mrq/DL-Art-School
1
3
Fork 5
Code Issues Pull Requests Packages Projects Releases Wiki Activity

update audio_with_noise

Browse Source
This commit is contained in:
James Betker 2022-03-12 20:41:47 -07:00
parent e754c4fbbc
commit 9bbbe26012
1 changed files with 22 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

32
codes/data/audio/audio_with_noise_dataset.py
View File

@ -82,6 +82,8 @@ class AudioWithNoiseDataset(Dataset):
self.sampling_rate = self.underlying_dataset.sampling_rate
self.use_gpu_for_reverb_compute = opt_get(opt, ['use_gpu_for_reverb_compute'], True)
self.openair_kernels = None
self.current_item_fetch = 0
self.fetch_error_count = 0
def load_openair_kernels(self):
if self.use_gpu_for_reverb_compute and self.openair_kernels is None:
@ -91,6 +93,10 @@ class AudioWithNoiseDataset(Dataset):
self.openair_kernels.append(load_rir(oa, self.underlying_dataset.sampling_rate, self.underlying_dataset.sampling_rate*2).cuda())
def __getitem__(self, item):
if self.current_item_fetch != item:
self.current_item_fetch = item
self.fetch_error_count = 0
# Load on the fly to prevent GPU memory sharing across process errors.
self.load_openair_kernels()
@ -107,7 +113,7 @@ class AudioWithNoiseDataset(Dataset):
clip = clip * clipvol
label = random.randint(0, 4) # Current excludes GSM corruption.
label = 3
#label = 3
if label > 0 and label < 4: # 0 is basically "leave it alone"
aug_needed = True
augvol = (random.random() * (self.max_volume-self.min_volume) + self.min_volume)
@ -137,7 +143,7 @@ class AudioWithNoiseDataset(Dataset):
clip = torch.cat([clip, aug], dim=1)
# Restore some meta-parameters.
padding_room = dlen - clip.shape[-1]
out['clip_lengths'] = clip.shape[-1]
out['clip_lengths'] = torch.tensor(clip.shape[-1])
aug_needed = False
if aug_needed:
aug = load_audio(augpath, self.underlying_dataset.sampling_rate)
@ -169,9 +175,11 @@ class AudioWithNoiseDataset(Dataset):
# Apply the GSM codec to simulate cellular phone audio.
clip = torchaudio.functional.apply_codec(clip, self.underlying_dataset.sampling_rate, format="gsm")
except:
#print(f"Exception encountered processing {item}, re-trying because this is often just a failed aug.")
#print(sys.exc_info())
if self.fetch_error_count > 10:
print(f"Exception encountered processing {item}, re-trying because this is often just a failed aug.")
print(sys.exc_info())
#raise # Uncomment to surface exceptions.
self.fetch_error_count += 1
return self[item]
clip.clip_(-1, 1)
@ -193,27 +201,29 @@ if __name__ == '__main__':
params = {
'mode': 'unsupervised_audio_with_noise',
'path': ['y:/clips/books1'],
'cache_path': 'E:\\audio\\remote-cache4.pth',
'cache_path': 'D:\\data\\clips_for_noise_classifier.pth',
'sampling_rate': 22050,
'pad_to_samples': 400000,
'do_augmentation': False,
'phase': 'train',
'n_workers': 0,
'batch_size': 4,
'extra_samples': 4,
'n_workers': 4,
'batch_size': 256,
'extra_samples': 0,
'env_noise_paths': ['E:\\audio\\UrbanSound\\filtered', 'E:\\audio\\UrbanSound\\MSSND'],
'env_noise_cache': 'E:\\audio\\UrbanSound\\cache.pth',
'music_paths': ['E:\\audio\\music\\FMA\\fma_large', 'E:\\audio\\music\\maestro\\maestro-v3.0.0'],
'music_cache': 'E:\\audio\\music\\cache.pth',
'openair_path': 'D:\\data\\audio\\openair\\resampled'
'openair_path': 'D:\\data\\audio\\openair\\resampled',
'use_gpu_for_reverb_compute': False,
}
from data import create_dataset, create_dataloader, util
ds = create_dataset(params)
dl = create_dataloader(ds, params)
dl = create_dataloader(ds, params, pin_memory=False)
i = 0
for b in tqdm(dl):
for b_ in range(b['clip'].shape[0]):
torchaudio.save(f'{i}_clip_{b_}_{b["label"][b_].item()}.wav', b['clip'][b_][:, :b['clip_lengths'][b_]], ds.sampling_rate)
#torchaudio.save(f'{i}_clip_{b_}_{b["label"][b_].item()}.wav', b['clip'][b_][:, :b['clip_lengths'][b_]], ds.sampling_rate)
#torchaudio.save(f'{i}_clip_{b_}_aug.wav', b['aug'][b_], ds.sampling_rate)
print(f'{i} aug path: {b["augpath"][b_]} aug volume: {b["augvol"][b_]} clip volume: {b["clipvol"][b_]}')
i += 1