Skip to content

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

fix: handle empty audio/text input gracefully in VAD and PUNC models

游雁4ad2a7b04cc5
diff --git a/funasr/models/ct_transformer/model.py b/funasr/models/ct_transformer/model.py
index abc5dfd10..f1a231893 100644
--- a/funasr/models/ct_transformer/model.py
+++ b/funasr/models/ct_transformer/model.py
@@ -254,6 +254,9 @@ class CTTransformer(torch.nn.Module):
**kwargs,
):
assert len(data_in) == 1
+ if not data_in[0] or (isinstance(data_in[0], str) and not data_in[0].strip()):
+ meta_data = {"batch_data_time": -1}
+ return [{"key": key[0] if key else "", "text": "", "punc_array": None}], meta_data
text = load_audio_text_image_video(data_in, data_type=kwargs.get("kwargs", "text"))[0]
vad_indexes = kwargs.get("vad_indexes", None)
# text = data_in[0]
diff --git a/funasr/models/fsmn_vad_streaming/model.py b/funasr/models/fsmn_vad_streaming/model.py
index d8dd5ae1e..1657631ab 100644
--- a/funasr/models/fsmn_vad_streaming/model.py
+++ b/funasr/models/fsmn_vad_streaming/model.py
@@ -688,6 +688,8 @@ class FsmnVADStreaming(nn.Module):
is_streaming_input = cfg["is_streaming_input"]
time2 = time.perf_counter()
meta_data["load_data"] = f"{time2 - time1:0.3f}"
+ if len(audio_sample_list) == 0 or (hasattr(audio_sample_list[0], '__len__') and len(audio_sample_list[0]) == 0):
+ return [{"key": key[0] if key else "", "value": []}], meta_data
assert len(audio_sample_list) == 1, "batch_size must be set 1"
audio_sample = torch.cat((cache["prev_samples"], audio_sample_list[0]))