Skip to content

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

修复流式识别中prev_samples截取bug (#2779)

ypatz4120c252415f
diff --git a/funasr/models/fsmn_vad_streaming/model.py b/funasr/models/fsmn_vad_streaming/model.py
index f62e36a2c..7dfb96a05 100644
--- a/funasr/models/fsmn_vad_streaming/model.py
+++ b/funasr/models/fsmn_vad_streaming/model.py
@@ -724,7 +724,7 @@ class FsmnVADStreaming(nn.Module):
if len(segments_i) > 0:
segments.extend(*segments_i)
- cache["prev_samples"] = audio_sample[:-m]
+ cache["prev_samples"] = audio_sample[-m:] if m > 0 else torch.empty(0)
if _is_final:
self.init_cache(cache)
diff --git a/funasr/models/paraformer_streaming/model.py b/funasr/models/paraformer_streaming/model.py
index 16021ceb6..cbf666992 100644
--- a/funasr/models/paraformer_streaming/model.py
+++ b/funasr/models/paraformer_streaming/model.py
@@ -642,7 +642,7 @@ class ParaformerStreaming(Paraformer):
result_i = {"key": key[0], "text": text_postprocessed}
result = [result_i]
- cache["prev_samples"] = audio_sample[:-m]
+ cache["prev_samples"] = audio_sample[-m:] if m > 0 else torch.empty(0)
if _is_final:
self.init_cache(cache, **kwargs)
diff --git a/funasr/models/scama/model.py b/funasr/models/scama/model.py
index c15f435a4..22a48bccc 100644
--- a/funasr/models/scama/model.py
+++ b/funasr/models/scama/model.py
@@ -637,7 +637,7 @@ class SCAMA(nn.Module):
cache["decoder"] = cache_decoder
cache["frontend"] = {}
- cache["prev_samples"] = torch.empty(0).to(device=device)
+ cache["prev_samples"] = torch.empty(0)
return cache
@@ -726,7 +726,7 @@ class SCAMA(nn.Module):
result_i = {"key": key[0], "text": text_postprocessed}
result = [result_i]
- cache["prev_samples"] = audio_sample[:-m]
+ cache["prev_samples"] = audio_sample[-m:] if m > 0 else torch.empty(0)
if _is_final:
self.init_cache(cache, **kwargs)