Skip to content

Open-source speech recognition toolkit for training, inference, streaming ASR, VAD, punctuation, speaker diarization pipelines, and OpenAI-compatible/MCP serving.

Dev gzf deepspeed (#1733)

zhifu gaod3da0760a5f9
diff --git a/funasr/frontends/whisper_frontend.py b/funasr/frontends/whisper_frontend.py
index acf32dbc7..1bd8aecdd 100644
--- a/funasr/frontends/whisper_frontend.py
+++ b/funasr/frontends/whisper_frontend.py
@@ -33,7 +33,7 @@ class WhisperFrontend(nn.Module):
self.win_length = N_FFT
self.hop_length = HOP_LENGTH
self.pad_samples = N_SAMPLES
- self.frame_shift = self.hop_length
+ self.frame_shift = int(self.hop_length / self.fs * 1000)
self.lfr_n = 1
self.n_mels = n_mels
if whisper_model == "large-v3" or whisper_model == "large":