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
Funasr1.0 (#1284)
zhifu gao2c3183b61148
diff --git a/funasr/models/ct_transformer/model.py b/funasr/models/ct_transformer/model.pyindex 8c3f04327..330d7e554 100644--- a/funasr/models/ct_transformer/model.py+++ b/funasr/models/ct_transformer/model.py@@ -333,12 +333,13 @@ class CTTransformer(torch.nn.Module):elif new_mini_sentence[-1] == ",":new_mini_sentence_out = new_mini_sentence[:-1] + "."new_mini_sentence_punc_out = new_mini_sentence_punc[:-1] + [self.sentence_end_id]- elif new_mini_sentence[-1] != "。" and new_mini_sentence[-1] != "?" and len(new_mini_sentence[-1].encode())==0:+ elif new_mini_sentence[-1] != "。" and new_mini_sentence[-1] != "?" and len(new_mini_sentence[-1].encode())!=1:new_mini_sentence_out = new_mini_sentence + "。"new_mini_sentence_punc_out = new_mini_sentence_punc[:-1] + [self.sentence_end_id]elif new_mini_sentence[-1] != "." and new_mini_sentence[-1] != "?" and len(new_mini_sentence[-1].encode())==1:new_mini_sentence_out = new_mini_sentence + "."new_mini_sentence_punc_out = new_mini_sentence_punc[:-1] + [self.sentence_end_id]+# keep a punctuations array for punc segmentif punc_array is None:punc_array = punctuationsdiff --git a/funasr/utils/load_utils.py b/funasr/utils/load_utils.pyindex 9cd38547a..6f60dac43 100644--- a/funasr/utils/load_utils.py+++ b/funasr/utils/load_utils.py@@ -39,7 +39,8 @@ def load_audio_text_image_video(data_or_path_or_list, fs: int = 16000, audio_fs:if isinstance(data_or_path_or_list, str) and os.path.exists(data_or_path_or_list): # local fileif data_type is None or data_type == "sound":data_or_path_or_list, audio_fs = torchaudio.load(data_or_path_or_list)- data_or_path_or_list = data_or_path_or_list[0, :]+ if kwargs.get("reduce_channels", True):+ data_or_path_or_list = data_or_path_or_list.mean(0)elif data_type == "text" and tokenizer is not None:data_or_path_or_list = tokenizer.encode(data_or_path_or_list)elif data_type == "image": # undo