def __getitem__(self, index):
# Read audio
filename = self.audio_files[index]
audio_mel, sampling_rate = load_wav_to_torch(self.dir_normal + '/' +
filename)
audio, sampling_rate = load_wav_to_torch(self.dir_hi + '/' + filename)
if sampling_rate != self.sampling_rate:
raise ValueError("{} SR doesn't match target {} SR".format(
sampling_rate, self.sampling_rate))
# Take segment
if audio.size(0) >= self.segment_length:
max_audio_start = audio.size(0) - self.segment_length
audio_start = random.randint(0, max_audio_start)
audio = audio[audio_start:audio_start + self.segment_length]
audio_mel = audio_mel[audio_start:audio_start +
self.segment_length]
else:
audio = torch.nn.functional.pad(
audio, (0, self.segment_length - audio.size(0)),
'constant').data
audio_mel = torch.nn.functional.pad(
audio_mel, (0, self.segment_length - audio.size(0)),
'constant').data
mel = self.get_mel(audio_mel)
audio = audio / MAX_WAV_VALUE
return (mel, audio)
def load_buffer(self):
num_files = len(self.audio_files)
for i in tqdm(range(num_files)):
filename = self.audio_files[i]
audio, sampling_rate = ms.load_wav_to_torch(filename)
if sampling_rate != self.sampling_rate:
raise ValueError("{} SR doesn't match target {} SR".format(
sampling_rate, self.sampling_rate))
self.all_length += audio.size(0)
if i == 0:
self.all_audio = audio
else:
self.all_audio = torch.cat((self.all_audio, audio), 0)
print("All audio has been loaded, totally length: {}".format(
self.all_length))
def DNN_test(checkpoint_path, filename):
model = DNNnet(**DNN_net_config)
checkpoint_dict = torch.load(checkpoint_path, map_location='cpu')
model_for_loading = checkpoint_dict['model']
model.load_state_dict(model_for_loading.state_dict())
print("Loaded checkpoint '{}' ".format(checkpoint_path))
model.eval()
Data_gen = spec2load(**DNN_data_config)
audio, sr = ms.load_wav_to_torch(filename)
# Take segment
if audio.size(0) >= 2240:
max_audio_start = audio.size(0) - 2240
audio_start = random.randint(0, max_audio_start)
audio = audio[audio_start:audio_start + 2240]
else:
audio = torch.nn.functional.pad(audio, (2240 - audio.size(0), 0),
'constant').data
spec = Data_gen.get_spec(audio)
feed_spec = spec[:, :-2]
targ_spec = spec[:, -2:]
feed_spec = feed_spec.unsqueeze(0)
gener_spec = model.forward(feed_spec)
# gener_linear = gener_linear.squeeze().view(-1, 2).T
gener_spec = gener_spec.detach().numpy()
targ_spec = torch.cat((targ_spec[:, 0], targ_spec[:, 1]), 0)
targ_spec = targ_spec.numpy()
# targ_mel_db = librosa.power_to_db(targ_mel[0], ref=np.max)
# gener_mel_db = librosa.power_to_db(gener_mel[0], ref=np.max)
plt.figure()
plt.plot(targ_spec)
# librosa.display.specshow(targ_mel, x_axis='time', y_axis='mel')
plt.plot(gener_spec[0], 'r')
# librosa.display.specshow(gener_mel, x_axis='time', y_axis='mel')
plt.show()
def validate(model, loader_STFT, STFTs, logger, iteration, validation_files, speaker_lookup, sigma, output_directory, data_config, save_audio=True, max_length_s= 3, files_to_process=9):
print("Validating... ", end="")
from mel2samp import DTW
val_sigma = sigma * 1.00
model.eval()
val_start_time = time.time()
STFT_elapsed = samples_processed = 0
with torch.no_grad():
with torch.random.fork_rng(devices=[0,]):
torch.random.manual_seed(0)# use same Z / random seed during validation so results are more consistent and comparable.
with open(validation_files, encoding='utf-8') as f:
audiopaths_and_melpaths = [line.strip().split('|') for line in f]
if next(model.parameters()).type() == "torch.cuda.HalfTensor":
model_type = "half"
else:
model_type = "float"
timestr = time.strftime("%Y_%m_%d-%H_%M_%S")
total_MAE = total_MSE = total = files_processed = 0
input_mels = []
gt_mels = []
pred_mels = []
MAE_specs = []
for i, (audiopath, melpath, *remaining) in enumerate(audiopaths_and_melpaths):
if files_processed >= files_to_process: # number of validation files to run.
break
audio = load_wav_to_torch(audiopath)[0]/32768.0 # load audio from wav file to tensor
if audio.shape[0] > (data_config['sampling_rate']*max_length_s):
continue # ignore audio over max_length_seconds
gt_mel = loader_STFT.mel_spectrogram(audio.unsqueeze(0)).cuda()# [T] -> [1, T] -> [1, n_mel, T]
if 'load_hidden_from_disk' in data_config.keys() and data_config['load_hidden_from_disk']:
mel = None
hidden_path = remaining[1]
model_input = np.load(hidden_path) # load tacotron hidden from file into numpy arr
model_input = torch.from_numpy(model_input).unsqueeze(0).cuda() # from numpy arr to tensor on GPU
else:
if loader_STFT and data_config['load_mel_from_disk'] < 0.2:
mel = None
model_input = gt_mel.clone()
else:
mel = np.load(melpath) # load mel from file into numpy arr
mel = torch.from_numpy(mel).unsqueeze(0).cuda() # from numpy arr to tensor on GPU
assert mel[:, :gt_mel.shape[1], :].shape == gt_mel.shape, f'shapes of {mel[:, :gt_mel.shape[1], :].shape} and {gt_mel.shape} do not match.'
#if torch.nn.functional.mse_loss(mel[:, :gt_mel.shape[1], :], gt_mel) > 0.3:
# continue # skip validation files that significantly vary from the target.
if model.has_logvar_channels:
if mel.shape[1] == model.n_mel_channels*2:
mel, logvar = mel.chunk(2, dim=1)# [1, n_mel*2, T] -> [1, n_mel, T], [1, n_mel, T]
mel = DTW(mel, gt_mel, scale_factor=8, range_=5)
model_input = torch.cat((mel, logvar), dim=1)
else:
raise Exception("Loaded mel from disk has wrong shape.")
else:
if mel.shape[1] == model.n_mel_channels*2:
mel = mel.chunk(2, dim=1)[0]# [1, n_mel*2, T] -> [1, n_mel, T]
#mel = DTW(mel, gt_mel, scale_factor=8, range_=5)
model_input = mel
if hasattr(model, 'multispeaker') and model.multispeaker == True:
assert len(remaining), f"Speaker ID missing while multispeaker == True.\nLine: {i}\n'{'|'.join([autiopath, melpath])}'"
speaker_id = remaining[0]
speaker_id = torch.IntTensor([speaker_lookup[int(speaker_id)],])
speaker_id = speaker_id.cuda(non_blocking=True).long()
else:
speaker_id = None
if model_type == "half":
model_input = model_input.half() # for fp16 training
audio_waveglow = model.infer(model_input, speaker_id, sigma=val_sigma).cpu().float()
audio = audio.squeeze().unsqueeze(0) # crush extra dimensions and shape for STFT
audio_waveglow = audio_waveglow.squeeze().unsqueeze(0).clamp(min=-0.999, max=0.999) # [1, T] crush extra dimensions and shape for STFT
audio_waveglow[torch.isnan(audio_waveglow) | torch.isinf(audio_waveglow)] = 0.0 # and clamp any values over/under |1.0| (which should only exist very early in training)
STFT_start_time = time.time()
for j, STFT in enumerate(STFTs): # check Spectrogram Error with multiple window sizes
input_mels.append(mel)
mel_GT = STFT.mel_spectrogram(audio)# [1, T] -> [1, n_mel, T//hop_len]
gt_mels.append(mel_GT)
mel_waveglow = STFT.mel_spectrogram(audio_waveglow)[:,:,:mel_GT.shape[-1]]# [1, T] -> [1, n_mel, T//hop_len]
pred_mels.append(mel_waveglow)
MSE = (torch.nn.MSELoss()(mel_waveglow, mel_GT)).item() # get MSE (Mean Squared Error) between Ground Truth and WaveGlow inferred spectrograms.
MAE_spec = torch.nn.L1Loss(reduction='none')(mel_waveglow, mel_GT)
MAE = (MAE_spec.mean()).item() # get MAE (Mean Absolute Error) between Ground Truth and WaveGlow inferred spectrograms.
MAE_specs.append(MAE_spec)
total_MAE+=MAE
total_MSE+=MSE
total+=1
STFT_elapsed += time.time()-STFT_start_time
if save_audio:
audio_path = os.path.join(output_directory, "samples", str(iteration)+"-"+timestr, os.path.basename(audiopath)) # Write audio to checkpoint_directory/iteration/audiofilename.wav
os.makedirs(os.path.join(output_directory, "samples", str(iteration)+"-"+timestr), exist_ok=True)
sf.write(audio_path, audio_waveglow.squeeze().cpu().numpy(), data_config['sampling_rate'], "PCM_16") # save waveglow sample
audio_path = os.path.join(output_directory, "samples", "Ground Truth", os.path.basename(audiopath)) # Write audio to checkpoint_directory/iteration/audiofilename.wav
if not os.path.exists(audio_path):
os.makedirs(os.path.join(output_directory, "samples", "Ground Truth"), exist_ok=True)
sf.write(audio_path, audio.squeeze().cpu().numpy(), data_config['sampling_rate'], "PCM_16") # save ground truth
files_processed+=1
samples_processed+=audio_waveglow.shape[-1]
if total:
average_MSE = total_MSE/total
average_MAE = total_MAE/total
logger.add_scalar('val_MSE', average_MSE, iteration)
logger.add_scalar('val_MAE', average_MAE, iteration)
for idx, (gt_mel, pred_mel, input_mel, mae_mel) in enumerate(zip(gt_mels[-6:], pred_mels[-6:], input_mels[-6:], MAE_specs[-6:])):
logger.add_image(f'mel_{idx}/pred',
plot_spectrogram_to_numpy(pred_mel[0].data.cpu().numpy(), range=[-11.5, 2.0]),
iteration, dataformats='HWC')
if mae_mel is not None:
logger.add_image(f'mel_{idx}/zmae',
plot_spectrogram_to_numpy(mae_mel[0].data.cpu().numpy(), range=[0.0, 2.5]),
iteration, dataformats='HWC')
if iteration % 10000 == 0: # target doesn't change unless batch size or dataset changes so only needs to be plotted once in a while.
logger.add_image(f'mel_{idx}/target',
plot_spectrogram_to_numpy(gt_mel[0].data.cpu().numpy(), range=[-11.5, 2.0]),
iteration, dataformats='HWC')
if input_mel is not None:
logger.add_image(f'mel_{idx}/input',
plot_spectrogram_to_numpy(input_mel[0].data.cpu().numpy(), range=[-11.5, 2.0]),
iteration, dataformats='HWC')
time_elapsed = time.time()-val_start_time
time_elapsed_without_stft = time_elapsed-STFT_elapsed
samples_per_second = samples_processed/time_elapsed_without_stft
print(f"[Avg MSE: {average_MSE:.6f} MAE: {average_MAE:.6f}]",
f"[{time_elapsed_without_stft:.3f}s]",
f"[{time_elapsed:.3f}s_stft]",
f"[{time_elapsed_without_stft/files_processed:.3f}s/file]",
f"[{samples_per_second/data_config['sampling_rate']:.3f}rtf]",
f"[{samples_per_second:,.0f}samples/s]"
)
logger.add_scalar('val_rtf', samples_per_second/data_config['sampling_rate'], iteration)
else:
average_MSE = 1e3
average_MAE = 1e3
print("Average MSE: N/A", "Average MAE: N/A")
for convinv in model.convinv:
if hasattr(convinv, 'W_inverse'):
delattr(convinv, "W_inverse") # clear Inverse Weights.
if hasattr(model, 'iso226'):
delattr(model, 'iso226')
mel = speaker_id = None # del GPU based tensors.
torch.cuda.empty_cache() # clear cache for next training
model.train()
return average_MSE, average_MAE
def main(files, waveglow_path, sigma, output_dir, sampling_rate, is_fp16,
denoiser_strength, args):
#mel_files = files_to_list(mel_files)
#print(mel_files)
files = ['/local-scratch/fuyang/cmpt726/final_project/cremad/1091_WSI_SAD_XX.wav']
#files = ['/local-scratch/fuyang/cmpt726/waveglow/data/LJSpeech-1.1/LJ001-0001.wav']
with open('config.json') as f:
data = f.read()
config = json.loads(data)
waveglow_config = config["waveglow_config"]
model = WaveGlow(**waveglow_config)
checkpoint_dict = torch.load('waveglow_256channels_universal_v5.pt', map_location='cpu')
model_for_loading = checkpoint_dict['model']
model.load_state_dict(model_for_loading.state_dict())
model.cuda()
#waveglow = torch.load(waveglow_path)['model']
#waveglow = waveglow.remove_weightnorm(waveglow)
#waveglow.cuda()
waveglow = model
if is_fp16:
from apex import amp
waveglow, _ = amp.initialize(waveglow, [], opt_level="O1")
if denoiser_strength > 0:
denoiser = Denoiser(waveglow).cuda()
mel_extractor = Get_mel(1024, 256, 1024, args.sampling_rate, 0.0, 8000.0)
for i, file_path in enumerate(files):
audio, rate = load_wav_to_torch(file_path)
if rate != sampling_rate:
audio = resampy.resample(audio.numpy(), rate, sampling_rate)
audio = torch.from_numpy(audio).float()
#if audio.size(0) >= args.segment_length:
# max_audio_start = audio.size(0) - args.segment_length
# audio_start = random.randint(0, max_audio_start)
# audio = audio[audio_start:audio_start+args.segment_length]
#else:
# audio = torch.nn.functional.pad(audio, (0, args.segment_length-audio.size(0)), 'constant').data
mel = mel_extractor.get_mel(audio)
audio = audio / MAX_WAV_VALUE
mel = torch.autograd.Variable(mel.cuda().unsqueeze(0))
audio = torch.autograd.Variable(audio.cuda().unsqueeze(0))
audio = audio.half() if is_fp16 else audio
mel = mel.half() if is_fp16 else mel
outputs = waveglow((mel, audio))
z = outputs[0][:,4:]
print(outputs)
mel_up = waveglow.upsample(mel)
time_cutoff = waveglow.upsample.kernel_size[0]-waveglow.upsample.stride[0]
mel_up = mel_up[:,:,:-time_cutoff]
#mel_up = mel_up[:,:,:-(time_cutoff+128)]
mel_up = mel_up.unfold(2, waveglow.n_group, waveglow.n_group).permute(0,2,1,3)
mel_up = mel_up.contiguous().view(mel_up.size(0), mel_up.size(1), -1).permute(0, 2, 1)
audio = z
mel_up = mel_up[:,:,:audio.size(2)]
sigma = 0.7
z_i = 0
for k in reversed(range(waveglow.n_flows)):
n_half = int(audio.size(1)/2)
audio_0 = audio[:,:n_half, :]
audio_1 = audio[:, n_half:, :]
output = waveglow.WN[k]((audio_0, mel_up))
s = output[:,n_half:, :]
b = output[:, :n_half, :]
audio_1 = (audio_1-b)/torch.exp(s)
audio = torch.cat([audio_0, audio_1],1)
audio = waveglow.convinv[k](audio, reverse=True)
if k % waveglow.n_early_every == 0 and k > 0:
z = outputs[0][:, 2-z_i:4-z_i]
#if mel_up.type() == 'torch.cuda.HalfTensor':
# z = torch.cuda.HalfTensor(mel_up.size(0), waveglow.n_early_size, mel_up.size(2)).normal_()
#else:
# z = torch.cuda.FloatTensor(mel_up.size(0), waveglow.n_early_size, mel_up.size(2)).normal_()
audio = torch.cat((sigma*z, audio),1)
audio = audio.permute(0,2,1).contiguous().view(audio.size(0), -1).data
audio = audio * MAX_WAV_VALUE
audio = audio.squeeze()
audio = audio.cpu().numpy()
audio = audio.astype('int16')
audio_path = os.path.join(
output_dir, "{}_synthesis.wav".format('fuyangz'))
write(audio_path, sampling_rate, audio)
print(audio_path)
def main(style, waveglow_path, sigma, output_dir, sampling_rate, is_fp16,
denoiser_strength, args):
#mel_files = files_to_list(mel_files)
#print(mel_files)
dataset = voice_dataset(dataBase={
'ravdess': './our_data/ravdess',
'cremad': './our_data/cremad'
},
style=('happy', 'sad', 'angry'))
#print(len(dataset.final_data['happy']))
#sample = dataset.pick_one_random_sample('happy')
files = dataset.final_data[style]
#files = ['/local-scratch/fuyang/cmpt726/waveglow/data/LJSpeech-1.1/LJ001-0001.wav']
with open('config.json') as f:
data = f.read()
config = json.loads(data)
waveglow_config = config["waveglow_config"]
model = WaveGlow(**waveglow_config)
checkpoint_dict = torch.load('waveglow_256channels_universal_v5.pt',
map_location='cpu')
model_for_loading = checkpoint_dict['model']
model.load_state_dict(model_for_loading.state_dict())
model.cuda()
waveglow = model
if is_fp16:
from apex import amp
waveglow, _ = amp.initialize(waveglow, [], opt_level="O1")
if denoiser_strength > 0:
denoiser = Denoiser(waveglow).cuda()
mel_extractor = Get_mel(1024, 256, 1024, args.sampling_rate, 0.0, 8000.0)
avg_z = np.zeros(8)
_count = 0
for i, (_, file_path) in enumerate(files):
if i > 50:
break
try:
audio, rate = load_wav_to_torch(file_path)
if rate != sampling_rate:
audio = resampy.resample(audio.numpy(), rate, sampling_rate)
audio = torch.from_numpy(audio).float()
#if audio.size(0) >= args.segment_length:
# max_audio_start = audio.size(0) - args.segment_length
# audio_start = random.randint(0, max_audio_start)
# audio = audio[audio_start:audio_start+args.segment_length]
#else:
# audio = torch.nn.functional.pad(audio, (0, args.segment_length-audio.size(0)), 'constant').data
mel = mel_extractor.get_mel(audio)
audio = audio / MAX_WAV_VALUE
mel = torch.autograd.Variable(mel.cuda().unsqueeze(0))
audio = torch.autograd.Variable(audio.cuda().unsqueeze(0))
audio = audio.half() if is_fp16 else audio
mel = mel.half() if is_fp16 else mel
outputs = waveglow((mel, audio))
avg_z += outputs[0].squeeze(0).mean(1).detach().cpu().numpy()
_count += 1
z = outputs[0][:, 4:]
#print(outputs)
mel_up = waveglow.upsample(mel)
time_cutoff = waveglow.upsample.kernel_size[
0] - waveglow.upsample.stride[0]
mel_up = mel_up[:, :, :-time_cutoff]
#mel_up = mel_up[:,:,:-(time_cutoff+128)]
mel_up = mel_up.unfold(2, waveglow.n_group,
waveglow.n_group).permute(0, 2, 1, 3)
mel_up = mel_up.contiguous().view(mel_up.size(0), mel_up.size(1),
-1).permute(0, 2, 1)
audio = z
mel_up = mel_up[:, :, :audio.size(2)]
sigma = 0.7
z_i = 0
for k in reversed(range(waveglow.n_flows)):
n_half = int(audio.size(1) / 2)
audio_0 = audio[:, :n_half, :]
audio_1 = audio[:, n_half:, :]
output = waveglow.WN[k]((audio_0, mel_up))
s = output[:, n_half:, :]
b = output[:, :n_half, :]
audio_1 = (audio_1 - b) / torch.exp(s)
audio = torch.cat([audio_0, audio_1], 1)
audio = waveglow.convinv[k](audio, reverse=True)
if k % waveglow.n_early_every == 0 and k > 0:
z = outputs[0][:, 2 - z_i:4 - z_i]
#if mel_up.type() == 'torch.cuda.HalfTensor':
# z = torch.cuda.HalfTensor(mel_up.size(0), waveglow.n_early_size, mel_up.size(2)).normal_()
#else:
# z = torch.cuda.FloatTensor(mel_up.size(0), waveglow.n_early_size, mel_up.size(2)).normal_()
audio = torch.cat((sigma * z, audio), 1)
audio = audio.permute(0, 2,
1).contiguous().view(audio.size(0), -1).data
audio = audio * MAX_WAV_VALUE
audio = audio.squeeze()
audio = audio.cpu().numpy()
audio = audio.astype('int16')
audio_path = os.path.join(
output_dir, "{}_synthesis.wav".format(file_path[:-4]))
if os.path.exists(
os.path.join(*audio_path.split('/')[:-1])) is False:
os.makedirs(os.path.join(*audio_path.split('/')[:-1]),
exist_ok=True)
write(audio_path, sampling_rate, audio)
print(audio_path)
except:
continue
avg_z = avg_z / _count
np.save(style, avg_z)
def test_mel2samp():
"""Test mel2samp modules on example data."""
from mel2samp import Mel2Samp
hparams = hparams_class()
passed = 0
# test filelist loader
try:
from mel2samp import load_filepaths_and_text
audio_files = load_filepaths_and_text("code_tests/test_materials/filelists/validation_utf8.txt")
passed+=1
print("--PASSED--\n")
except Exception as ex:
print("--EXCEPTION-- @ Load Filepaths and Text (UTF-8)")
traceback.print_exc(file=sys.stdout)
print("\n")
# test filelist checker
try:
assert audio_files
from mel2samp import check_files
audio_files = check_files(audio_files, hparams)
assert len(audio_files) == 1
passed+=1
print("--PASSED--\n")
del audio_files
except Exception as ex:
print("--EXCEPTION-- @ Load Filepaths and Text (UTF-8)")
traceback.print_exc(file=sys.stdout)
print("\n")
# test initalization
try:
trainset = Mel2Samp(hparams)
passed+=1
print("--PASSED--\n")
except Exception as ex:
print("--EXCEPTION-- @ Mel2Samp Initialization")
traceback.print_exc(file=sys.stdout)
print("\n")
# test 16-BIT .wav to torch
try:
from mel2samp import load_wav_to_torch
x, sr = load_wav_to_torch("code_tests/test_materials/audio_0/example_16bits.wav")
assert len(x)
assert x.max() <= 2**15
assert x.min() >= -(2**15)
assert sr == 48000
passed+=1
print("--PASSED--\n")
except Exception as ex:
print("--EXCEPTION-- @ Load 16-BIT .wav to Pytorch")
traceback.print_exc(file=sys.stdout)
print("\n")
# test 24-BIT .wav to torch
try:
from mel2samp import load_wav_to_torch
x, sr = load_wav_to_torch("code_tests/test_materials/audio_0/example_24bits.wav")
assert len(x)
assert x.max() <= 2**23
assert x.min() >= -(2**23)
assert sr == 48000
passed+=1
print("--PASSED--\n")
except Exception as ex:
print("--EXCEPTION-- @ Load 24-BIT .wav to Pytorch")
traceback.print_exc(file=sys.stdout)
print("\n")
# test 32-BIT .wav to torch
try:
from mel2samp import load_wav_to_torch
x, sr = load_wav_to_torch("code_tests/test_materials/audio_0/example_32bits.wav")
assert len(x)
assert x.max() <= 2**31
assert x.min() >= -(2**31)
assert sr == 48000
passed+=1
print("--PASSED--\n")
except Exception as ex:
print("--EXCEPTION-- @ Load 32-BIT .wav to Pytorch")
traceback.print_exc(file=sys.stdout)
print("\n")
# test 32-BIT .mp3 to torch
try:
from mel2samp import load_wav_to_torch
x, sr = load_wav_to_torch("code_tests/test_materials/audio_0/example_32bits.mp3")
assert len(x)
assert x.max() <= 2**31
assert x.min() >= -(2**31)
assert sr == 48000
passed+=1
print("--PASSED--\n")
except Exception as ex:
print("--EXCEPTION-- @ Load 32-BIT .mp3 to Pytorch")
traceback.print_exc(file=sys.stdout)
print("\n")
# test 16-BIT .wav to mel
try:
x, sr = load_wav_to_torch("code_tests/test_materials/audio_0/example_16bits.wav")
x = trainset.get_mel(x)
passed+=1
print("--PASSED--\n")
except Exception as ex:
print("--EXCEPTION-- @ 16-BIT .wav to Mel-spec")
traceback.print_exc(file=sys.stdout)
print("\n")
# test 24-BIT .wav to mel
try:
x, sr = load_wav_to_torch("code_tests/test_materials/audio_0/example_24bits.wav")
x = trainset.get_mel(x)
passed+=1
print("--PASSED--\n")
except Exception as ex:
print("--EXCEPTION-- @ 24-BIT .wav to Mel-spec")
traceback.print_exc(file=sys.stdout)
print("\n")
# test 32-BIT .wav to mel
try:
x, sr = load_wav_to_torch("code_tests/test_materials/audio_0/example_32bits.wav")
x = trainset.get_mel(x)
passed+=1
print("--PASSED--\n")
except Exception as ex:
print("--EXCEPTION-- @ 32-BIT .wav to Mel-spec")
traceback.print_exc(file=sys.stdout)
print("\n")
# test 32-BIT .mp3 to mel
try:
x, sr = load_wav_to_torch("code_tests/test_materials/audio_0/example_32bits.mp3")
x = trainset.get_mel(x)
passed+=1
print("--PASSED--\n")
except Exception as ex:
print("--EXCEPTION-- @ 32-BIT .mp3 to Mel-spec")
traceback.print_exc(file=sys.stdout)
print("\n")
# test __getitem__ with load_mel_from_disk = False
try:
assert trainset # This test will fail if Mel2Samp cannot initalize
trainset.load_mel_from_disk = False
trainset.__getitem__(0)
passed+=1
print("--PASSED--\n")
except Exception as ex:
print("--EXCEPTION-- @ __getitem__ with load_mel_from_disk = False")
traceback.print_exc(file=sys.stdout)
print("\n")
# test __getitem__ with load_mel_from_disk = True
try:
assert trainset # This test will fail if Mel2Samp cannot initalize
trainset.load_mel_from_disk = True
trainset.__getitem__(0)
passed+=1
print("--PASSED--\n")
except Exception as ex:
print("--EXCEPTION-- @ __getitem__ with load_mel_from_disk = True")
traceback.print_exc(file=sys.stdout)
print("\n")
# test initalization with Pre-empthasis
try:
trainset = None
hparams.preempthasis = 0.98
trainset = Mel2Samp(hparams)
passed+=1
print("--PASSED--\n")
except Exception as ex:
print("--EXCEPTION-- @ Mel2Samp with Pre-empthasis Initialization")
traceback.print_exc(file=sys.stdout)
print("\n")
# test __getitem__ with Pre-empthasis
try:
assert trainset # This test will fail if Mel2Samp cannot initalize
trainset.load_mel_from_disk = False
trainset.__getitem__(0)
passed+=1
print("--PASSED--\n")
except Exception as ex:
print("--EXCEPTION-- @ __getitem__ with Pre-empthasis")
traceback.print_exc(file=sys.stdout)
print("\n")
def validate(model,
loader_STFT,
STFTs,
logger,
iteration,
validation_files,
speaker_lookup,
sigma,
output_directory,
data_config,
save_audio=True,
max_length_s=5):
from mel2samp import load_wav_to_torch
from scipy import signal
val_sigma = sigma * 0.9
model.eval()
with torch.no_grad():
with open(validation_files, encoding='utf-8') as f:
audiopaths_and_melpaths = [line.strip().split('|') for line in f]
if list(model.parameters())[0].type() == "torch.cuda.HalfTensor":
model_type = "half"
else:
model_type = "float"
timestr = time.strftime("%Y_%m_%d-%H_%M_%S")
total_MAE = total_MSE = total = 0
for i, (audiopath, melpath,
*remaining) in enumerate(audiopaths_and_melpaths):
if i > 60: break # debug
audio = load_wav_to_torch(
audiopath)[0] / 32768.0 # load audio from wav file to tensor
if audio.shape[0] > (data_config['sampling_rate'] * max_length_s):
continue # ignore audio over max_length_seconds
if loader_STFT:
mel = loader_STFT.mel_spectrogram(audio.unsqueeze(0)).cuda()
else:
mel = np.load(melpath) # load mel from file into numpy arr
mel = torch.from_numpy(mel).unsqueeze(
0).cuda() # from numpy arr to tensor on GPU
if hasattr(model, 'multispeaker') and model.multispeaker == True:
assert len(
remaining
), f"Speaker ID missing while multispeaker == True.\nLine: {i}\n'{'|'.join([autiopath, melpath])}'"
speaker_id = remaining[0]
speaker_id = torch.IntTensor([speaker_lookup[int(speaker_id)]])
speaker_id = speaker_id.cuda(non_blocking=True).long()
else:
speaker_id = None
if model_type == "half":
mel = mel.half() # for fp16 training
audio_waveglow = model.infer(mel, speaker_id, sigma=val_sigma)
audio_waveglow = audio_waveglow.cpu().float()
if data_config['preempthasis']: # inverse-preempthasis
audio_waveglow = audio_waveglow.squeeze()
audio_waveglow = torch.from_numpy(
signal.lfilter([1],
[1, -float(data_config['preempthasis'])],
audio_waveglow.numpy())
).float(
) # de-preempthasis (scipy signal is faster than pytorch implementation for some reason /shrug )
audio = audio.squeeze().unsqueeze(
0) # crush extra dimensions and shape for STFT
audio_waveglow = audio_waveglow.squeeze().unsqueeze(
0) # crush extra dimensions and shape for STFT
audio_waveglow = audio_waveglow.clamp(
-1, 1
) # clamp any values over/under |1.0| (which should only exist very early in training)
for STFT in STFTs: # check Spectrogram Error with multiple window sizes
mel_GT = STFT.mel_spectrogram(audio)
try:
mel_waveglow = STFT.mel_spectrogram(
audio_waveglow)[:, :, :mel_GT.shape[-1]]
except AssertionError as ex:
print(ex)
continue
MSE = (torch.nn.MSELoss()(mel_waveglow, mel_GT)).item(
) # get MSE (Mean Squared Error) between Ground Truth and WaveGlow inferred spectrograms.
MAE = (torch.nn.L1Loss()(mel_waveglow, mel_GT)).item(
) # get MAE (Mean Absolute Error) between Ground Truth and WaveGlow inferred spectrograms.
total_MAE += MAE
total_MSE += MSE
total += 1
if save_audio:
audio_path = os.path.join(
output_directory, "samples",
str(iteration) + "-" + timestr, os.path.basename(audiopath)
) # Write audio to checkpoint_directory/iteration/audiofilename.wav
os.makedirs(os.path.join(output_directory, "samples",
str(iteration) + "-" + timestr),
exist_ok=True)
sf.write(audio_path,
audio_waveglow.squeeze().cpu().numpy(),
data_config['sampling_rate'],
"PCM_16") # save waveglow sample
audio_path = os.path.join(
output_directory, "samples", "Ground Truth",
os.path.basename(audiopath)
) # Write audio to checkpoint_directory/iteration/audiofilename.wav
if not os.path.exists(audio_path):
os.makedirs(os.path.join(output_directory, "samples",
"Ground Truth"),
exist_ok=True)
sf.write(audio_path,
audio.squeeze().cpu().numpy(),
data_config['sampling_rate'],
"PCM_16") # save ground truth
for convinv in model.convinv:
if hasattr(convinv, 'W_inverse'):
delattr(convinv, "W_inverse") # clear Inverse Weights.
if total:
average_MSE = total_MSE / total
average_MAE = total_MAE / total
logger.add_scalar('val_MSE', average_MSE, iteration)
logger.add_scalar('val_MAE', average_MAE, iteration)
print("Average MSE:", average_MSE, "Average MAE:", average_MAE)
else:
average_MSE = 1e3
average_MAE = 1e3
print("Average MSE: N/A", "Average MAE: N/A")
model.train()
return average_MSE, average_MAE
def main(style, waveglow_path, sigma, output_dir, sampling_rate, is_fp16,
denoiser_strength, args):
#mel_files = files_to_list(mel_files)
#print(mel_files)
dataset = voice_dataset(dataBase={
'ravdess': './our_data/ravdess',
'cremad': './our_data/cremad'
},
style=('happy', 'sad', 'angry'))
#print(len(dataset.final_data['happy']))
#sample = dataset.pick_one_random_sample('happy')
styles = ['happy', 'sad', 'angry']
with open('config.json') as f:
data = f.read()
config = json.loads(data)
waveglow_config = config["waveglow_config"]
model = WaveGlow(**waveglow_config)
checkpoint_dict = torch.load('waveglow_256channels_universal_v5.pt',
map_location='cpu')
model_for_loading = checkpoint_dict['model']
model.load_state_dict(model_for_loading.state_dict())
model.cuda()
waveglow = model
if is_fp16:
from apex import amp
waveglow, _ = amp.initialize(waveglow, [], opt_level="O1")
if denoiser_strength > 0:
denoiser = Denoiser(waveglow).cuda()
mel_extractor = Get_mel(1024, 256, 1024, args.sampling_rate, 0.0, 8000.0)
vector_all = {}
for style in styles:
files = dataset.final_data[style].copy()
random.shuffle(files)
vectors = []
for i, (_, file_path) in enumerate(files):
if i > 200:
break
try:
audio, rate = load_wav_to_torch(file_path)
if rate != sampling_rate:
audio = resampy.resample(audio.numpy(), rate,
sampling_rate)
audio = torch.from_numpy(audio).float()
#if audio.size(0) >= args.segment_length:
# max_audio_start = audio.size(0) - args.segment_length
# audio_start = random.randint(0, max_audio_start)
# audio = audio[audio_start:audio_start+args.segment_length]
#else:
# audio = torch.nn.functional.pad(audio, (0, args.segment_length-audio.size(0)), 'constant').data
mel = mel_extractor.get_mel(audio)
audio = audio / MAX_WAV_VALUE
mel = torch.autograd.Variable(mel.cuda().unsqueeze(0))
audio = torch.autograd.Variable(audio.cuda().unsqueeze(0))
audio = audio.half() if is_fp16 else audio
mel = mel.half() if is_fp16 else mel
outputs = waveglow((mel, audio))
vectors.append(
outputs[0].squeeze(0).mean(1).detach().cpu().numpy())
print(style, i)
except:
continue
vector_all[style] = vectors
np.save('all_style_vector', vector_all)
def validate(model,
STFTs,
logger,
iteration,
speaker_lookup,
hparams,
output_directory,
save_audio=True,
max_length_s=5):
from mel2samp import load_wav_to_torch
val_sigma = hparams.sigma * 0.9
model.eval()
with torch.no_grad():
with open(hparams.validation_files, encoding='utf-8') as f:
audiopaths_and_melpaths = [line.strip().split('|') for line in f]
if list(model.parameters())[0].type() == "torch.cuda.HalfTensor":
model_type = "half"
else:
model_type = "float"
timestr = time.strftime("%Y_%m_%d-%H_%M_%S")
total_MAE = total_MSE = total = 0
for i, (audiopath, melpath,
*remaining) in enumerate(audiopaths_and_melpaths):
if i > 30: break # debug
audio = load_wav_to_torch(
audiopath)[0] / 32768.0 # load audio from wav file to tensor
if audio.shape[0] > (hparams.sampling_rate * max_length_s):
continue # ignore audio over max_length_seconds
mel = np.load(melpath) # load mel from file into numpy arr
mel = torch.from_numpy(mel).unsqueeze(
0).cuda() # from numpy arr to tensor on GPU
#mel = (mel+5.2)*0.5 # shift values between approx -4 and 4
if hasattr(model, 'multispeaker') and model.multispeaker == True:
assert len(
remaining
), f"Speaker ID missing while multispeaker == True.\nLine: {i}\n'{'|'.join([autiopath, melpath])}'"
speaker_id = remaining[0]
assert int(speaker_id) in speaker_lookup.keys(
), f"Validation speaker ID:{speaker_id} not found in training filelist.\n(This speaker does not have an embedding, either use single-speaker models or provide an example of this speaker in the training data)."
speaker_id = torch.IntTensor([speaker_lookup[int(speaker_id)]])
speaker_id = speaker_id.cuda(non_blocking=True).long()
else:
speaker_id = None
if model_type == "half":
mel = mel.half() # for fp16 training
audio_waveglow = model.infer(mel, speaker_id, sigma=val_sigma)
audio_waveglow = audio_waveglow.cpu().float()
audio = audio.squeeze().unsqueeze(
0) # crush extra dimensions and shape for STFT
audio_waveglow = audio_waveglow.squeeze().unsqueeze(
0) # crush extra dimensions and shape for STFT
audio_waveglow = audio_waveglow.clamp(
-1, 1
) # clamp any values over/under |1.0| (which should only exist very early in training)
for STFT in STFTs: # check Spectrogram Error with multiple window sizes
mel_GT = STFT.mel_spectrogram(audio)
try:
mel_waveglow = STFT.mel_spectrogram(
audio_waveglow)[:, :, :mel_GT.shape[-1]]
except AssertionError as ex:
cprint(ex, b_tqdm=hparams.tqdm)
continue
MSE = (torch.nn.MSELoss()(mel_waveglow, mel_GT)).item(
) # get MSE (Mean Squared Error) between Ground Truth and WaveGlow inferred spectrograms.
MAE = (torch.nn.L1Loss()(mel_waveglow, mel_GT)).item(
) # get MAE (Mean Absolute Error) between Ground Truth and WaveGlow inferred spectrograms.
total_MAE += MAE
total_MSE += MSE
total += 1
if save_audio:
audio_path = os.path.join(
output_directory, "samples",
str(iteration) + "-" + timestr, os.path.basename(audiopath)
) # Write audio to checkpoint_directory/iteration/audiofilename.wav
os.makedirs(os.path.join(output_directory, "samples",
str(iteration) + "-" + timestr),
exist_ok=True)
sf.write(audio_path,
audio_waveglow.squeeze().cpu().numpy(),
hparams.sampling_rate,
"PCM_16") # save waveglow sample
audio_path = os.path.join(
output_directory, "samples", "Ground Truth",
os.path.basename(audiopath)
) # Write audio to checkpoint_directory/iteration/audiofilename.wav
if not os.path.exists(audio_path):
os.makedirs(os.path.join(output_directory, "samples",
"Ground Truth"),
exist_ok=True)
sf.write(audio_path,
audio.squeeze().cpu().numpy(),
hparams.sampling_rate,
"PCM_16") # save ground truth
for convinv in model.convinv:
if hasattr(convinv, 'W_inverse'):
delattr(convinv, "W_inverse") # clear Inverse Weights.
if total:
average_MSE = total_MSE / total
average_MAE = total_MAE / total
logger.add_scalar('val_MSE', average_MSE, iteration)
logger.add_scalar('val_MAE', average_MAE, iteration)
cprint("Average MSE:",
average_MSE,
"Average MAE:",
average_MAE,
b_tqdm=hparams.tqdm)
else:
average_MSE = 1e3
average_MAE = 1e3
cprint("Average MSE: N/A", "Average MAE: N/A", b_tqdm=hparams.tqdm)
model.train()
return average_MSE, average_MAE
