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
优化speakid和语句匹配逻辑,部分解决speakid不从0递增问题 (#1870)
wuhongsheng3a4281f49595
diff --git a/funasr/models/campplus/utils.py b/funasr/models/campplus/utils.pyindex 94b4952de..4addd0a8a 100644--- a/funasr/models/campplus/utils.py+++ b/funasr/models/campplus/utils.py@@ -193,23 +193,21 @@ def smooth(res, mindur=0.7):def distribute_spk(sentence_list, sd_time_list):- sd_sentence_list = []+ sd_time_list = [(spk_st * 1000, spk_ed * 1000, spk) for spk_st, spk_ed, spk in sd_time_list]for d in sentence_list:- sentence_start = d["start"]- sentence_end = d["end"]+ sentence_start = d['start']+ sentence_end = d['end']sentence_spk = 0max_overlap = 0- for sd_time in sd_time_list:- spk_st, spk_ed, spk = sd_time- spk_st = spk_st * 1000- spk_ed = spk_ed * 1000+ for spk_st, spk_ed, spk in sd_time_list:overlap = max(min(sentence_end, spk_ed) - max(sentence_start, spk_st), 0)if overlap > max_overlap:max_overlap = overlapsentence_spk = spk- d["spk"] = int(sentence_spk)- sd_sentence_list.append(d)- return sd_sentence_list+ if overlap > 0 and sentence_spk == spk:+ max_overlap += overlap+ d['spk'] = int(sentence_spk)+ return sentence_listclass Storage(metaclass=ABCMeta):