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
[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.pyindex 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': []})continuespeech_j, speech_lengths_j = slice_padding_audio_samples(@@ -418,7 +417,7 @@ class AutoModel:end_idx += 1max_len_in_batch = sample_lengthif len(results) < 1:- results.append({'key': 'bad_data', 'text': '', 'timestamp': []})+ continueresults_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))+ continuerestored_data = [0] * nfor j in range(n):index = sorted_data[j][1]