dnrops
/
transformers
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Clap processor: remove wasteful np.stack operations (#27454) 

remove wasteful np.stack

Np.stack on large 1-D tensor, causing ~0.5s processing time on short audio (<10s). Compared to 0.02s for medium length audio
This commit is contained in:
Max Bain 2023-11-14 10:41:12 +00:00 committed by GitHub
parent 4309abedbc
commit b86c54d9ff
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 2 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
src/transformers/models/clap/feature_extraction_clap.py
View File

@ -242,10 +242,10 @@ class ClapFeatureExtractor(SequenceFeatureExtractor):
if waveform.shape[0] < max_length:
if padding == "repeat":
n_repeat = int(max_length / len(waveform))
waveform = np.stack(np.tile(waveform, n_repeat + 1))[:max_length]
waveform = np.tile(waveform, n_repeat + 1)[:max_length]
if padding == "repeatpad":
n_repeat = int(max_length / len(waveform))
waveform = np.stack(np.tile(waveform, n_repeat))
waveform = np.tile(waveform, n_repeat)
waveform = np.pad(waveform, (0, max_length - waveform.shape[0]), mode="constant", constant_values=0)
if truncation == "fusion":