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

⚗️ | Experiment with other layer layouts.

Browse Source
This commit is contained in:
NikkeDoy
2024-12-21 23:27:38 +02:00
parent b7d7e95c89
commit 70e20f53d4
6 changed files with 142 additions and 91 deletions
Download Patch File Download Diff File Expand all files Collapse all files

25
data.py
View File

@ -19,26 +19,25 @@ class AudioDataset(Dataset):
def __getitem__(self, idx):
high_quality_wav, sr_original = torchaudio.load(self.input_files[idx], normalize=True)
high_quality_audio, original_sample_rate = torchaudio.load(self.input_files[idx], normalize=True)
sample_rate = random.choice(self.audio_sample_rates)
resample_transform = torchaudio.transforms.Resample(sr_original, sample_rate)
low_quality_wav = resample_transform(high_quality_wav)
low_quality_wav = low_quality_wav
mangled_sample_rate = random.choice(self.audio_sample_rates)
resample_transform = torchaudio.transforms.Resample(original_sample_rate, mangled_sample_rate)
low_quality_audio = resample_transform(high_quality_audio)
# Calculate target length based on desired duration and 16000 Hz
if self.target_duration is not None:
target_length = int(self.target_duration * 44100)
else:
# Calculate duration of original high quality audio
target_length = high_quality_wav.size(1)
# if self.target_duration is not None:
# target_length = int(self.target_duration * 44100)
# else:
# # Calculate duration of original high quality audio
# target_length = high_quality_wav.size(1)
# Pad both to the calculated target length
high_quality_wav = self.stretch_tensor(high_quality_wav, target_length)
low_quality_wav = self.stretch_tensor(low_quality_wav, target_length)
# high_quality_wav = self.stretch_tensor(high_quality_wav, target_length)
# low_quality_wav = self.stretch_tensor(low_quality_wav, target_length)
return low_quality_wav, high_quality_wav
return (high_quality_audio, original_sample_rate), (low_quality_audio, mangled_sample_rate)
def stretch_tensor(self, tensor, target_length):
current_length = tensor.size(1)