Skip to content

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

[fix] better solution for handling empty result (#1796)

Xingchen Song(宋星辰)0bd1a4d6a989
diff --git a/funasr/auto/auto_model.py b/funasr/auto/auto_model.py
index 9b4f6d92c..191e172a7 100644
--- a/funasr/auto/auto_model.py
+++ b/funasr/auto/auto_model.py
@@ -388,7 +388,6 @@ class AutoModel:
):
max_len_in_batch = max(max_len_in_batch, sample_length)
end_idx += 1
- results_sorted.append({'key': 'bad_data', 'text': '', 'timestamp': []})
continue
speech_j, speech_lengths_j = slice_padding_audio_samples(
@@ -418,7 +417,7 @@ class AutoModel:
end_idx += 1
max_len_in_batch = sample_length
if len(results) < 1:
- results.append({'key': 'bad_data', 'text': '', 'timestamp': []})
+ continue
results_sorted.extend(results)
# end_asr_total = time.time()
@@ -428,6 +427,10 @@ class AutoModel:
# f"time_speech_total_per_sample: {time_speech_total_per_sample: 0.3f}, "
# f"time_escape_total_per_sample: {time_escape_total_per_sample:0.3f}")
+ if len(results_sorted) != n:
+ results_ret_list.append({"key": key, "text": "", "timestamp": []})
+ logging.info("decoding, utt: {}, empty result".format(key))
+ continue
restored_data = [0] * n
for j in range(n):
index = sorted_data[j][1]