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
test
语帆343a281ca148
diff --git a/funasr/models/lcbnet/model.py b/funasr/models/lcbnet/model.pyindex 54fba1cb2..d1ebc5ce5 100644--- a/funasr/models/lcbnet/model.py+++ b/funasr/models/lcbnet/model.py@@ -413,7 +413,6 @@ class LCBNet(nn.Module):logging.info("enable beam_search")self.init_beam_search(**kwargs)self.nbest = kwargs.get("nbest", 1)- pdb.set_trace()meta_data = {}if isinstance(data_in, torch.Tensor) and kwargs.get("data_type", "sound") == "fbank": # fbank@@ -431,6 +430,7 @@ class LCBNet(nn.Module):tokenizer=tokenizer)time2 = time.perf_counter()meta_data["load_data"] = f"{time2 - time1:0.3f}"+ pdb.set_trace()speech, speech_lengths = extract_fbank(audio_sample_list, data_type=kwargs.get("data_type", "sound"),frontend=frontend)time3 = time.perf_counter()diff --git a/funasr/utils/load_utils.py b/funasr/utils/load_utils.pyindex 20fa0fd2e..963f5c258 100644--- a/funasr/utils/load_utils.py+++ b/funasr/utils/load_utils.py@@ -31,14 +31,13 @@ def load_audio_text_image_video(data_or_path_or_list, fs: int = 16000, audio_fs:return [load_audio_text_image_video(audio, fs=fs, audio_fs=audio_fs, data_type=data_type, **kwargs) for audio in data_or_path_or_list]if isinstance(data_or_path_or_list, str) and data_or_path_or_list.startswith('http'): # download url to local filedata_or_path_or_list = download_from_url(data_or_path_or_list)- pdb.set_trace()+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)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:- pdb.set_trace()data_or_path_or_list = tokenizer.encode(data_or_path_or_list)elif data_type == "image": # undopass@@ -68,7 +67,7 @@ def load_audio_text_image_video(data_or_path_or_list, fs: int = 16000, audio_fs:else:pass# print(f"unsupport data type: {data_or_path_or_list}, return raw data")- pdb.set_trace()+if audio_fs != fs and data_type != "text":resampler = torchaudio.transforms.Resample(audio_fs, fs)data_or_path_or_list = resampler(data_or_path_or_list[None, :])[0, :]@@ -112,6 +111,7 @@ def extract_fbank(data, data_len = None, data_type: str="sound", frontend=None,# import pdb;# pdb.set_trace()# if data_type == "sound":+ pdb.set_trace()data, data_len = frontend(data, data_len, **kwargs)if isinstance(data_len, (list, tuple)):