funasr pages

游雁

2023-11-16 4ace5a95b052d338947fc88809a440ccd55cf6b4

 funasr/bin/asr_inference_launch.py

@@ -462,18 +462,18 @@


def inference_paraformer_vad_punc(
        maxlenratio: float,
        minlenratio: float,
        batch_size: int,
        beam_size: int,
        ngpu: int,
        ctc_weight: float,
        lm_weight: float,
        penalty: float,
        log_level: Union[int, str],
        maxlenratio: float=0.0,
        minlenratio: float=0.0,
        batch_size: int=1,
        beam_size: int=1,
        ngpu: int=1,
        ctc_weight: float=0.0,
        lm_weight: float=0.0,
        penalty: float=0.0,
        log_level: Union[int, str]=logging.ERROR,
        # data_path_and_name_and_type,
        asr_train_config: Optional[str],
        asr_model_file: Optional[str],
        asr_train_config: Optional[str]=None,
        asr_model_file: Optional[str]=None,
        cmvn_file: Optional[str] = None,
        lm_train_config: Optional[str] = None,
        lm_file: Optional[str] = None,
@@ -487,7 +487,7 @@
        seed: int = 0,
        ngram_weight: float = 0.9,
        nbest: int = 1,
        num_workers: int = 1,
        num_workers: int = 0,
        vad_infer_config: Optional[str] = None,
        vad_model_file: Optional[str] = None,
        vad_cmvn_file: Optional[str] = None,
@@ -815,8 +815,7 @@
        format="%(asctime)s (%(module)s:%(lineno)d) %(levelname)s: %(message)s",
    )

    if sv_model_file is None:
        sv_model_file = "{}/damo/speech_paraformer-large-vad-punc-spk_asr_nat-zh-cn/campplus_cn_common.bin".format(get_cache_dir(None))
    sv_model_file = asr_model_file.replace("model.pb", "campplus_cn_common.bin")

    if param_dict is not None:
        hotword_list_or_file = param_dict.get('hotword')
@@ -1099,18 +1098,18 @@


def inference_paraformer_online(
        maxlenratio: float,
        minlenratio: float,
        batch_size: int,
        beam_size: int,
        ngpu: int,
        ctc_weight: float,
        lm_weight: float,
        penalty: float,
        log_level: Union[int, str],
        maxlenratio: float=0.0,
        minlenratio: float=0.0,
        batch_size: int=1,
        beam_size: int=1,
        ngpu: int=1,
        ctc_weight: float=0.0,
        lm_weight: float=0.0,
        penalty: float=0.0,
        log_level: Union[int, str]=logging.ERROR,
        # data_path_and_name_and_type,
        asr_train_config: Optional[str],
        asr_model_file: Optional[str],
        asr_train_config: Optional[str]=None,
        asr_model_file: Optional[str]=None,
        cmvn_file: Optional[str] = None,
        lm_train_config: Optional[str] = None,
        lm_file: Optional[str] = None,
@@ -2056,6 +2055,12 @@

    ncpu = kwargs.get("ncpu", 1)
    torch.set_num_threads(ncpu)
    if param_dict:
        language = param_dict.get("language", None)
        task = param_dict.get("task", "transcribe")
    else:
        language = None
        task = "transcribe"
    if batch_size > 1:
        raise NotImplementedError("batch decoding is not implemented")
    if word_lm_train_config is not None:
@@ -2099,6 +2104,8 @@
        penalty=penalty,
        nbest=nbest,
        streaming=streaming,
        language=language,
        task=task,
    )
    logging.info("speech2text_kwargs: {}".format(speech2text_kwargs))
    speech2text = Speech2TextWhisper(**speech2text_kwargs)