[Frontend] Remove librosa from audio dependency (#37058)

Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn>
This commit is contained in:
Isotr0py
2026-03-21 11:36:15 +08:00
committed by GitHub
parent 1c472f8fe1
commit c7f98b4d0a
18 changed files with 247 additions and 188 deletions
@@ -188,7 +188,7 @@ class FireRedASR2FeatureExtractor(SequenceFeatureExtractor):
for speech in raw_speech:
"""
We must multiply by 32768 here because FireRedASR2 loads audio data
using kaldiio.load_mat, while vLLM loads audio data using librosa.
using kaldiio.load_mat, while vLLM loads audio data using pyav.
"""
speech = speech * 32768
fbank = self.fbank(sampling_rate, speech)