", n_segments, "segments"
502 | 
503 |     if n_segments == 0:
504 |         raise ValueError("Not enough data to analyze! Minimum sample length needs to be " +
505 |                          str(segment_size) + " (5.94 seconds) but it is " + str(wavedata.shape[0]) +
506 |                          " (" + str(round(wavedata.shape[0] * 1.0 / samplerate,2)) + " seconds)")
507 | 
508 |     # initialize output
509 |     features = {}
510 | 
511 |     ssd_list = []
512 |     sh_list = []
513 |     rh_list  = []
514 |     rh2_list = []
515 |     rp_list  = []
516 |     mvd_list = []
517 | 
518 |     hearing_threshold_factor = 0.0875 * (2**15)
519 | 
520 |     # SEGMENT ITERATION
521 | 
522 |     for seg_id in range(n_segments):
523 | 
524 |         # keep track of segment position
525 |         if return_segment_features:
526 |             seg_pos_list.append(seg_pos)
527 |         
528 |         # EXTRACT WAVE SEGMENT that will be processed
529 |         # data is assumed to be mono waveform
530 |         wavsegment = wavedata[seg_pos[0]-1:seg_pos[1]] # verified
531 |         
532 |         # v210715
533 |         # Python : [-0.0269165  -0.02128601 -0.01864624 -0.01893616 -0.02166748 -0.02694702 -0.03457642 -0.04333496 -0.05166626 -0.05891418]
534 |         # Matlab : [-0,0269165  -0,02125549 -0,01861572 -0,01893616 -0,02165222 -0,02694702 -0,03456115 -0,04331970 -0,05166626 -0,05891418]
535 | 
536 |         
537 |         # adjust hearing threshold # TODO: move after stereo-mono conversion above?
538 |         wavsegment = wavsegment * hearing_threshold_factor
539 | 
540 |         # v210715
541 |         # Python : [ -77.175    -61.03125  -53.4625   -54.29375  -62.125    -77.2625   -99.1375  -124.25    -148.1375  -168.91875]
542 |         # Matlab : [ -77,175    -60,94375  -53,3750   -54,29375  -62,081    -77,2625   -99,0938  -124,21    -148,1375  -168,91875]        
543 | 
544 |         matrix = calc_spectrogram(wavsegment,fft_window_size)
545 | 
546 |         # v210715
547 |         #Python:   0.01372537     0.51454915    72.96077581   84.86663379   2.09940049    3.29631279   97373.2756834      23228.2065494       2678.44451741     30467.235416   
548 |         #      :  84.50635406    58.32826049  1263.82538188  234.11858349  85.48176796   97.26094525  214067.91208223   3570917.53366476   2303291.96676741   1681002.94519665 
549 |         #      : 171.47168402  1498.04129116  3746.45491915  153.01444364  37.20801758  177.74229702  238810.1975412    3064388.50572536   5501187.79635479   4172009.81345923
550 |                                                                          
551 |         #Matlab:   0,01528259     0,49653179    73,32978523   85,38774541   2,00416767    3,36618763   97416,24267209     23239,84650814      2677,01521862     30460,9231041364
552 |         #      :  84,73805309    57,84524803  1263,40594029  235,62185973  85,13826606   97,61122652  214078,02415144   3571346,74831746   2303286,74666381   1680967,41922679
553 |         #      : 170,15377915  1500,98052242  3744,98456435  154,14108817  36,69362260  177,48982263  238812,02171250   3064642,99278220   5501230,26588318   4172058,72803277
554 |         #
555 | 
556 |         # PSYCHO-ACOUSTIC TRANSFORMS
557 | 
558 |         # Map to Bark Scale
559 |         if transform_bark:
560 |             matrix = transform2bark(matrix,freq_axis,n_bark_bands)
561 | 
562 |         # v210715
563 |         # Python:    255.991763   1556.884100   5083.2410768    471.9996609   124.789186   278.299555  550251.385306   6658534.245939   7807158.207639  5883479.99407189 
564 |         #       :  77128.354925  10446.109041  22613.8525735  13266.2502432  2593.395039  1367.697057  675114.554043  23401741.536499   6300109.471193  8039710.71759598 
565 |         #       : 127165.795400  91270.354107  15240.3501050  16291.2234730  1413.851495  2166.723800  868138.817452  20682384.237884   8971171.605009  5919089.97818692 
566 | 
567 |         # Matlab:    254,907114   1559,322302    5081,720289    475,1506933   123,836056    278,46723  550306,288536   6659229,587607   7807194,027765   5883487,07036370
568 |         #       :  77118,196343  10447,961479   22605,559124  13266,4432995  2591,064037   1368,48462  675116,996782  23400723,570438   6300124,132022   8039688,83884099
569 |         #       : 127172,560642  91251,040768   15246,639683  16286,4542687  1414,053166   2166,42874  868063,055613  20681863,052695   8971108,607811   5919136,16752791
570 | 
571 | 
572 |         # Spectral Masking
573 |         if spectral_masking:
574 |             matrix = do_spectral_masking(matrix)
575 |             
576 |         # v210715
577 |         # Python:  12978.051641    3416.109125   8769.913963   2648.888265   547.12360   503.50224   660888.17361  10480839.33617   8840234.405272   7193404.23970964 
578 |         #       : 100713.471006   27602.656332  27169.741240  16288.350176  2887.60281  1842.05959  1021358.42618  29229962.41626  10653981.441005  11182818.62910279 
579 |         #       : 426733.607945  262537.326945  43522.106075  41091.381283  4254.39289  4617.45877  1315036.85377  31353824.35688  12417010.121754   9673923.23590653 
580 |         
581 |         # Matlab:  12975,335615     3418,81282   8767,062187   2652,061105   545,79379   503,79683   660943,32199  10481368,76411   8840272,477464   7193407,85259461
582 |         #       : 100704,175421    27602,34142  27161,901160  16288,924458  2884,94883  1842,86020  1021368,99046  29229118,99738  10653999,341989  11182806,7524195
583 |         #       : 426751,992198   262523,89306  43524,970883  41085,415594  4253,42029  4617,35691  1314966,73269  31353021,99155  12416968,806879   9673951,88376021
584 | 
585 | 
586 |         # Map to Decibel Scale
587 |         if transform_db:
588 |             matrix = transform2db(matrix)
589 | 
590 |         # v210715
591 |         # Python: 41.13209498  35.33531736  39.42995333  34.23063639  27.38085455  27.02001413  58.2012798   70.20396064  69.46463781  68.56934467 
592 |         #       : 50.03087564  44.40950878  44.34085502  42.11877097  34.60537456  32.65303677  60.09178176  74.65828257  70.27511936  70.48551281 
593 |         #       : 56.30156848  54.19191059  46.38709903  46.1375074   36.28837595  36.64403027  61.18937924  74.96290521  70.94017035  69.85602637 
594 |         
595 |         # Matlab: 41,13118599  35,33875324  39,42854087  34,23583526  27,37028596  27,02255437  58,20164218  70,20418000  69,46465651  68,56934684
596 |         #       : 50,03047477  44,40945923  44,33960164  42,11892409  34,60138115  32,65492392  60,09182668  74,65815725  70,27512665  70,48550820
597 |         #       : 56,30175557  54,19168835  46,38738489  46,13687684  36,28738298  36,64393446  61,18914765  74,96279407  70,94015590  69,85603922
598 | 
599 |         
600 |         # Transform Phon
601 |         if transform_phon:
602 |             matrix = transform2phon(matrix)
603 | 
604 |         # v210715
605 |         # Python: 25.90299283  17.82310731  23.4713619   16.37852452   7.42111749   6.94924924  47.58029453  60.22662293  59.43646085  58.49404702 
606 |         #       : 47.03087564  41.40950878  41.34085502  38.89846372  29.5067182   27.06629597  57.09178176  71.65828257  67.27511936  67.48551281 
607 |         #       : 55.02273887  52.91308099  45.10826943  44.8586778   34.3678058   34.769195    59.91054964  73.68407561  69.66134075  68.57719676 
608 |         
609 |         # Matlab: 25,90169428  17,82760039  23,46934410  16,38532303   7,40729702   6,95257110  47,58067598  60,22686667  59,43648053  58,49404931
610 |         #       : 47,03047477  41,40945923  41,33960164  38,89865511  29,50172644  27,06865491  57,09182668  71,65815725  67,27512665  67,48550820
611 |         #       : 55,02292596  52,91285875  45,10855528  44,85804723  34,36668514  34,76908687  59,91031805  73,68396446  69,66132629  68,57720962
612 | 
613 | 
614 |         # Transform Sone
615 |         if transform_sone:
616 |             matrix = transform2sone(matrix)
617 | 
618 |         # v210715
619 |         # Python: 0.31726931   0.11815598   0.24452297   0.09450863   0.01167179   0.009812     1.6911791    4.06332931   3.84676603   3.60351463 
620 |         #       : 1.62798518   1.10263162   1.09739697   0.92887876   0.44759842   0.35631529   3.26974511   8.97447943   6.62312431   6.72041945 
621 |         #       : 2.83288863   2.44749871   1.42486669   1.40042797   0.669685     0.69054778   3.97527582  10.327417     7.81439442   7.24868691
622 |         
623 |         # Matlab: 0,31722728   0,11823469   0,24446743   0,09461230   0,01161444   0,00982439   1,69122381   4,06339796   3,84677128   3,60351520
624 |         #       : 1,62793994   1,10262783   1,09730163   0,92889083   0,44739839   0,35639734   3,26975529   8,97440147   6,62312765   6,72041730
625 |         #       : 2,83292537   2,44746100   1,42489491   1,40036676   0,66962731   0,69054210   3,97521200  10,32733744   7,81438659   7,24869337
626 | 
627 | 
628 |         # FEATURES: now we got a Sonogram and extract statistical features
629 |     
630 |         # SSD: Statistical Spectrum Descriptors
631 |         if (extract_ssd or extract_tssd):
632 |             ssd = calc_statistical_features(matrix)
633 |             ssd_list.append(ssd.flatten(FLATTEN_ORDER))
634 | 
635 |         # v210715
636 |         # Python: 2.97307486   5.10356599   0.65305978   2.35489911   2.439558     0.009812     8.1447095 
637 |         #       : 4.72262845   7.30899976   0.17862996   2.10446264   4.58595337   0.25538117  12.83339251
638 |         #       : 4.77858109   5.52646859   0.23911764   2.9056742    4.96338019   0.589568    13.6683906 
639 |         #       : 4.43503421   3.69422906   0.41473155   3.06743402   4.33220988   0.88354694  10.89393754
640 |         #       : 3.77216546   2.3993334    0.84001713   4.35548197   3.65140589   1.01199696  11.07806891
641 |         #       : 3.60563073   2.09907968   1.49906811   7.07183968   3.35596471   1.00619842  11.2872743 
642 |         #       : 3.56816128   2.20237398   1.69790808   7.57870223   3.33806767   1.10826324  10.84965392
643 |         #       : 3.43734647   2.38648202   1.59655791   6.86704341   3.23361995   1.10198021  11.89470587
644 |         #       : 3.18466303   2.39479532   1.99223131   8.83987184   2.8819031    0.93982524  11.28737448
645 |         #       : 2.90996406   1.85412568   1.97247446   8.36738395   2.68063918   0.81760102   9.64247378
646 |         
647 |         # Matlab: 2,97309758   5,11366933   0,65306558   2,35489605   2,43956735   0,00982439   8,14473582
648 |         #       : 4,72264163   7,32338449   0,17863061   2,10444843   4,58593777   0,25568703  12,83335168
649 |         #       : 4,77859306   5,53731457   0,23911126   2,90567055   4,96338616   0,58959588  13,66839858
650 |         #       : 4,43505068   3,70148292   0,41473410   3,06742263   4,33222037   0,88357883  10,89397920
651 |         #       : 3,77217541   2,40405654   0,84000183   4,35540491   3,65136495   1,01191651  11,07802201
652 |         #       : 3,60563459   2,10319516   1,49905911   7,07181623   3,35609824   1,00628652  11,28728291
653 |         #       : 3,56820841   2,20675908   1,69792784   7,57880557   3,33819690   1,10830805  10,84975850
654 |         #       : 3,43736757   2,39117736   1,59656951   6,86710630   3,23366165   1,10199096  11,89486723
655 |         #       : 3,18467212   2,39951286   1,99223621   8,83991021   2,88200015   0,93978494  11,28733449
656 |         #       : 2,90997546   1,85776617   1,97246361   8,36742039   2,68074853   0,81790606   9,64262886
657 | 
658 |         # values verified
659 | 
660 |         # RP: RHYTHM PATTERNS
661 |         feature_part_xaxis1 = range(0,mod_ampl_limit)    # take first (opts.mod_ampl_limit) values of fft result including DC component
662 |         feature_part_xaxis2 = range(1,mod_ampl_limit+1)  # leave DC component and take next (opts.mod_ampl_limit) values of fft result
663 | 
664 |         if (include_DC):
665 |             feature_part_xaxis_rp = feature_part_xaxis1
666 |         else:
667 |             feature_part_xaxis_rp = feature_part_xaxis2
668 | 
669 |         # 2nd FFT
670 |         fft_size = 2**(nextpow2(matrix.shape[1]))
671 | 
672 |         if (mod_ampl_limit >= fft_size):
673 |             raise(ValueError("mod_ampl_limit option must be smaller than FFT window size (" + str(fft_size) +  ")."))
674 |             # NOTE: in fact only half of it (256) makes sense due to the symmetry of the FFT result
675 |         
676 |         rhythm_patterns = np.zeros((matrix.shape[0], fft_size), dtype=np.complex128)
677 |         #rhythm_patterns = np.zeros((matrix.shape[0], fft_size), dtype=np.float64)
678 | 
679 |         # real_matrix = abs(matrix)
680 | 
681 |         for b in range(0,matrix.shape[0]):
682 |         
683 |             rhythm_patterns[b,:] = fft(matrix[b,:], fft_size)
684 | 
685 |             # tried this instead, but ...
686 |             #rhythm_patterns[b,:] = fft(real_matrix[b,:], fft_size)   # ... no performance improvement
687 |             #rhythm_patterns[b,:] = rfft(real_matrix[b,:], fft_size)  # ... different output values
688 |         
689 |         rhythm_patterns = rhythm_patterns / 256  # why 256?
690 | 
691 |         # convert from complex128 to float64 (real)
692 |         rp = np.abs(rhythm_patterns[:,feature_part_xaxis_rp]) # verified
693 | 
694 |         # MVD: Modulation Variance Descriptors
695 |         if extract_mvd:
696 |             mvd = calc_statistical_features(rp.transpose()) # verified
697 |             mvd_list.append(mvd.flatten(FLATTEN_ORDER))
698 | 
699 |         # RH: Rhythm Histograms - OPTION 1: before fluctuation_strength_weighting (as in Matlab)
700 |         if extract_rh:
701 |             rh = np.sum(np.abs(rhythm_patterns[:,feature_part_xaxis2]),axis=0) #without DC component # verified
702 |             rh_list.append(rh.flatten(FLATTEN_ORDER))
703 | 
704 |         # final steps for RP:
705 | 
706 |         # Fluctuation Strength weighting curve
707 |         if fluctuation_strength_weighting:
708 | 
709 |             # modulation frequency x-axis (after 2nd FFT)
710 |             # mod_freq_res = resolution of modulation frequency axis (0.17 Hz)
711 |             mod_freq_res  = 1 / (float(segment_size) / samplerate)
712 | 
713 |             #  modulation frequencies along x-axis from index 0 to 256)
714 |             mod_freq_axis = mod_freq_res * np.array(feature_part_xaxis_rp)
715 | 
716 |             #  fluctuation strength curve
717 |             fluct_curve = 1 / (mod_freq_axis/4 + 4/mod_freq_axis)
718 | 
719 |             for b in range(rp.shape[0]):
720 |                 rp[b,:] = rp[b,:] * fluct_curve #[feature_part_xaxis_rp]
721 | 
722 |         #values verified
723 | 
724 | 
725 |         # RH: Rhythm Histograms - OPTION 2 (after Fluctuation weighting)
726 |         if extract_rh2:
727 |             rh2 = np.sum(rp,axis=0) #TODO: adapt to do always without DC component
728 |             rh2_list.append(rh2.flatten(FLATTEN_ORDER))
729 | 
730 | 
731 |         # Gradient+Gauss filter
732 |         #if extract_rp:
733 |             # TODO Gradient+Gauss filter
734 | 
735 |             #for i in range(1,rp.shape[1]):
736 |             #    rp[:,i-1] = np.abs(rp[:,i] - rp[:,i-1]);
737 |             #
738 |             #rp = blur1 * rp * blur2;
739 | 
740 |         rp_list.append(rp.flatten(FLATTEN_ORDER))
741 | 
742 |         seg_pos = seg_pos + segment_size * step_width
743 | 
744 | 
745 |     if extract_rp:
746 |         if return_segment_features:
747 |             features["rp"] = np.array(rp_list)
748 |         else:
749 |             features["rp"] = np.median(np.asarray(rp_list), axis=0)
750 | 
751 |     if extract_ssd:
752 |         if return_segment_features:
753 |             features["ssd"] = np.array(ssd_list)
754 |         else:
755 |             features["ssd"]  = np.mean(np.asarray(ssd_list), axis=0)
756 |         
757 |     if extract_rh:
758 |         if return_segment_features:
759 |             features["rh"] = np.array(rh_list)
760 |         else:
761 |             features["rh"] = np.median(np.asarray(rh_list), axis=0)
762 | 
763 |     if extract_mvd:
764 |         if return_segment_features:
765 |             features["mvd"] = np.array(mvd_list)
766 |         else:
767 |             features["mvd"]  = np.mean(np.asarray(mvd_list), axis=0)
768 | 
769 |     # NOTE: no return_segment_features for temporal features as they measure variation of features over time
770 | 
771 |     if extract_tssd:
772 |         features["tssd"] = calc_statistical_features(np.asarray(ssd_list).transpose()).flatten(FLATTEN_ORDER)
773 | 
774 |     if extract_trh:
775 |         features["trh"]  = calc_statistical_features(np.asarray(rh_list).transpose()).flatten(FLATTEN_ORDER)
776 | 
777 |     if return_segment_features:
778 |         # also include the segment positions in the result
779 |         features["segpos"] = np.array(seg_pos_list)
780 |         features["timepos"] = features["segpos"] / (samplerate * 1.0)
781 | 
782 |     return features
783 | 
784 | 
785 | 
786 | # function to self test rp_extract if working properly
787 | def self_test():
788 |     import doctest
789 |     #doctest.testmod()
790 |     doctest.run_docstring_examples(rp_extract, globals(), verbose=True)
791 | 
792 | 
793 | 
794 | if __name__ == '__main__':
795 | 
796 |     import sys
797 |     from audiofile_read import *       # import our library for reading wav and mp3 files
798 | 
799 |     # process file given on command line or default song (included)
800 |     if len(sys.argv) > 1:
801 |         if sys.argv[1] == '-test': # RUN DOCSTRING SELF TEST
802 |             print "Doing self test. If nothing is printed, it is ok."
803 |             import doctest
804 |             doctest.run_docstring_examples(rp_extract, globals()) #, verbose=True)
805 |             exit()   # Note: no output means that everything went fine
806 |         else:
807 |             audiofile = sys.argv[1]
808 |     else:
809 |         audiofile = "music/BoxCat_Games_-_10_-_Epic_Song.mp3"
810 | 
811 |     # Read audio file and extract features
812 |     try:
813 | 
814 |         samplerate, samplewidth, wavedata = audiofile_read(audiofile)
815 | 
816 |         np.set_printoptions(suppress=True)
817 | 
818 |         bark_bands = 24  # choose the number of Bark bands (2..24)
819 |         mod_ampl_limit = 60 # number modulation frequencies on x-axis
820 | 
821 |         feat = rp_extract(wavedata,
822 |                           samplerate,
823 |                           extract_rp=True,
824 |                           extract_ssd=True,
825 |                           extract_tssd=False,
826 |                           extract_rh=True,
827 |                           n_bark_bands=bark_bands,
828 |                           spectral_masking=True,
829 |                           transform_db=True,
830 |                           transform_phon=True,
831 |                           transform_sone=True,
832 |                           fluctuation_strength_weighting=True,
833 |                           skip_leadin_fadeout=1,
834 |                           step_width=1,
835 |                           mod_ampl_limit=mod_ampl_limit)
836 | 
837 |         # feat is a dict containing arrays for different feature sets
838 |         print "Successfully extracted features:" , feat.keys()
839 | 
840 |     except ValueError, e:
841 |         print e
842 |         exit()
843 | 
844 | 
845 |     print "Rhythm Histogram feature vector:"
846 |     print feat["rh"]
847 | 
848 |     # EXAMPLE on how to plot the features
849 |     do_plots = False
850 | 
851 |     if do_plots:
852 |         from rp_plot import *
853 | 
854 |         plotrp(feat["rp"],rows=bark_bands,cols=mod_ampl_limit)
855 |         plotrh(feat["rh"])
856 |         plotssd(feat["ssd"],rows=bark_bands)
857 | 
858 |     # EXAMPLE on how to store RP features in CSV file
859 |     # import pandas as pd
860 |     # filename = "features.rp.csv"
861 |     # rp = pd.DataFrame(feat["rp"].reshape([1,feat["rp"].shape[0]]))
862 |     # rp.to_csv(filename)


--------------------------------------------------------------------------------
/wavio.py:
--------------------------------------------------------------------------------
  1 | # wavio.py
  2 | # Author: Warren Weckesser
  3 | # License: BSD 3-Clause (http://opensource.org/licenses/BSD-3-Clause)
  4 | 
  5 | # TODO: As of 2016, wavio is now available in PiPy. We could remove this file and add it to requirements.txt However the syntax changed and adaptions need to be made
  6 | 
  7 | import wave as _wave
  8 | import numpy as _np
  9 | 
 10 | 
 11 | __version__ = "0.0.1"
 12 | 
 13 | 
 14 | def _wav2array(nchannels, sampwidth, data):
 15 |     """data must be the string containing the bytes from the wav file."""
 16 |     num_samples, remainder = divmod(len(data), sampwidth * nchannels)
 17 |     if remainder > 0:
 18 |         raise ValueError('The length of data is not a multiple of '
 19 |                          'sampwidth * num_channels.')
 20 |     if sampwidth > 4:
 21 |         raise ValueError("sampwidth must not be greater than 4.")
 22 | 
 23 |     if sampwidth == 3:
 24 |         a = _np.empty((num_samples, nchannels, 4), dtype=_np.uint8)
 25 |         raw_bytes = _np.fromstring(data, dtype=_np.uint8)
 26 |         a[:, :, :sampwidth] = raw_bytes.reshape(-1, nchannels, sampwidth)
 27 |         a[:, :, sampwidth:] = (a[:, :, sampwidth - 1:sampwidth] >> 7) * 255
 28 |         result = a.view('<i4').reshape(a.shape[:-1])
 29 |     else:
 30 |         # 8 bit samples are stored as unsigned ints; others as signed ints.
 31 |         dt_char = 'u' if sampwidth == 1 else 'i'
 32 |         a = _np.fromstring(data, dtype='<%s%d' % (dt_char, sampwidth))
 33 |         result = a.reshape(-1, nchannels)
 34 |     return result
 35 | 
 36 | 
 37 | def readwav(file):
 38 |     """
 39 |     Read a WAV file.
 40 | 
 41 |     Parameters
 42 |     ----------
 43 |     file : string or file object
 44 |         Either the name of a file or an open file pointer.
 45 | 
 46 |     Return Values
 47 |     -------------
 48 |     rate : float
 49 |         The sampling frequency (i.e. frame rate)
 50 |     sampwidth : float
 51 |         The sample width, in bytes.  E.g. for a 24 bit WAV file,
 52 |         sampwidth is 3.
 53 |     data : numpy array
 54 |         The array containing the data.  The shape of the array is
 55 |         (num_samples, num_channels).  num_channels is the number of
 56 |         audio channels (1 for mono, 2 for stereo).
 57 | 
 58 |     Notes
 59 |     -----
 60 |     This function uses the `wave` module of the Python standard libary
 61 |     to read the WAV file, so it has the same limitations as that library.
 62 |     In particular, the function does not read compressed WAV files.
 63 | 
 64 |     """
 65 |     wav = _wave.open(file)
 66 |     rate = wav.getframerate()
 67 |     nchannels = wav.getnchannels()
 68 |     sampwidth = wav.getsampwidth()
 69 |     nframes = wav.getnframes()
 70 |     data = wav.readframes(nframes)
 71 |     wav.close()
 72 |     array = _wav2array(nchannels, sampwidth, data)
 73 |     return rate, sampwidth, array
 74 | 
 75 | 
 76 | def writewav24(filename, rate, data):
 77 |     """
 78 |     Create a 24 bit wav file.
 79 | 
 80 |     Parameters
 81 |     ----------
 82 |     filename : string
 83 |         Name of the file to create.
 84 |     rate : float
 85 |         The sampling frequency (i.e. frame rate) of the data.
 86 |     data : array-like collection of integer or floating point values
 87 |         data must be "array-like", either 1- or 2-dimensional.  If it
 88 |         is 2-d, the rows are the frames (i.e. samples) and the columns
 89 |         are the channels.
 90 | 
 91 |     Notes
 92 |     -----
 93 |     The data is assumed to be signed, and the values are assumed to be
 94 |     within the range of a 24 bit integer.  Floating point values are
 95 |     converted to integers.  The data is not rescaled or normalized before
 96 |     writing it to the file.
 97 | 
 98 |     Example
 99 |     -------
100 |     Create a 3 second 440 Hz sine wave.
101 | 
102 |     >>> rate = 22050  # samples per second
103 |     >>> T = 3         # sample duration (seconds)
104 |     >>> f = 440.0     # sound frequency (Hz)
105 |     >>> t = np.linspace(0, T, T*rate, endpoint=False)
106 |     >>> x = (2**23 - 1) * np.sin(2 * np.pi * f * t)
107 |     >>> writewav24("sine24.wav", rate, x)
108 | 
109 |     """
110 |     a32 = _np.asarray(data, dtype=_np.int32)
111 |     if a32.ndim == 1:
112 |         # Convert to a 2D array with a single column.
113 |         a32.shape = a32.shape + (1,)
114 |     # By shifting first 0 bits, then 8, then 16, the resulting output
115 |     # is 24 bit little-endian.
116 |     a8 = (a32.reshape(a32.shape + (1,)) >> _np.array([0, 8, 16])) & 255
117 |     wavdata = a8.astype(_np.uint8).tostring()
118 | 
119 |     w = _wave.open(filename, 'wb')
120 |     w.setnchannels(a32.shape[1])
121 |     w.setsampwidth(3)
122 |     w.setframerate(rate)
123 |     w.writeframes(wavdata)
124 |     w.close()
125 | 


--------------------------------------------------------------------------------