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
libtorch demo
维石509d09f50d4e
diff --git a/runtime/python/libtorch/demo.py b/runtime/python/libtorch/demo.pyindex 1ef9a2024..c8eae783f 100644--- a/runtime/python/libtorch/demo.py+++ b/runtime/python/libtorch/demo.py@@ -1,8 +1,9 @@from funasr_torch import Paraformer+from pathlib import Pathmodel_dir = (- "/nfs/zhifu.gzf/export/damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch"+ "iic/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch")model = Paraformer(model_dir, batch_size=1) # cpu@@ -11,7 +12,7 @@ model = Paraformer(model_dir, batch_size=1) # cpu# when using paraformer-large-vad-punc model, you can set plot_timestamp_to="./xx.png" to get figure of alignment besides timestamps# model = Paraformer(model_dir, batch_size=1, plot_timestamp_to="test.png")-wav_path = "YourPath/xx.wav"+wav_path = "{}/.cache/modelscope/hub/{}/example/asr_example.wav".format(Path.home(), model_dir)result = model(wav_path)print(result)diff --git a/runtime/python/libtorch/funasr_torch/paraformer_bin.py b/runtime/python/libtorch/funasr_torch/paraformer_bin.pyindex 68886dfe5..b7fb14b31 100644--- a/runtime/python/libtorch/funasr_torch/paraformer_bin.py+++ b/runtime/python/libtorch/funasr_torch/paraformer_bin.py@@ -15,9 +15,16 @@ from .utils.timestamp_utils import time_stamp_lfr6_onnxlogging = get_logger()import torch+import jsonclass Paraformer:+ """+ Author: Speech Lab of DAMO Academy, Alibaba Group+ Paraformer: Fast and Accurate Parallel Transformer for Non-autoregressive End-to-End Speech Recognition+ https://arxiv.org/abs/2206.08317+ """+def __init__(self,model_dir: Union[str, Path] = None,@@ -25,20 +32,43 @@ class Paraformer:device_id: Union[str, int] = "-1",plot_timestamp_to: str = "",quantize: bool = False,- intra_op_num_threads: int = 1,+ intra_op_num_threads: int = 4,+ cache_dir: str = None,+ **kwargs,):-if not Path(model_dir).exists():- raise FileNotFoundError(f"{model_dir} does not exist.")+ try:+ from modelscope.hub.snapshot_download import snapshot_download+ except:+ raise "You are exporting model from modelscope, please install modelscope and try it again. To install modelscope, you could:\n" "\npip3 install -U modelscope\n" "For the users in China, you could install with the command:\n" "\npip3 install -U modelscope -i https://mirror.sjtu.edu.cn/pypi/web/simple"+ try:+ model_dir = snapshot_download(model_dir, cache_dir=cache_dir)+ except:+ raise "model_dir must be model_name in modelscope or local path downloaded from modelscope, but is {}".format(+ model_dir+ )model_file = os.path.join(model_dir, "model.torchscripts")if quantize:model_file = os.path.join(model_dir, "model_quant.torchscripts")+ if not os.path.exists(model_file):+ print(".onnx is not exist, begin to export onnx")+ try:+ from funasr import AutoModel+ except:+ raise "You are exporting onnx, please install funasr and try it again. To install funasr, you could:\n" "\npip3 install -U funasr\n" "For the users in China, you could install with the command:\n" "\npip3 install -U funasr -i https://mirror.sjtu.edu.cn/pypi/web/simple"++ model = AutoModel(model=model_dir)+ model_dir = model.export(type="torchscript", quantize=quantize, **kwargs)+config_file = os.path.join(model_dir, "config.yaml")cmvn_file = os.path.join(model_dir, "am.mvn")config = read_yaml(config_file)+ token_list = os.path.join(model_dir, "tokens.json")+ with open(token_list, "r", encoding="utf-8") as f:+ token_list = json.load(f)- self.converter = TokenIDConverter(config["token_list"])+ self.converter = TokenIDConverter(token_list)self.tokenizer = CharTokenizer()self.frontend = WavFrontend(cmvn_file=cmvn_file, **config["frontend_conf"])self.ort_infer = torch.jit.load(model_file)@@ -49,6 +79,10 @@ class Paraformer:self.pred_bias = config["model_conf"]["predictor_bias"]else:self.pred_bias = 0+ if "lang" in config:+ self.language = config["lang"]+ else:+ self.language = Nonedef __call__(self, wav_content: Union[str, np.ndarray, List[str]], **kwargs) -> List:waveform_list = self.load_data(wav_content, self.frontend.opts.frame_opts.samp_freq)@@ -202,4 +236,4 @@ class Paraformer:token = self.converter.ids2tokens(token_int)token = token[: valid_token_num - self.pred_bias]# texts = sentence_postprocess(token)- return token+ return token\ No newline at end of file