Open-source speech recognition toolkit for training, inference, streaming ASR, VAD, punctuation, speaker diarization pipelines, and OpenAI-compatible/MCP serving.
Python asraudiochineseemotion-recognitionfunasrmcp-servermultilingual-asropenai-compatible-apiparaformerpunctuationpytorchreal-time-asrspeaker-diarizationspeech-recognitionspeech-to-textstreaming-asrtranscriptionvllmvoice-activity-detectionwhisper-alternative
decoding key
游雁cfe577f16fef
diff --git a/funasr/models/sense_voice/decoder.py b/funasr/models/sense_voice/decoder.pyindex 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