Skip to content

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

decoding key

游雁cfe577f16fef
diff --git a/funasr/models/sense_voice/decoder.py b/funasr/models/sense_voice/decoder.py
index 3c62072e5..03b753246 100644
--- a/funasr/models/sense_voice/decoder.py
+++ b/funasr/models/sense_voice/decoder.py
@@ -599,5 +599,6 @@ class SenseVoiceDecoderFSMN(nn.Module):
def score(self, ys, state, x):
"""Score."""
ys_mask = subsequent_mask(len(ys), device=x.device).unsqueeze(0)
- logp = self.forward(ys.unsqueeze(0), x.unsqueeze(0), cache=state)
+ logp = self.forward(ys.unsqueeze(0), x.unsqueeze(0), cache=None)
+ logp = torch.log_softmax(logp, dim=-1)
return logp.squeeze(0)[-1, :], state