msb-public
/
PaddleSpeech
mirror of https://github.com/PaddlePaddle/PaddleSpeech
1
0
Code Issues Projects Releases Wiki Activity

Merge pull request #691 from PaddlePaddle/bugfix

Browse Source 
fix audio shape bug for audio len
pull/692/head
Hui Zhang 4 years ago committed by GitHub
parent dd433982e0 9c0b6c5bb0
commit d1c280cce8
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 3 additions and 3 deletions
Show Stats Download Patch File Download Diff File

6
deepspeech/io/collator.py
Unescape View File

@ -154,8 +154,8 @@ class SpeechCollator():
random_seed (int, optional): for random generator. Defaults to 0. random_seed (int, optional): for random generator. Defaults to 0.
keep_transcription_text (bool, optional): True, when not in training mode, will not do tokenizer; Defaults to False. keep_transcription_text (bool, optional): True, when not in training mode, will not do tokenizer; Defaults to False.
if ``keep_transcription_text`` is False, text is token ids else is raw string. if ``keep_transcription_text`` is False, text is token ids else is raw string.
Do augmentations Do augmentations
Padding audio features with zeros to make them have the same shape (or Padding audio features with zeros to make them have the same shape (or
a user-defined shape) within one batch. a user-defined shape) within one batch.
""" """
@ -271,7 +271,7 @@ class SpeechCollator():
utts.append(utt) utts.append(utt)
# audio # audio
audios.append(audio) # [T, D] audios.append(audio) # [T, D]
audio_lens.append(audio.shape[1]) audio_lens.append(audio.shape[0])
# text # text
# for training, text is token ids # for training, text is token ids
# else text is string, convert to unicode ord # else text is string, convert to unicode ord

Write Preview
Loading…
Cancel
Save