Skip to content

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

decoding

游雁b139b9cfaef2
diff --git a/funasr/models/sense_voice/decoder.py b/funasr/models/sense_voice/decoder.py
index 19d9c1680..dd00ca869 100644
--- a/funasr/models/sense_voice/decoder.py
+++ b/funasr/models/sense_voice/decoder.py
@@ -472,7 +472,7 @@ class ResidualAttentionBlockFSMN(nn.Module):
is_pad_mask = kwargs.get("is_pad_mask", False)
is_pad_memory_mask = kwargs.get("is_pad_memory_mask", False)
- fsmn_cache = cache[layer]["fsmn_cache"] if len(cache) > 0 else None
+ fsmn_cache = cache[layer]["fsmn_cache"] if len(cache) > 0 or cache is None else None
# if fsmn_cache is not None:
# x = x[:, -1:]
att_res, fsmn_cache = self.attn(self.attn_ln(x), mask=None, cache=fsmn_cache)