NikkeDoy/SISU
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

:albemic: | Real-time testing...

Browse Source
This commit is contained in:
NikkeDoy
2025-05-04 22:48:57 +03:00
parent d70c86c257
commit 660b41aef8
4 changed files with 107 additions and 75 deletions
Download Patch File Download Diff File Expand all files Collapse all files

32
data.py
View File

@ -21,33 +21,25 @@ class AudioDataset(Dataset):
def __getitem__(self, idx):
# Load high-quality audio
high_quality_audio, original_sample_rate = torchaudio.load(self.input_files[idx], normalize=True)
# Change to mono
high_quality_audio = AudioUtils.stereo_tensor_to_mono(high_quality_audio)
# Generate low-quality audio with random downsampling
mangled_sample_rate = random.choice(self.audio_sample_rates)
resample_transform_low = torchaudio.transforms.Resample(original_sample_rate, mangled_sample_rate)
low_quality_audio = resample_transform_low(high_quality_audio)
resample_transform_low = torchaudio.transforms.Resample(original_sample_rate, mangled_sample_rate)
resample_transform_high = torchaudio.transforms.Resample(mangled_sample_rate, original_sample_rate)
low_quality_audio = resample_transform_low(high_quality_audio)
low_quality_audio = resample_transform_high(low_quality_audio)
high_quality_audio = AudioUtils.stereo_tensor_to_mono(high_quality_audio)
low_quality_audio = AudioUtils.stereo_tensor_to_mono(low_quality_audio)
# Pad or truncate high-quality audio
if high_quality_audio.shape[1] < self.MAX_LENGTH:
padding = self.MAX_LENGTH - high_quality_audio.shape[1]
high_quality_audio = F.pad(high_quality_audio, (0, padding))
elif high_quality_audio.shape[1] > self.MAX_LENGTH:
high_quality_audio = high_quality_audio[:, :self.MAX_LENGTH]
splitted_high_quality_audio = AudioUtils.split_audio(high_quality_audio, 128)
splitted_high_quality_audio[-1] = AudioUtils.pad_tensor(splitted_high_quality_audio[-1], 128)
splitted_high_quality_audio = [tensor.to(self.device) for tensor in splitted_high_quality_audio]
# Pad or truncate low-quality audio
if low_quality_audio.shape[1] < self.MAX_LENGTH:
padding = self.MAX_LENGTH - low_quality_audio.shape[1]
low_quality_audio = F.pad(low_quality_audio, (0, padding))
elif low_quality_audio.shape[1] > self.MAX_LENGTH:
low_quality_audio = low_quality_audio[:, :self.MAX_LENGTH]
splitted_low_quality_audio = AudioUtils.split_audio(low_quality_audio, 128)
splitted_low_quality_audio[-1] = AudioUtils.pad_tensor(splitted_low_quality_audio[-1], 128)
splitted_low_quality_audio = [tensor.to(self.device) for tensor in splitted_low_quality_audio]
high_quality_audio = high_quality_audio.to(self.device)
low_quality_audio = low_quality_audio.to(self.device)
return (high_quality_audio, original_sample_rate), (low_quality_audio, mangled_sample_rate)
return (splitted_high_quality_audio, original_sample_rate), (splitted_low_quality_audio, mangled_sample_rate)