├── pretrain └── put rmvpe.pt there.txt ├── pitch ├── __init__.py ├── debug.py └── inference.py ├── rmvpe ├── constants.py ├── __init__.py ├── seq.py ├── spec.py ├── inference.py ├── model.py ├── utils.py └── deepunet.py ├── README.md ├── prepare └── preprocess_rmvpe.py ├── RMVPEF0Predictor.py └── LICENSE /pretrain/put rmvpe.pt there.txt: -------------------------------------------------------------------------------- 1 | -------------------------------------------------------------------------------- /pitch/__init__.py: -------------------------------------------------------------------------------- 1 | from .inference import load_csv_pitch -------------------------------------------------------------------------------- /rmvpe/constants.py: -------------------------------------------------------------------------------- 1 | SAMPLE_RATE = 16000 2 | 3 | N_CLASS = 360 4 | 5 | N_MELS = 128 6 | MEL_FMIN = 30 7 | MEL_FMAX = SAMPLE_RATE // 2 8 | WINDOW_LENGTH = 1024 9 | CONST = 1997.3794084376191 10 | -------------------------------------------------------------------------------- /rmvpe/__init__.py: -------------------------------------------------------------------------------- 1 | from .constants import * # noqa: F403 2 | from .inference import RMVPE # noqa: F401 3 | from .model import E2E, E2E0 # noqa: F401 4 | from .spec import MelSpectrogram # noqa: F401 5 | from .utils import ( # noqa: F401 6 | cycle, 7 | summary, 8 | to_local_average_cents, 9 | to_viterbi_cents, 10 | ) 11 | -------------------------------------------------------------------------------- /rmvpe/seq.py: -------------------------------------------------------------------------------- 1 | import torch.nn as nn 2 | 3 | 4 | class BiGRU(nn.Module): 5 | def __init__(self, input_features, hidden_features, num_layers): 6 | super(BiGRU, self).__init__() 7 | self.gru = nn.GRU(input_features, hidden_features, num_layers=num_layers, batch_first=True, bidirectional=True) 8 | 9 | def forward(self, x): 10 | return self.gru(x)[0] 11 | 12 | 13 | class BiLSTM(nn.Module): 14 | def __init__(self, input_features, hidden_features, num_layers): 15 | super(BiLSTM, self).__init__() 16 | self.lstm = nn.LSTM(input_features, hidden_features, num_layers=num_layers, batch_first=True, bidirectional=True) 17 | 18 | def forward(self, x): 19 | return self.lstm(x)[0] 20 | 21 | -------------------------------------------------------------------------------- /pitch/debug.py: -------------------------------------------------------------------------------- 1 | import argparse 2 | import numpy as np 3 | 4 | 5 | def save_csv_pitch(pitch, path): 6 | with open(path, "w", encoding='utf-8') as pitch_file: 7 | for i in range(len(pitch)): 8 | t = i * 10 9 | minute = t // 60000 10 | seconds = (t - minute * 60000) // 1000 11 | millisecond = t % 1000 12 | print( 13 | f"{minute}m {seconds}s {millisecond:3d},{int(pitch[i])}", file=pitch_file) 14 | 15 | 16 | if __name__ == "__main__": 17 | parser = argparse.ArgumentParser() 18 | parser.description = 'please enter embed parameter ...' 19 | parser.add_argument("-p", "--pit", help="pit", dest="pit") # pit for train 20 | args = parser.parse_args() 21 | print(args.pit) 22 | 23 | pitch = np.load(args.pit) 24 | save_csv_pitch(pitch, 'pitch_debug.csv') 25 | -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | ##### RMVPE 2 | 3 | If you are using the `rmvpe` F0 Predictor, you will need to download the pre-trained RMVPE model. 4 | 5 | - download model at [rmvpe.pt](https://huggingface.co/datasets/ylzz1997/rmvpe_pretrain_model/resolve/main/rmvpe.pt) 6 | - Place it under the `pretrain` directory 7 | 8 | ##### USE 9 | 10 | · Run `pitch/inference.py` 11 | 12 | You will get a CSV file containing the timestamp and fundamental frequency information 13 | 14 | · Run `prepare/preprocess_rmvpe.py` 15 | 16 | To get NPY files using RMVPE 17 | 18 | #### Any country, region, organization, or individual using this project must comply with the following laws. 19 | 20 | #### 《民法典》 21 | 22 | ##### 第一千零一十九条 23 | 24 | 任何组织或者个人不得以丑化、污损，或者利用信息技术手段伪造等方式侵害他人的肖像权。未经肖像权人同意，不得制作、使用、公开肖像权人的肖像，但是法律另有规定的除外。未经肖像权人同意，肖像作品权利人不得以发表、复制、发行、出租、展览等方式使用或者公开肖像权人的肖像。对自然人声音的保护，参照适用肖像权保护的有关规定。 25 | 26 | ##### 第一千零二十四条 27 | 28 | 【名誉权】民事主体享有名誉权。任何组织或者个人不得以侮辱、诽谤等方式侵害他人的名誉权。 29 | 30 | ##### 第一千零二十七条 31 | 32 | 【作品侵害名誉权】行为人发表的文学、艺术作品以真人真事或者特定人为描述对象，含有侮辱、诽谤内容，侵害他人名誉权的，受害人有权依法请求该行为人承担民事责任。行为人发表的文学、艺术作品不以特定人为描述对象，仅其中的情节与该特定人的情况相似的，不承担民事责任。 33 | 34 | #### 《[中华人民共和国宪法](http://www.gov.cn/guoqing/2018-03/22/content_5276318.htm)》 35 | 36 | #### 《[中华人民共和国刑法](http://gongbao.court.gov.cn/Details/f8e30d0689b23f57bfc782d21035c3.html?sw=%E4%B8%AD%E5%8D%8E%E4%BA%BA%E6%B0%91%E5%85%B1%E5%92%8C%E5%9B%BD%E5%88%91%E6%B3%95)》 37 | 38 | #### 《[中华人民共和国民法典](http://gongbao.court.gov.cn/Details/51eb6750b8361f79be8f90d09bc202.html)》 39 | 40 | #### 《[中华人民共和国合同法](http://www.npc.gov.cn/zgrdw/npc/lfzt/rlyw/2016-07/01/content_1992739.htm)》 41 | 42 | -------------------------------------------------------------------------------- /pitch/inference.py: -------------------------------------------------------------------------------- 1 | import sys,os 2 | sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__)))) 3 | import torch 4 | import librosa 5 | import argparse 6 | import numpy as np 7 | #import crepe 8 | from RMVPEF0Predictor import RMVPEF0Predictor 9 | import soundfile 10 | 11 | def save_csv_pitch(pitch, uv, path): 12 | with open(path, "w", encoding='utf-8') as pitch_file: 13 | pitch *= uv 14 | for i in range(len(pitch)): 15 | t = i * 10 16 | minute = t // 60000 17 | seconds = (t - minute * 60000) // 1000 18 | millisecond = t % 1000 19 | print( 20 | f"{minute}m {seconds}s {millisecond:3d},{int(pitch[i])}", file=pitch_file) 21 | 22 | 23 | def load_csv_pitch(path): 24 | pitch = [] 25 | with open(path, "r", encoding='utf-8') as pitch_file: 26 | for line in pitch_file.readlines(): 27 | pit = line.strip().split(",")[-1] 28 | pitch.append(int(pit)) 29 | return pitch 30 | 31 | 32 | if __name__ == "__main__": 33 | parser = argparse.ArgumentParser() 34 | parser.description = 'please enter embed parameter ...' 35 | parser.add_argument("-w", "--wav", help="wav", dest="wav") 36 | parser.add_argument("-p", "--pit", help="pit", dest="pit") # csv for excel 37 | args = parser.parse_args() 38 | print(args.wav) 39 | print(args.pit) 40 | 41 | device = "cuda" if torch.cuda.is_available() else "cpu" 42 | #pitch = compute_f0_sing(args.wav, device) 43 | predictor = RMVPEF0Predictor(hop_length=320, f0_min=50, f0_max=1100, device=device) 44 | audio, sampling_rate = soundfile.read(args.wav) 45 | if len(audio.shape) > 1: 46 | audio = librosa.to_mono(audio.transpose(1, 0)) 47 | if sampling_rate != 16000: 48 | audio = librosa.resample(audio, orig_sr=sampling_rate, target_sr=16000) 49 | pitch, uv = predictor.compute_f0_uv(audio) 50 | pitch = np.repeat(pitch, 2, -1) 51 | uv = np.repeat(uv, 2, -1) 52 | save_csv_pitch(pitch, uv, args.pit) 53 | #tmp = load_csv_pitch(args.pit) 54 | #save_csv_pitch(tmp, "tmp.csv") 55 | -------------------------------------------------------------------------------- /rmvpe/spec.py: -------------------------------------------------------------------------------- 1 | import numpy as np 2 | import torch 3 | import torch.nn.functional as F 4 | from librosa.filters import mel 5 | 6 | 7 | class MelSpectrogram(torch.nn.Module): 8 | def __init__( 9 | self, 10 | n_mel_channels, 11 | sampling_rate, 12 | win_length, 13 | hop_length, 14 | n_fft=None, 15 | mel_fmin=0, 16 | mel_fmax=None, 17 | clamp = 1e-5 18 | ): 19 | super().__init__() 20 | n_fft = win_length if n_fft is None else n_fft 21 | self.hann_window = {} 22 | mel_basis = mel( 23 | sr=sampling_rate, 24 | n_fft=n_fft, 25 | n_mels=n_mel_channels, 26 | fmin=mel_fmin, 27 | fmax=mel_fmax, 28 | htk=True) 29 | mel_basis = torch.from_numpy(mel_basis).float() 30 | self.register_buffer("mel_basis", mel_basis) 31 | self.n_fft = win_length if n_fft is None else n_fft 32 | self.hop_length = hop_length 33 | self.win_length = win_length 34 | self.sampling_rate = sampling_rate 35 | self.n_mel_channels = n_mel_channels 36 | self.clamp = clamp 37 | 38 | def forward(self, audio, keyshift=0, speed=1, center=True): 39 | factor = 2 ** (keyshift / 12) 40 | n_fft_new = int(np.round(self.n_fft * factor)) 41 | win_length_new = int(np.round(self.win_length * factor)) 42 | hop_length_new = int(np.round(self.hop_length * speed)) 43 | 44 | keyshift_key = str(keyshift)+'_'+str(audio.device) 45 | if keyshift_key not in self.hann_window: 46 | self.hann_window[keyshift_key] = torch.hann_window(win_length_new).to(audio.device) 47 | 48 | fft = torch.stft( 49 | audio, 50 | n_fft=n_fft_new, 51 | hop_length=hop_length_new, 52 | win_length=win_length_new, 53 | window=self.hann_window[keyshift_key], 54 | center=center, 55 | return_complex=True) 56 | magnitude = torch.sqrt(fft.real.pow(2) + fft.imag.pow(2)) 57 | 58 | if keyshift != 0: 59 | size = self.n_fft // 2 + 1 60 | resize = magnitude.size(1) 61 | if resize < size: 62 | magnitude = F.pad(magnitude, (0, 0, 0, size-resize)) 63 | magnitude = magnitude[:, :size, :] * self.win_length / win_length_new 64 | 65 | mel_output = torch.matmul(self.mel_basis, magnitude) 66 | log_mel_spec = torch.log(torch.clamp(mel_output, min=self.clamp)) 67 | return log_mel_spec -------------------------------------------------------------------------------- /rmvpe/inference.py: -------------------------------------------------------------------------------- 1 | import torch 2 | import torch.nn.functional as F 3 | from torchaudio.transforms import Resample 4 | 5 | from .constants import * # noqa: F403 6 | from .model import E2E0 7 | from .spec import MelSpectrogram 8 | from .utils import to_local_average_cents, to_viterbi_cents 9 | 10 | 11 | class RMVPE: 12 | def __init__(self, model_path, device=None, dtype = torch.float32, hop_length=160): 13 | self.resample_kernel = {} 14 | if device is None: 15 | self.device = 'cuda' if torch.cuda.is_available() else 'cpu' 16 | else: 17 | self.device = device 18 | model = E2E0(4, 1, (2, 2)) 19 | ckpt = torch.load(model_path, map_location=torch.device(self.device)) 20 | model.load_state_dict(ckpt['model']) 21 | model = model.to(dtype).to(self.device) 22 | model.eval() 23 | self.model = model 24 | self.dtype = dtype 25 | self.mel_extractor = MelSpectrogram(N_MELS, SAMPLE_RATE, WINDOW_LENGTH, hop_length, None, MEL_FMIN, MEL_FMAX) # noqa: F405 26 | self.resample_kernel = {} 27 | 28 | def mel2hidden(self, mel): 29 | with torch.no_grad(): 30 | n_frames = mel.shape[-1] 31 | mel = F.pad(mel, (0, 32 * ((n_frames - 1) // 32 + 1) - n_frames), mode='reflect') 32 | hidden = self.model(mel) 33 | return hidden[:, :n_frames] 34 | 35 | def decode(self, hidden, thred=0.03, use_viterbi=False): 36 | if use_viterbi: 37 | cents_pred = to_viterbi_cents(hidden, thred=thred) 38 | else: 39 | cents_pred = to_local_average_cents(hidden, thred=thred) 40 | f0 = torch.Tensor([10 * (2 ** (cent_pred / 1200)) if cent_pred else 0 for cent_pred in cents_pred]).to(self.device) 41 | return f0 42 | 43 | def infer_from_audio(self, audio, sample_rate=16000, thred=0.05, use_viterbi=False): 44 | audio = audio.unsqueeze(0).to(self.dtype).to(self.device) 45 | if sample_rate == 16000: 46 | audio_res = audio 47 | else: 48 | key_str = str(sample_rate) 49 | if key_str not in self.resample_kernel: 50 | self.resample_kernel[key_str] = Resample(sample_rate, 16000, lowpass_filter_width=128) 51 | self.resample_kernel[key_str] = self.resample_kernel[key_str].to(self.dtype).to(self.device) 52 | audio_res = self.resample_kernel[key_str](audio) 53 | mel_extractor = self.mel_extractor.to(self.device) 54 | mel = mel_extractor(audio_res, center=True).to(self.dtype) 55 | hidden = self.mel2hidden(mel) 56 | f0 = self.decode(hidden.squeeze(0), thred=thred, use_viterbi=use_viterbi) 57 | return f0 58 | -------------------------------------------------------------------------------- /rmvpe/model.py: -------------------------------------------------------------------------------- 1 | from torch import nn 2 | 3 | from .constants import * # noqa: F403 4 | from .deepunet import DeepUnet, DeepUnet0 5 | from .seq import BiGRU 6 | from .spec import MelSpectrogram 7 | 8 | 9 | class E2E(nn.Module): 10 | def __init__(self, hop_length, n_blocks, n_gru, kernel_size, en_de_layers=5, inter_layers=4, in_channels=1, 11 | en_out_channels=16): 12 | super(E2E, self).__init__() 13 | self.mel = MelSpectrogram(N_MELS, SAMPLE_RATE, WINDOW_LENGTH, hop_length, None, MEL_FMIN, MEL_FMAX) # noqa: F405 14 | self.unet = DeepUnet(kernel_size, n_blocks, en_de_layers, inter_layers, in_channels, en_out_channels) 15 | self.cnn = nn.Conv2d(en_out_channels, 3, (3, 3), padding=(1, 1)) 16 | if n_gru: 17 | self.fc = nn.Sequential( 18 | BiGRU(3 * N_MELS, 256, n_gru), # noqa: F405 19 | nn.Linear(512, N_CLASS), # noqa: F405 20 | nn.Dropout(0.25), 21 | nn.Sigmoid() 22 | ) 23 | else: 24 | self.fc = nn.Sequential( 25 | nn.Linear(3 * N_MELS, N_CLASS), # noqa: F405 26 | nn.Dropout(0.25), 27 | nn.Sigmoid() 28 | ) 29 | 30 | def forward(self, x): 31 | mel = self.mel(x.reshape(-1, x.shape[-1])).transpose(-1, -2).unsqueeze(1) 32 | x = self.cnn(self.unet(mel)).transpose(1, 2).flatten(-2) 33 | # x = self.fc(x) 34 | hidden_vec = 0 35 | if len(self.fc) == 4: 36 | for i in range(len(self.fc)): 37 | x = self.fc[i](x) 38 | if i == 0: 39 | hidden_vec = x 40 | return hidden_vec, x 41 | 42 | 43 | class E2E0(nn.Module): 44 | def __init__(self, n_blocks, n_gru, kernel_size, en_de_layers=5, inter_layers=4, in_channels=1, 45 | en_out_channels=16): 46 | super(E2E0, self).__init__() 47 | self.unet = DeepUnet0(kernel_size, n_blocks, en_de_layers, inter_layers, in_channels, en_out_channels) 48 | self.cnn = nn.Conv2d(en_out_channels, 3, (3, 3), padding=(1, 1)) 49 | if n_gru: 50 | self.fc = nn.Sequential( 51 | BiGRU(3 * N_MELS, 256, n_gru), # noqa: F405 52 | nn.Linear(512, N_CLASS), # noqa: F405 53 | nn.Dropout(0.25), 54 | nn.Sigmoid() 55 | ) 56 | else: 57 | self.fc = nn.Sequential( 58 | nn.Linear(3 * N_MELS, N_CLASS), # noqa: F405 59 | nn.Dropout(0.25), 60 | nn.Sigmoid() 61 | ) 62 | 63 | def forward(self, mel): 64 | mel = mel.transpose(-1, -2).unsqueeze(1) 65 | x = self.cnn(self.unet(mel)).transpose(1, 2).flatten(-2) 66 | x = self.fc(x) 67 | return x 68 | -------------------------------------------------------------------------------- /prepare/preprocess_rmvpe.py: -------------------------------------------------------------------------------- 1 | import sys,os 2 | sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__)))) 3 | import numpy as np 4 | import librosa 5 | import torch 6 | import argparse 7 | from tqdm import tqdm 8 | from multiprocessing import set_start_method 9 | from concurrent.futures import ProcessPoolExecutor, as_completed 10 | from RMVPEF0Predictor import RMVPEF0Predictor 11 | 12 | def compute_f0(filename, save, device): 13 | audio, sr = librosa.load(filename, sr=16000) 14 | assert sr == 16000 15 | # Load audio 16 | audio = torch.tensor(np.copy(audio)) 17 | audio = audio + torch.randn_like(audio) * 0.001 18 | predictor = RMVPEF0Predictor(hop_length=160, f0_min=50, f0_max=1100, device=device) 19 | 20 | pitch, uv = predictor.compute_f0_uv(audio) 21 | 22 | pitch = abs(pitch) * uv 23 | pitch = torch.from_numpy(pitch).to(torch.float32) 24 | pitch = pitch.squeeze(0) 25 | np.save(save, pitch, allow_pickle=False) 26 | 27 | 28 | def process_file(file, wavPath, spks, pitPath, device): 29 | if file.endswith(".wav"): 30 | file = file[:-4] 31 | compute_f0(f"{wavPath}/{spks}/{file}.wav", f"{pitPath}/{spks}/{file}.pit", device) 32 | 33 | def process_files_with_process_pool(wavPath, spks, pitPath, device, process_num=None): 34 | files = [f for f in os.listdir(f"./{wavPath}/{spks}") if f.endswith(".wav")] 35 | 36 | with ProcessPoolExecutor(max_workers=process_num) as executor: 37 | futures = {executor.submit(process_file, file, wavPath, spks, pitPath, device): file for file in files} 38 | 39 | for future in tqdm(as_completed(futures), total=len(futures), desc='Processing files'): 40 | future.result() 41 | 42 | if __name__ == "__main__": 43 | parser = argparse.ArgumentParser() 44 | parser.description = 'please enter embed parameter ...' 45 | parser.add_argument("-w", "--wav", help="wav", dest="wav") 46 | parser.add_argument("-p", "--pit", help="pit", dest="pit") 47 | parser.add_argument("-t", "--thread_count", help="thread count to process, set 0 to use all cpu cores", dest="thread_count", type=int, default=1) 48 | args = parser.parse_args() 49 | print(args.wav) 50 | print(args.pit) 51 | os.makedirs(args.pit, exist_ok=True) 52 | wavPath = args.wav 53 | pitPath = args.pit 54 | 55 | device = "cuda" if torch.cuda.is_available() else "cpu" 56 | if device == "cuda": 57 | set_start_method('spawn') 58 | 59 | for spks in os.listdir(wavPath): 60 | if os.path.isdir(f"./{wavPath}/{spks}"): 61 | os.makedirs(f"./{pitPath}/{spks}", exist_ok=True) 62 | print(f">>>>>>>>>>{spks}<<<<<<<<<<") 63 | if args.thread_count == 0: 64 | process_num = os.cpu_count() 65 | else: 66 | process_num = args.thread_count 67 | process_files_with_process_pool(wavPath, spks, pitPath, device, process_num) 68 | -------------------------------------------------------------------------------- /rmvpe/utils.py: -------------------------------------------------------------------------------- 1 | import sys 2 | from functools import reduce 3 | 4 | import librosa 5 | import numpy as np 6 | import torch 7 | from torch.nn.modules.module import _addindent 8 | 9 | from .constants import * # noqa: F403 10 | 11 | 12 | def cycle(iterable): 13 | while True: 14 | for item in iterable: 15 | yield item 16 | 17 | 18 | def summary(model, file=sys.stdout): 19 | def repr(model): 20 | # We treat the extra repr like the sub-module, one item per line 21 | extra_lines = [] 22 | extra_repr = model.extra_repr() 23 | # empty string will be split into list [''] 24 | if extra_repr: 25 | extra_lines = extra_repr.split('\n') 26 | child_lines = [] 27 | total_params = 0 28 | for key, module in model._modules.items(): 29 | mod_str, num_params = repr(module) 30 | mod_str = _addindent(mod_str, 2) 31 | child_lines.append('(' + key + '): ' + mod_str) 32 | total_params += num_params 33 | lines = extra_lines + child_lines 34 | 35 | for name, p in model._parameters.items(): 36 | if hasattr(p, 'shape'): 37 | total_params += reduce(lambda x, y: x * y, p.shape) 38 | 39 | main_str = model._get_name() + '(' 40 | if lines: 41 | # simple one-liner info, which most builtin Modules will use 42 | if len(extra_lines) == 1 and not child_lines: 43 | main_str += extra_lines[0] 44 | else: 45 | main_str += '\n ' + '\n '.join(lines) + '\n' 46 | 47 | main_str += ')' 48 | if file is sys.stdout: 49 | main_str += ', \033[92m{:,}\033[0m params'.format(total_params) 50 | else: 51 | main_str += ', {:,} params'.format(total_params) 52 | return main_str, total_params 53 | 54 | string, count = repr(model) 55 | if file is not None: 56 | if isinstance(file, str): 57 | file = open(file, 'w') 58 | print(string, file=file) 59 | file.flush() 60 | 61 | return count 62 | 63 | 64 | def to_local_average_cents(salience, center=None, thred=0.05): 65 | """ 66 | find the weighted average cents near the argmax bin 67 | """ 68 | 69 | if not hasattr(to_local_average_cents, 'cents_mapping'): 70 | # the bin number-to-cents mapping 71 | to_local_average_cents.cents_mapping = ( 72 | 20 * torch.arange(N_CLASS) + CONST).to(salience.device) # noqa: F405 73 | 74 | if salience.ndim == 1: 75 | if center is None: 76 | center = int(torch.argmax(salience)) 77 | start = max(0, center - 4) 78 | end = min(len(salience), center + 5) 79 | salience = salience[start:end] 80 | product_sum = torch.sum( 81 | salience * to_local_average_cents.cents_mapping[start:end]) 82 | weight_sum = torch.sum(salience) 83 | return product_sum / weight_sum if torch.max(salience) > thred else 0 84 | if salience.ndim == 2: 85 | return torch.Tensor([to_local_average_cents(salience[i, :], None, thred) for i in 86 | range(salience.shape[0])]).to(salience.device) 87 | 88 | raise Exception("label should be either 1d or 2d ndarray") 89 | 90 | def to_viterbi_cents(salience, thred=0.05): 91 | # Create viterbi transition matrix 92 | if not hasattr(to_viterbi_cents, 'transition'): 93 | xx, yy = torch.meshgrid(range(N_CLASS), range(N_CLASS)) # noqa: F405 94 | transition = torch.maximum(30 - abs(xx - yy), 0) 95 | transition = transition / transition.sum(axis=1, keepdims=True) 96 | to_viterbi_cents.transition = transition 97 | 98 | # Convert to probability 99 | prob = salience.T 100 | prob = prob / prob.sum(axis=0) 101 | 102 | # Perform viterbi decoding 103 | path = librosa.sequence.viterbi(prob.detach().cpu().numpy(), to_viterbi_cents.transition).astype(np.int64) 104 | 105 | return torch.Tensor([to_local_average_cents(salience[i, :], path[i], thred) for i in 106 | range(len(path))]).to(salience.device) 107 | -------------------------------------------------------------------------------- /RMVPEF0Predictor.py: -------------------------------------------------------------------------------- 1 | from typing import Union 2 | import sys, torch, numpy as np, traceback, pdb 3 | 4 | import torch 5 | import torch.nn.functional as F 6 | 7 | from rmvpe import RMVPE 8 | 9 | class RMVPEF0Predictor: 10 | def __init__(self, hop_length=320, f0_min=50, f0_max=1100, dtype=torch.float32, device=None, 11 | sampling_rate=16000, 12 | threshold=0.05): 13 | self.rmvpe = RMVPE(model_path="pretrain/rmvpe.pt", dtype=dtype, device=device) 14 | self.hop_length = hop_length 15 | self.f0_min = f0_min 16 | self.f0_max = f0_max 17 | if device is None: 18 | self.device = 'cuda' if torch.cuda.is_available() else 'cpu' 19 | else: 20 | self.device = device 21 | self.threshold = threshold 22 | self.sampling_rate = sampling_rate 23 | self.dtype = dtype 24 | self.name = "rmvpe" 25 | 26 | def repeat_expand( 27 | self, content: Union[torch.Tensor, np.ndarray], target_len: int, mode: str = "nearest" 28 | ): 29 | ndim = content.ndim 30 | 31 | if content.ndim == 1: 32 | content = content[None, None] 33 | elif content.ndim == 2: 34 | content = content[None] 35 | 36 | assert content.ndim == 3 37 | 38 | is_np = isinstance(content, np.ndarray) 39 | if is_np: 40 | content = torch.from_numpy(content) 41 | 42 | results = torch.nn.functional.interpolate(content, size=target_len, mode=mode) 43 | 44 | if is_np: 45 | results = results.numpy() 46 | 47 | if ndim == 1: 48 | return results[0, 0] 49 | elif ndim == 2: 50 | return results[0] 51 | # 默认 52 | def post_process0(self, x, sampling_rate, f0, pad_to): 53 | if isinstance(f0, np.ndarray): 54 | f0 = torch.from_numpy(f0).float().to(x.device) 55 | 56 | if pad_to is None: 57 | return f0 58 | 59 | f0 = self.repeat_expand(f0, pad_to) 60 | 61 | vuv_vector = torch.zeros_like(f0) 62 | vuv_vector[f0 > 0.0] = 1.0 63 | vuv_vector[f0 <= 0.0] = 0.0 64 | 65 | # 去掉0频率, 并线性插值 66 | nzindex = torch.nonzero(f0).squeeze() 67 | f0 = torch.index_select(f0, dim=0, index=nzindex).cpu().numpy() 68 | time_org = self.hop_length / sampling_rate * nzindex.cpu().numpy() 69 | time_frame = np.arange(pad_to) * self.hop_length / sampling_rate 70 | 71 | vuv_vector = F.interpolate(vuv_vector[None, None, :], size=pad_to)[0][0] 72 | 73 | if f0.shape[0] <= 0: 74 | return torch.zeros(pad_to, dtype=torch.float, device=x.device).cpu().numpy(), vuv_vector.cpu().numpy() 75 | if f0.shape[0] == 1: 76 | return (torch.ones(pad_to, dtype=torch.float, device=x.device) * f0[ 77 | 0]).cpu().numpy(), vuv_vector.cpu().numpy() 78 | 79 | # 大概可以用 torch 重写? 80 | f0 = np.interp(time_frame, time_org, f0, left=f0[0], right=f0[-1]) 81 | #vuv_vector = np.ceil(scipy.ndimage.zoom(vuv_vector,pad_to/len(vuv_vector),order = 0)) 82 | 83 | return f0, vuv_vector.cpu().numpy() 84 | 85 | # 分段线性插值 86 | def post_process1(self, x, sampling_rate, f0, pad_to): 87 | if isinstance(f0, np.ndarray): 88 | f0 = torch.from_numpy(f0).float().to(x.device) 89 | 90 | if pad_to is None: 91 | return f0 92 | 93 | f0 = self.repeat_expand(f0, pad_to) 94 | 95 | vuv_vector = torch.zeros_like(f0) 96 | vuv_vector[f0 > 0.0] = 1.0 97 | vuv_vector[f0 <= 0.0] = 0.0 98 | 99 | nzindex = torch.nonzero(f0).squeeze() 100 | segments = [] 101 | for i in range(len(nzindex) - 1): 102 | if nzindex[i + 1] - nzindex[i] > 1: 103 | segments.append((nzindex[i], nzindex[i + 1])) 104 | 105 | for start, end in segments: 106 | slope = (f0[end] - f0[start]) / (end - start) 107 | intercept = f0[start] - slope * start 108 | f0[start + 1: end] = slope * torch.arange(start + 1, end, device=f0.device) + intercept 109 | 110 | return f0.cpu().numpy(), vuv_vector.cpu().numpy() 111 | 112 | # 使用三次样条插值 113 | def post_process2(self, x, sampling_rate, f0, pad_to): 114 | if isinstance(f0, np.ndarray): 115 | f0 = torch.from_numpy(f0).float().to(x.device) 116 | 117 | if pad_to is None: 118 | return f0 119 | 120 | f0 = self.repeat_expand(f0, pad_to) 121 | 122 | vuv_vector = torch.zeros_like(f0) 123 | vuv_vector[f0 > 0.0] = 1.0 124 | vuv_vector[f0 <= 0.0] = 0.0 125 | 126 | nzindex = torch.nonzero(f0).squeeze() 127 | f0 = torch.index_select(f0, dim=0, index=nzindex).cpu().numpy() 128 | time_org = self.hop_length / sampling_rate * nzindex.cpu().numpy() 129 | time_frame = np.arange(pad_to) * self.hop_length / sampling_rate 130 | 131 | vuv_vector = F.interpolate(vuv_vector[None, None, :], size=pad_to)[0][0] 132 | 133 | if f0.shape[0] <= 0: 134 | return torch.zeros(pad_to, dtype=torch.float, device=x.device).cpu().numpy(), vuv_vector.cpu().numpy() 135 | if f0.shape[0] == 1: 136 | return (torch.ones(pad_to, dtype=torch.float, device=x.device) * f0[ 137 | 0]).cpu().numpy(), vuv_vector.cpu().numpy() 138 | 139 | from scipy.interpolate import CubicSpline 140 | cs = CubicSpline(time_org, f0) 141 | f0 = cs(time_frame) 142 | 143 | return f0, vuv_vector.cpu().numpy() 144 | 145 | def compute_f0(self, wav, p_len=None): 146 | x = torch.FloatTensor(wav).to(self.dtype).to(self.device) 147 | if p_len is None: 148 | p_len = x.shape[0] // self.hop_length 149 | else: 150 | assert abs(p_len - x.shape[0] // self.hop_length) < 4, "pad length error" 151 | f0 = self.rmvpe.infer_from_audio(x, self.sampling_rate, self.threshold) 152 | if torch.all(f0 == 0): 153 | rtn = f0.cpu().numpy() if p_len is None else np.zeros(p_len) 154 | return rtn, rtn 155 | return self.post_process0(x, self.sampling_rate, f0, p_len)[0] 156 | 157 | def compute_f0_uv(self, wav, p_len=None): 158 | x = torch.FloatTensor(wav).to(self.dtype).to(self.device) 159 | if p_len is None: 160 | p_len = x.shape[0] // self.hop_length 161 | else: 162 | assert abs(p_len - x.shape[0] // self.hop_length) < 4, "pad length error" 163 | f0 = self.rmvpe.infer_from_audio(x, self.sampling_rate, self.threshold) 164 | if torch.all(f0 == 0): 165 | rtn = f0.cpu().numpy() if p_len is None else np.zeros(p_len) 166 | return rtn, rtn 167 | return self.post_process0(x, self.sampling_rate, f0, p_len) 168 | -------------------------------------------------------------------------------- /rmvpe/deepunet.py: -------------------------------------------------------------------------------- 1 | import torch 2 | import torch.nn as nn 3 | 4 | from .constants import N_MELS 5 | 6 | 7 | class ConvBlockRes(nn.Module): 8 | def __init__(self, in_channels, out_channels, momentum=0.01): 9 | super(ConvBlockRes, self).__init__() 10 | self.conv = nn.Sequential( 11 | nn.Conv2d(in_channels=in_channels, 12 | out_channels=out_channels, 13 | kernel_size=(3, 3), 14 | stride=(1, 1), 15 | padding=(1, 1), 16 | bias=False), 17 | nn.BatchNorm2d(out_channels, momentum=momentum), 18 | nn.ReLU(), 19 | 20 | nn.Conv2d(in_channels=out_channels, 21 | out_channels=out_channels, 22 | kernel_size=(3, 3), 23 | stride=(1, 1), 24 | padding=(1, 1), 25 | bias=False), 26 | nn.BatchNorm2d(out_channels, momentum=momentum), 27 | nn.ReLU(), 28 | ) 29 | if in_channels != out_channels: 30 | self.shortcut = nn.Conv2d(in_channels, out_channels, (1, 1)) 31 | self.is_shortcut = True 32 | else: 33 | self.is_shortcut = False 34 | 35 | def forward(self, x): 36 | if self.is_shortcut: 37 | return self.conv(x) + self.shortcut(x) 38 | else: 39 | return self.conv(x) + x 40 | 41 | 42 | class ResEncoderBlock(nn.Module): 43 | def __init__(self, in_channels, out_channels, kernel_size, n_blocks=1, momentum=0.01): 44 | super(ResEncoderBlock, self).__init__() 45 | self.n_blocks = n_blocks 46 | self.conv = nn.ModuleList() 47 | self.conv.append(ConvBlockRes(in_channels, out_channels, momentum)) 48 | for i in range(n_blocks - 1): 49 | self.conv.append(ConvBlockRes(out_channels, out_channels, momentum)) 50 | self.kernel_size = kernel_size 51 | if self.kernel_size is not None: 52 | self.pool = nn.AvgPool2d(kernel_size=kernel_size) 53 | 54 | def forward(self, x): 55 | for i in range(self.n_blocks): 56 | x = self.conv[i](x) 57 | if self.kernel_size is not None: 58 | return x, self.pool(x) 59 | else: 60 | return x 61 | 62 | 63 | class ResDecoderBlock(nn.Module): 64 | def __init__(self, in_channels, out_channels, stride, n_blocks=1, momentum=0.01): 65 | super(ResDecoderBlock, self).__init__() 66 | out_padding = (0, 1) if stride == (1, 2) else (1, 1) 67 | self.n_blocks = n_blocks 68 | self.conv1 = nn.Sequential( 69 | nn.ConvTranspose2d(in_channels=in_channels, 70 | out_channels=out_channels, 71 | kernel_size=(3, 3), 72 | stride=stride, 73 | padding=(1, 1), 74 | output_padding=out_padding, 75 | bias=False), 76 | nn.BatchNorm2d(out_channels, momentum=momentum), 77 | nn.ReLU(), 78 | ) 79 | self.conv2 = nn.ModuleList() 80 | self.conv2.append(ConvBlockRes(out_channels * 2, out_channels, momentum)) 81 | for i in range(n_blocks-1): 82 | self.conv2.append(ConvBlockRes(out_channels, out_channels, momentum)) 83 | 84 | def forward(self, x, concat_tensor): 85 | x = self.conv1(x) 86 | x = torch.cat((x, concat_tensor), dim=1) 87 | for i in range(self.n_blocks): 88 | x = self.conv2[i](x) 89 | return x 90 | 91 | 92 | class Encoder(nn.Module): 93 | def __init__(self, in_channels, in_size, n_encoders, kernel_size, n_blocks, out_channels=16, momentum=0.01): 94 | super(Encoder, self).__init__() 95 | self.n_encoders = n_encoders 96 | self.bn = nn.BatchNorm2d(in_channels, momentum=momentum) 97 | self.layers = nn.ModuleList() 98 | self.latent_channels = [] 99 | for i in range(self.n_encoders): 100 | self.layers.append(ResEncoderBlock(in_channels, out_channels, kernel_size, n_blocks, momentum=momentum)) 101 | self.latent_channels.append([out_channels, in_size]) 102 | in_channels = out_channels 103 | out_channels *= 2 104 | in_size //= 2 105 | self.out_size = in_size 106 | self.out_channel = out_channels 107 | 108 | def forward(self, x): 109 | concat_tensors = [] 110 | x = self.bn(x) 111 | for i in range(self.n_encoders): 112 | _, x = self.layers[i](x) 113 | concat_tensors.append(_) 114 | return x, concat_tensors 115 | 116 | 117 | class Intermediate(nn.Module): 118 | def __init__(self, in_channels, out_channels, n_inters, n_blocks, momentum=0.01): 119 | super(Intermediate, self).__init__() 120 | self.n_inters = n_inters 121 | self.layers = nn.ModuleList() 122 | self.layers.append(ResEncoderBlock(in_channels, out_channels, None, n_blocks, momentum)) 123 | for i in range(self.n_inters-1): 124 | self.layers.append(ResEncoderBlock(out_channels, out_channels, None, n_blocks, momentum)) 125 | 126 | def forward(self, x): 127 | for i in range(self.n_inters): 128 | x = self.layers[i](x) 129 | return x 130 | 131 | 132 | class Decoder(nn.Module): 133 | def __init__(self, in_channels, n_decoders, stride, n_blocks, momentum=0.01): 134 | super(Decoder, self).__init__() 135 | self.layers = nn.ModuleList() 136 | self.n_decoders = n_decoders 137 | for i in range(self.n_decoders): 138 | out_channels = in_channels // 2 139 | self.layers.append(ResDecoderBlock(in_channels, out_channels, stride, n_blocks, momentum)) 140 | in_channels = out_channels 141 | 142 | def forward(self, x, concat_tensors): 143 | for i in range(self.n_decoders): 144 | x = self.layers[i](x, concat_tensors[-1-i]) 145 | return x 146 | 147 | 148 | class TimbreFilter(nn.Module): 149 | def __init__(self, latent_rep_channels): 150 | super(TimbreFilter, self).__init__() 151 | self.layers = nn.ModuleList() 152 | for latent_rep in latent_rep_channels: 153 | self.layers.append(ConvBlockRes(latent_rep[0], latent_rep[0])) 154 | 155 | def forward(self, x_tensors): 156 | out_tensors = [] 157 | for i, layer in enumerate(self.layers): 158 | out_tensors.append(layer(x_tensors[i])) 159 | return out_tensors 160 | 161 | 162 | class DeepUnet(nn.Module): 163 | def __init__(self, kernel_size, n_blocks, en_de_layers=5, inter_layers=4, in_channels=1, en_out_channels=16): 164 | super(DeepUnet, self).__init__() 165 | self.encoder = Encoder(in_channels, N_MELS, en_de_layers, kernel_size, n_blocks, en_out_channels) 166 | self.intermediate = Intermediate(self.encoder.out_channel // 2, self.encoder.out_channel, inter_layers, n_blocks) 167 | self.tf = TimbreFilter(self.encoder.latent_channels) 168 | self.decoder = Decoder(self.encoder.out_channel, en_de_layers, kernel_size, n_blocks) 169 | 170 | def forward(self, x): 171 | x, concat_tensors = self.encoder(x) 172 | x = self.intermediate(x) 173 | concat_tensors = self.tf(concat_tensors) 174 | x = self.decoder(x, concat_tensors) 175 | return x 176 | 177 | 178 | class DeepUnet0(nn.Module): 179 | def __init__(self, kernel_size, n_blocks, en_de_layers=5, inter_layers=4, in_channels=1, en_out_channels=16): 180 | super(DeepUnet0, self).__init__() 181 | self.encoder = Encoder(in_channels, N_MELS, en_de_layers, kernel_size, n_blocks, en_out_channels) 182 | self.intermediate = Intermediate(self.encoder.out_channel // 2, self.encoder.out_channel, inter_layers, n_blocks) 183 | self.tf = TimbreFilter(self.encoder.latent_channels) 184 | self.decoder = Decoder(self.encoder.out_channel, en_de_layers, kernel_size, n_blocks) 185 | 186 | def forward(self, x): 187 | x, concat_tensors = self.encoder(x) 188 | x = self.intermediate(x) 189 | x = self.decoder(x, concat_tensors) 190 | return x 191 | -------------------------------------------------------------------------------- /LICENSE: -------------------------------------------------------------------------------- 1 | GNU AFFERO GENERAL PUBLIC LICENSE 2 | Version 3, 19 November 2007 3 | 4 | Copyright (C) 2007 Free Software Foundation, Inc. 5 | Everyone is permitted to copy and distribute verbatim copies 6 | of this license document, but changing it is not allowed. 7 | 8 | Preamble 9 | 10 | The GNU Affero General Public License is a free, copyleft license for 11 | software and other kinds of works, specifically designed to ensure 12 | cooperation with the community in the case of network server software. 13 | 14 | The licenses for most software and other practical works are designed 15 | to take away your freedom to share and change the works. By contrast, 16 | our General Public Licenses are intended to guarantee your freedom to 17 | share and change all versions of a program--to make sure it remains free 18 | software for all its users. 19 | 20 | When we speak of free software, we are referring to freedom, not 21 | price. Our General Public Licenses are designed to make sure that you 22 | have the freedom to distribute copies of free software (and charge for 23 | them if you wish), that you receive source code or can get it if you 24 | want it, that you can change the software or use pieces of it in new 25 | free programs, and that you know you can do these things. 26 | 27 | Developers that use our General Public Licenses protect your rights 28 | with two steps: (1) assert copyright on the software, and (2) offer 29 | you this License which gives you legal permission to copy, distribute 30 | and/or modify the software. 31 | 32 | A secondary benefit of defending all users' freedom is that 33 | improvements made in alternate versions of the program, if they 34 | receive widespread use, become available for other developers to 35 | incorporate. Many developers of free software are heartened and 36 | encouraged by the resulting cooperation. However, in the case of 37 | software used on network servers, this result may fail to come about. 38 | The GNU General Public License permits making a modified version and 39 | letting the public access it on a server without ever releasing its 40 | source code to the public. 41 | 42 | The GNU Affero General Public License is designed specifically to 43 | ensure that, in such cases, the modified source code becomes available 44 | to the community. It requires the operator of a network server to 45 | provide the source code of the modified version running there to the 46 | users of that server. Therefore, public use of a modified version, on 47 | a publicly accessible server, gives the public access to the source 48 | code of the modified version. 49 | 50 | An older license, called the Affero General Public License and 51 | published by Affero, was designed to accomplish similar goals. This is 52 | a different license, not a version of the Affero GPL, but Affero has 53 | released a new version of the Affero GPL which permits relicensing under 54 | this license. 55 | 56 | The precise terms and conditions for copying, distribution and 57 | modification follow. 58 | 59 | TERMS AND CONDITIONS 60 | 61 | 0. Definitions. 62 | 63 | "This License" refers to version 3 of the GNU Affero General Public License. 64 | 65 | "Copyright" also means copyright-like laws that apply to other kinds of 66 | works, such as semiconductor masks. 67 | 68 | "The Program" refers to any copyrightable work licensed under this 69 | License. Each licensee is addressed as "you". "Licensees" and 70 | "recipients" may be individuals or organizations. 71 | 72 | To "modify" a work means to copy from or adapt all or part of the work 73 | in a fashion requiring copyright permission, other than the making of an 74 | exact copy. The resulting work is called a "modified version" of the 75 | earlier work or a work "based on" the earlier work. 76 | 77 | A "covered work" means either the unmodified Program or a work based 78 | on the Program. 79 | 80 | To "propagate" a work means to do anything with it that, without 81 | permission, would make you directly or secondarily liable for 82 | infringement under applicable copyright law, except executing it on a 83 | computer or modifying a private copy. Propagation includes copying, 84 | distribution (with or without modification), making available to the 85 | public, and in some countries other activities as well. 86 | 87 | To "convey" a work means any kind of propagation that enables other 88 | parties to make or receive copies. Mere interaction with a user through 89 | a computer network, with no transfer of a copy, is not conveying. 90 | 91 | An interactive user interface displays "Appropriate Legal Notices" 92 | to the extent that it includes a convenient and prominently visible 93 | feature that (1) displays an appropriate copyright notice, and (2) 94 | tells the user that there is no warranty for the work (except to the 95 | extent that warranties are provided), that licensees may convey the 96 | work under this License, and how to view a copy of this License. If 97 | the interface presents a list of user commands or options, such as a 98 | menu, a prominent item in the list meets this criterion. 99 | 100 | 1. Source Code. 101 | 102 | The "source code" for a work means the preferred form of the work 103 | for making modifications to it. "Object code" means any non-source 104 | form of a work. 105 | 106 | A "Standard Interface" means an interface that either is an official 107 | standard defined by a recognized standards body, or, in the case of 108 | interfaces specified for a particular programming language, one that 109 | is widely used among developers working in that language. 110 | 111 | The "System Libraries" of an executable work include anything, other 112 | than the work as a whole, that (a) is included in the normal form of 113 | packaging a Major Component, but which is not part of that Major 114 | Component, and (b) serves only to enable use of the work with that 115 | Major Component, or to implement a Standard Interface for which an 116 | implementation is available to the public in source code form. A 117 | "Major Component", in this context, means a major essential component 118 | (kernel, window system, and so on) of the specific operating system 119 | (if any) on which the executable work runs, or a compiler used to 120 | produce the work, or an object code interpreter used to run it. 121 | 122 | The "Corresponding Source" for a work in object code form means all 123 | the source code needed to generate, install, and (for an executable 124 | work) run the object code and to modify the work, including scripts to 125 | control those activities. However, it does not include the work's 126 | System Libraries, or general-purpose tools or generally available free 127 | programs which are used unmodified in performing those activities but 128 | which are not part of the work. For example, Corresponding Source 129 | includes interface definition files associated with source files for 130 | the work, and the source code for shared libraries and dynamically 131 | linked subprograms that the work is specifically designed to require, 132 | such as by intimate data communication or control flow between those 133 | subprograms and other parts of the work. 134 | 135 | The Corresponding Source need not include anything that users 136 | can regenerate automatically from other parts of the Corresponding 137 | Source. 138 | 139 | The Corresponding Source for a work in source code form is that 140 | same work. 141 | 142 | 2. Basic Permissions. 143 | 144 | All rights granted under this License are granted for the term of 145 | copyright on the Program, and are irrevocable provided the stated 146 | conditions are met. This License explicitly affirms your unlimited 147 | permission to run the unmodified Program. The output from running a 148 | covered work is covered by this License only if the output, given its 149 | content, constitutes a covered work. This License acknowledges your 150 | rights of fair use or other equivalent, as provided by copyright law. 151 | 152 | You may make, run and propagate covered works that you do not 153 | convey, without conditions so long as your license otherwise remains 154 | in force. You may convey covered works to others for the sole purpose 155 | of having them make modifications exclusively for you, or provide you 156 | with facilities for running those works, provided that you comply with 157 | the terms of this License in conveying all material for which you do 158 | not control copyright. Those thus making or running the covered works 159 | for you must do so exclusively on your behalf, under your direction 160 | and control, on terms that prohibit them from making any copies of 161 | your copyrighted material outside their relationship with you. 162 | 163 | Conveying under any other circumstances is permitted solely under 164 | the conditions stated below. Sublicensing is not allowed; section 10 165 | makes it unnecessary. 166 | 167 | 3. Protecting Users' Legal Rights From Anti-Circumvention Law. 168 | 169 | No covered work shall be deemed part of an effective technological 170 | measure under any applicable law fulfilling obligations under article 171 | 11 of the WIPO copyright treaty adopted on 20 December 1996, or 172 | similar laws prohibiting or restricting circumvention of such 173 | measures. 174 | 175 | When you convey a covered work, you waive any legal power to forbid 176 | circumvention of technological measures to the extent such circumvention 177 | is effected by exercising rights under this License with respect to 178 | the covered work, and you disclaim any intention to limit operation or 179 | modification of the work as a means of enforcing, against the work's 180 | users, your or third parties' legal rights to forbid circumvention of 181 | technological measures. 182 | 183 | 4. Conveying Verbatim Copies. 184 | 185 | You may convey verbatim copies of the Program's source code as you 186 | receive it, in any medium, provided that you conspicuously and 187 | appropriately publish on each copy an appropriate copyright notice; 188 | keep intact all notices stating that this License and any 189 | non-permissive terms added in accord with section 7 apply to the code; 190 | keep intact all notices of the absence of any warranty; and give all 191 | recipients a copy of this License along with the Program. 192 | 193 | You may charge any price or no price for each copy that you convey, 194 | and you may offer support or warranty protection for a fee. 195 | 196 | 5. Conveying Modified Source Versions. 197 | 198 | You may convey a work based on the Program, or the modifications to 199 | produce it from the Program, in the form of source code under the 200 | terms of section 4, provided that you also meet all of these conditions: 201 | 202 | a) The work must carry prominent notices stating that you modified 203 | it, and giving a relevant date. 204 | 205 | b) The work must carry prominent notices stating that it is 206 | released under this License and any conditions added under section 207 | 7. This requirement modifies the requirement in section 4 to 208 | "keep intact all notices". 209 | 210 | c) You must license the entire work, as a whole, under this 211 | License to anyone who comes into possession of a copy. This 212 | License will therefore apply, along with any applicable section 7 213 | additional terms, to the whole of the work, and all its parts, 214 | regardless of how they are packaged. This License gives no 215 | permission to license the work in any other way, but it does not 216 | invalidate such permission if you have separately received it. 217 | 218 | d) If the work has interactive user interfaces, each must display 219 | Appropriate Legal Notices; however, if the Program has interactive 220 | interfaces that do not display Appropriate Legal Notices, your 221 | work need not make them do so. 222 | 223 | A compilation of a covered work with other separate and independent 224 | works, which are not by their nature extensions of the covered work, 225 | and which are not combined with it such as to form a larger program, 226 | in or on a volume of a storage or distribution medium, is called an 227 | "aggregate" if the compilation and its resulting copyright are not 228 | used to limit the access or legal rights of the compilation's users 229 | beyond what the individual works permit. Inclusion of a covered work 230 | in an aggregate does not cause this License to apply to the other 231 | parts of the aggregate. 232 | 233 | 6. Conveying Non-Source Forms. 234 | 235 | You may convey a covered work in object code form under the terms 236 | of sections 4 and 5, provided that you also convey the 237 | machine-readable Corresponding Source under the terms of this License, 238 | in one of these ways: 239 | 240 | a) Convey the object code in, or embodied in, a physical product 241 | (including a physical distribution medium), accompanied by the 242 | Corresponding Source fixed on a durable physical medium 243 | customarily used for software interchange. 244 | 245 | b) Convey the object code in, or embodied in, a physical product 246 | (including a physical distribution medium), accompanied by a 247 | written offer, valid for at least three years and valid for as 248 | long as you offer spare parts or customer support for that product 249 | model, to give anyone who possesses the object code either (1) a 250 | copy of the Corresponding Source for all the software in the 251 | product that is covered by this License, on a durable physical 252 | medium customarily used for software interchange, for a price no 253 | more than your reasonable cost of physically performing this 254 | conveying of source, or (2) access to copy the 255 | Corresponding Source from a network server at no charge. 256 | 257 | c) Convey individual copies of the object code with a copy of the 258 | written offer to provide the Corresponding Source. This 259 | alternative is allowed only occasionally and noncommercially, and 260 | only if you received the object code with such an offer, in accord 261 | with subsection 6b. 262 | 263 | d) Convey the object code by offering access from a designated 264 | place (gratis or for a charge), and offer equivalent access to the 265 | Corresponding Source in the same way through the same place at no 266 | further charge. You need not require recipients to copy the 267 | Corresponding Source along with the object code. If the place to 268 | copy the object code is a network server, the Corresponding Source 269 | may be on a different server (operated by you or a third party) 270 | that supports equivalent copying facilities, provided you maintain 271 | clear directions next to the object code saying where to find the 272 | Corresponding Source. Regardless of what server hosts the 273 | Corresponding Source, you remain obligated to ensure that it is 274 | available for as long as needed to satisfy these requirements. 275 | 276 | e) Convey the object code using peer-to-peer transmission, provided 277 | you inform other peers where the object code and Corresponding 278 | Source of the work are being offered to the general public at no 279 | charge under subsection 6d. 280 | 281 | A separable portion of the object code, whose source code is excluded 282 | from the Corresponding Source as a System Library, need not be 283 | included in conveying the object code work. 284 | 285 | A "User Product" is either (1) a "consumer product", which means any 286 | tangible personal property which is normally used for personal, family, 287 | or household purposes, or (2) anything designed or sold for incorporation 288 | into a dwelling. In determining whether a product is a consumer product, 289 | doubtful cases shall be resolved in favor of coverage. For a particular 290 | product received by a particular user, "normally used" refers to a 291 | typical or common use of that class of product, regardless of the status 292 | of the particular user or of the way in which the particular user 293 | actually uses, or expects or is expected to use, the product. A product 294 | is a consumer product regardless of whether the product has substantial 295 | commercial, industrial or non-consumer uses, unless such uses represent 296 | the only significant mode of use of the product. 297 | 298 | "Installation Information" for a User Product means any methods, 299 | procedures, authorization keys, or other information required to install 300 | and execute modified versions of a covered work in that User Product from 301 | a modified version of its Corresponding Source. The information must 302 | suffice to ensure that the continued functioning of the modified object 303 | code is in no case prevented or interfered with solely because 304 | modification has been made. 305 | 306 | If you convey an object code work under this section in, or with, or 307 | specifically for use in, a User Product, and the conveying occurs as 308 | part of a transaction in which the right of possession and use of the 309 | User Product is transferred to the recipient in perpetuity or for a 310 | fixed term (regardless of how the transaction is characterized), the 311 | Corresponding Source conveyed under this section must be accompanied 312 | by the Installation Information. But this requirement does not apply 313 | if neither you nor any third party retains the ability to install 314 | modified object code on the User Product (for example, the work has 315 | been installed in ROM). 316 | 317 | The requirement to provide Installation Information does not include a 318 | requirement to continue to provide support service, warranty, or updates 319 | for a work that has been modified or installed by the recipient, or for 320 | the User Product in which it has been modified or installed. Access to a 321 | network may be denied when the modification itself materially and 322 | adversely affects the operation of the network or violates the rules and 323 | protocols for communication across the network. 324 | 325 | Corresponding Source conveyed, and Installation Information provided, 326 | in accord with this section must be in a format that is publicly 327 | documented (and with an implementation available to the public in 328 | source code form), and must require no special password or key for 329 | unpacking, reading or copying. 330 | 331 | 7. Additional Terms. 332 | 333 | "Additional permissions" are terms that supplement the terms of this 334 | License by making exceptions from one or more of its conditions. 335 | Additional permissions that are applicable to the entire Program shall 336 | be treated as though they were included in this License, to the extent 337 | that they are valid under applicable law. If additional permissions 338 | apply only to part of the Program, that part may be used separately 339 | under those permissions, but the entire Program remains governed by 340 | this License without regard to the additional permissions. 341 | 342 | When you convey a copy of a covered work, you may at your option 343 | remove any additional permissions from that copy, or from any part of 344 | it. (Additional permissions may be written to require their own 345 | removal in certain cases when you modify the work.) You may place 346 | additional permissions on material, added by you to a covered work, 347 | for which you have or can give appropriate copyright permission. 348 | 349 | Notwithstanding any other provision of this License, for material you 350 | add to a covered work, you may (if authorized by the copyright holders of 351 | that material) supplement the terms of this License with terms: 352 | 353 | a) Disclaiming warranty or limiting liability differently from the 354 | terms of sections 15 and 16 of this License; or 355 | 356 | b) Requiring preservation of specified reasonable legal notices or 357 | author attributions in that material or in the Appropriate Legal 358 | Notices displayed by works containing it; or 359 | 360 | c) Prohibiting misrepresentation of the origin of that material, or 361 | requiring that modified versions of such material be marked in 362 | reasonable ways as different from the original version; or 363 | 364 | d) Limiting the use for publicity purposes of names of licensors or 365 | authors of the material; or 366 | 367 | e) Declining to grant rights under trademark law for use of some 368 | trade names, trademarks, or service marks; or 369 | 370 | f) Requiring indemnification of licensors and authors of that 371 | material by anyone who conveys the material (or modified versions of 372 | it) with contractual assumptions of liability to the recipient, for 373 | any liability that these contractual assumptions directly impose on 374 | those licensors and authors. 375 | 376 | All other non-permissive additional terms are considered "further 377 | restrictions" within the meaning of section 10. If the Program as you 378 | received it, or any part of it, contains a notice stating that it is 379 | governed by this License along with a term that is a further 380 | restriction, you may remove that term. If a license document contains 381 | a further restriction but permits relicensing or conveying under this 382 | License, you may add to a covered work material governed by the terms 383 | of that license document, provided that the further restriction does 384 | not survive such relicensing or conveying. 385 | 386 | If you add terms to a covered work in accord with this section, you 387 | must place, in the relevant source files, a statement of the 388 | additional terms that apply to those files, or a notice indicating 389 | where to find the applicable terms. 390 | 391 | Additional terms, permissive or non-permissive, may be stated in the 392 | form of a separately written license, or stated as exceptions; 393 | the above requirements apply either way. 394 | 395 | 8. Termination. 396 | 397 | You may not propagate or modify a covered work except as expressly 398 | provided under this License. Any attempt otherwise to propagate or 399 | modify it is void, and will automatically terminate your rights under 400 | this License (including any patent licenses granted under the third 401 | paragraph of section 11). 402 | 403 | However, if you cease all violation of this License, then your 404 | license from a particular copyright holder is reinstated (a) 405 | provisionally, unless and until the copyright holder explicitly and 406 | finally terminates your license, and (b) permanently, if the copyright 407 | holder fails to notify you of the violation by some reasonable means 408 | prior to 60 days after the cessation. 409 | 410 | Moreover, your license from a particular copyright holder is 411 | reinstated permanently if the copyright holder notifies you of the 412 | violation by some reasonable means, this is the first time you have 413 | received notice of violation of this License (for any work) from that 414 | copyright holder, and you cure the violation prior to 30 days after 415 | your receipt of the notice. 416 | 417 | Termination of your rights under this section does not terminate the 418 | licenses of parties who have received copies or rights from you under 419 | this License. If your rights have been terminated and not permanently 420 | reinstated, you do not qualify to receive new licenses for the same 421 | material under section 10. 422 | 423 | 9. Acceptance Not Required for Having Copies. 424 | 425 | You are not required to accept this License in order to receive or 426 | run a copy of the Program. Ancillary propagation of a covered work 427 | occurring solely as a consequence of using peer-to-peer transmission 428 | to receive a copy likewise does not require acceptance. However, 429 | nothing other than this License grants you permission to propagate or 430 | modify any covered work. These actions infringe copyright if you do 431 | not accept this License. Therefore, by modifying or propagating a 432 | covered work, you indicate your acceptance of this License to do so. 433 | 434 | 10. Automatic Licensing of Downstream Recipients. 435 | 436 | Each time you convey a covered work, the recipient automatically 437 | receives a license from the original licensors, to run, modify and 438 | propagate that work, subject to this License. You are not responsible 439 | for enforcing compliance by third parties with this License. 440 | 441 | An "entity transaction" is a transaction transferring control of an 442 | organization, or substantially all assets of one, or subdividing an 443 | organization, or merging organizations. If propagation of a covered 444 | work results from an entity transaction, each party to that 445 | transaction who receives a copy of the work also receives whatever 446 | licenses to the work the party's predecessor in interest had or could 447 | give under the previous paragraph, plus a right to possession of the 448 | Corresponding Source of the work from the predecessor in interest, if 449 | the predecessor has it or can get it with reasonable efforts. 450 | 451 | You may not impose any further restrictions on the exercise of the 452 | rights granted or affirmed under this License. For example, you may 453 | not impose a license fee, royalty, or other charge for exercise of 454 | rights granted under this License, and you may not initiate litigation 455 | (including a cross-claim or counterclaim in a lawsuit) alleging that 456 | any patent claim is infringed by making, using, selling, offering for 457 | sale, or importing the Program or any portion of it. 458 | 459 | 11. Patents. 460 | 461 | A "contributor" is a copyright holder who authorizes use under this 462 | License of the Program or a work on which the Program is based. The 463 | work thus licensed is called the contributor's "contributor version". 464 | 465 | A contributor's "essential patent claims" are all patent claims 466 | owned or controlled by the contributor, whether already acquired or 467 | hereafter acquired, that would be infringed by some manner, permitted 468 | by this License, of making, using, or selling its contributor version, 469 | but do not include claims that would be infringed only as a 470 | consequence of further modification of the contributor version. For 471 | purposes of this definition, "control" includes the right to grant 472 | patent sublicenses in a manner consistent with the requirements of 473 | this License. 474 | 475 | Each contributor grants you a non-exclusive, worldwide, royalty-free 476 | patent license under the contributor's essential patent claims, to 477 | make, use, sell, offer for sale, import and otherwise run, modify and 478 | propagate the contents of its contributor version. 479 | 480 | In the following three paragraphs, a "patent license" is any express 481 | agreement or commitment, however denominated, not to enforce a patent 482 | (such as an express permission to practice a patent or covenant not to 483 | sue for patent infringement). To "grant" such a patent license to a 484 | party means to make such an agreement or commitment not to enforce a 485 | patent against the party. 486 | 487 | If you convey a covered work, knowingly relying on a patent license, 488 | and the Corresponding Source of the work is not available for anyone 489 | to copy, free of charge and under the terms of this License, through a 490 | publicly available network server or other readily accessible means, 491 | then you must either (1) cause the Corresponding Source to be so 492 | available, or (2) arrange to deprive yourself of the benefit of the 493 | patent license for this particular work, or (3) arrange, in a manner 494 | consistent with the requirements of this License, to extend the patent 495 | license to downstream recipients. "Knowingly relying" means you have 496 | actual knowledge that, but for the patent license, your conveying the 497 | covered work in a country, or your recipient's use of the covered work 498 | in a country, would infringe one or more identifiable patents in that 499 | country that you have reason to believe are valid. 500 | 501 | If, pursuant to or in connection with a single transaction or 502 | arrangement, you convey, or propagate by procuring conveyance of, a 503 | covered work, and grant a patent license to some of the parties 504 | receiving the covered work authorizing them to use, propagate, modify 505 | or convey a specific copy of the covered work, then the patent license 506 | you grant is automatically extended to all recipients of the covered 507 | work and works based on it. 508 | 509 | A patent license is "discriminatory" if it does not include within 510 | the scope of its coverage, prohibits the exercise of, or is 511 | conditioned on the non-exercise of one or more of the rights that are 512 | specifically granted under this License. You may not convey a covered 513 | work if you are a party to an arrangement with a third party that is 514 | in the business of distributing software, under which you make payment 515 | to the third party based on the extent of your activity of conveying 516 | the work, and under which the third party grants, to any of the 517 | parties who would receive the covered work from you, a discriminatory 518 | patent license (a) in connection with copies of the covered work 519 | conveyed by you (or copies made from those copies), or (b) primarily 520 | for and in connection with specific products or compilations that 521 | contain the covered work, unless you entered into that arrangement, 522 | or that patent license was granted, prior to 28 March 2007. 523 | 524 | Nothing in this License shall be construed as excluding or limiting 525 | any implied license or other defenses to infringement that may 526 | otherwise be available to you under applicable patent law. 527 | 528 | 12. No Surrender of Others' Freedom. 529 | 530 | If conditions are imposed on you (whether by court order, agreement or 531 | otherwise) that contradict the conditions of this License, they do not 532 | excuse you from the conditions of this License. If you cannot convey a 533 | covered work so as to satisfy simultaneously your obligations under this 534 | License and any other pertinent obligations, then as a consequence you may 535 | not convey it at all. For example, if you agree to terms that obligate you 536 | to collect a royalty for further conveying from those to whom you convey 537 | the Program, the only way you could satisfy both those terms and this 538 | License would be to refrain entirely from conveying the Program. 539 | 540 | 13. Remote Network Interaction; Use with the GNU General Public License. 541 | 542 | Notwithstanding any other provision of this License, if you modify the 543 | Program, your modified version must prominently offer all users 544 | interacting with it remotely through a computer network (if your version 545 | supports such interaction) an opportunity to receive the Corresponding 546 | Source of your version by providing access to the Corresponding Source 547 | from a network server at no charge, through some standard or customary 548 | means of facilitating copying of software. This Corresponding Source 549 | shall include the Corresponding Source for any work covered by version 3 550 | of the GNU General Public License that is incorporated pursuant to the 551 | following paragraph. 552 | 553 | Notwithstanding any other provision of this License, you have 554 | permission to link or combine any covered work with a work licensed 555 | under version 3 of the GNU General Public License into a single 556 | combined work, and to convey the resulting work. The terms of this 557 | License will continue to apply to the part which is the covered work, 558 | but the work with which it is combined will remain governed by version 559 | 3 of the GNU General Public License. 560 | 561 | 14. Revised Versions of this License. 562 | 563 | The Free Software Foundation may publish revised and/or new versions of 564 | the GNU Affero General Public License from time to time. Such new versions 565 | will be similar in spirit to the present version, but may differ in detail to 566 | address new problems or concerns. 567 | 568 | Each version is given a distinguishing version number. If the 569 | Program specifies that a certain numbered version of the GNU Affero General 570 | Public License "or any later version" applies to it, you have the 571 | option of following the terms and conditions either of that numbered 572 | version or of any later version published by the Free Software 573 | Foundation. If the Program does not specify a version number of the 574 | GNU Affero General Public License, you may choose any version ever published 575 | by the Free Software Foundation. 576 | 577 | If the Program specifies that a proxy can decide which future 578 | versions of the GNU Affero General Public License can be used, that proxy's 579 | public statement of acceptance of a version permanently authorizes you 580 | to choose that version for the Program. 581 | 582 | Later license versions may give you additional or different 583 | permissions. However, no additional obligations are imposed on any 584 | author or copyright holder as a result of your choosing to follow a 585 | later version. 586 | 587 | 15. Disclaimer of Warranty. 588 | 589 | THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY 590 | APPLICABLE LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT 591 | HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY 592 | OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, 593 | THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 594 | PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM 595 | IS WITH YOU. SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF 596 | ALL NECESSARY SERVICING, REPAIR OR CORRECTION. 597 | 598 | 16. Limitation of Liability. 599 | 600 | IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING 601 | WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS 602 | THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY 603 | GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE 604 | USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF 605 | DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD 606 | PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS), 607 | EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF 608 | SUCH DAMAGES. 609 | 610 | 17. Interpretation of Sections 15 and 16. 611 | 612 | If the disclaimer of warranty and limitation of liability provided 613 | above cannot be given local legal effect according to their terms, 614 | reviewing courts shall apply local law that most closely approximates 615 | an absolute waiver of all civil liability in connection with the 616 | Program, unless a warranty or assumption of liability accompanies a 617 | copy of the Program in return for a fee. 618 | 619 | END OF TERMS AND CONDITIONS 620 | 621 | How to Apply These Terms to Your New Programs 622 | 623 | If you develop a new program, and you want it to be of the greatest 624 | possible use to the public, the best way to achieve this is to make it 625 | free software which everyone can redistribute and change under these terms. 626 | 627 | To do so, attach the following notices to the program. It is safest 628 | to attach them to the start of each source file to most effectively 629 | state the exclusion of warranty; and each file should have at least 630 | the "copyright" line and a pointer to where the full notice is found. 631 | 632 | 633 | Copyright (C) 634 | 635 | This program is free software: you can redistribute it and/or modify 636 | it under the terms of the GNU Affero General Public License as published 637 | by the Free Software Foundation, either version 3 of the License, or 638 | (at your option) any later version. 639 | 640 | This program is distributed in the hope that it will be useful, 641 | but WITHOUT ANY WARRANTY; without even the implied warranty of 642 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 643 | GNU Affero General Public License for more details. 644 | 645 | You should have received a copy of the GNU Affero General Public License 646 | along with this program. If not, see . 647 | 648 | Also add information on how to contact you by electronic and paper mail. 649 | 650 | If your software can interact with users remotely through a computer 651 | network, you should also make sure that it provides a way for users to 652 | get its source. For example, if your program is a web application, its 653 | interface could display a "Source" link that leads users to an archive 654 | of the code. There are many ways you could offer source, and different 655 | solutions will be better for different programs; see section 13 for the 656 | specific requirements. 657 | 658 | You should also get your employer (if you work as a programmer) or school, 659 | if any, to sign a "copyright disclaimer" for the program, if necessary. 660 | For more information on this, and how to apply and follow the GNU AGPL, see 661 | . 662 | --------------------------------------------------------------------------------