mirrors/icefall
1
0
Fork 0
mirror of https://github.com/k2-fsa/icefall.git synced 2025-08-09 01:52:41 +00:00
Code Issues Packages Projects Releases Wiki Activity

Support using different musan augmentations for the same audio.

Browse Source 
In addition, it returns the original audio without augmentation.
This commit is contained in:
Fangjun Kuang 2025-07-01 16:58:31 +08:00
parent 075e74bcb5
commit 85f6deb8d1
1 changed files with 9 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
egs/librispeech/ASR/zipformer/speech_recognition.py
View File

@ -103,13 +103,15 @@ class K2SpeechRecognitionDataset(torch.utils.data.Dataset):
# Sort the cuts by duration so that the first one determines the batch time dimensions. # Sort the cuts by duration so that the first one determines the batch time dimensions.
cuts = cuts.sort_by_duration(ascending=False) cuts = cuts.sort_by_duration(ascending=False)
# Optional CutSet transforms - e.g. padding, or speed perturbation that adjusts if self.cut_transforms:
# the supervision boundaries. orig_cuts = cuts
cuts = cuts.repeat(times=2)
for tnfm in self.cut_transforms: for tnfm in self.cut_transforms:
cuts = tnfm(cuts) cuts = tnfm(cuts)
# Sort the cuts again after transforms cuts = orig_cuts + cuts
cuts = cuts.sort_by_duration(ascending=False)
# Get a tensor with batched feature matrices, shape (B, T, F) # Get a tensor with batched feature matrices, shape (B, T, F)
# Collation performs auto-padding, if necessary. # Collation performs auto-padding, if necessary.
@ -117,7 +119,7 @@ class K2SpeechRecognitionDataset(torch.utils.data.Dataset):
if len(input_tpl) == 3: if len(input_tpl) == 3:
# An input strategy with fault tolerant audio reading mode. # An input strategy with fault tolerant audio reading mode.
# "cuts" may be a subset of the original "cuts" variable, # "cuts" may be a subset of the original "cuts" variable,
# that only has cuts for which we succesfully read the audio. # that only has cuts for which we successfully read the audio.
inputs, _, cuts = input_tpl inputs, _, cuts = input_tpl
else: else:
inputs, _ = input_tpl inputs, _ = input_tpl