train update

游雁

2024-03-27 9b4e9cc8a0311e5243d69b73ed073e7ea441982e

 funasr/auto/auto_model.py

@@ -25,11 +25,12 @@
from funasr.train_utils.set_all_random_seed import set_all_random_seed
from funasr.train_utils.load_pretrained_model import load_pretrained_model
from funasr.utils import export_utils

try:
    from funasr.models.campplus.utils import sv_chunk, postprocess, distribute_spk
    from funasr.models.campplus.cluster_backend import ClusterBackend
except:
    print("Notice: If you want to use the speaker diarization, please `pip install hdbscan`")
    pass


def prepare_data_iterator(data_in, input_len=None, data_type=None, key=None):
@@ -110,7 +111,7 @@
        if vad_model is not None:
            logging.info("Building VAD model.")
            vad_kwargs["model"] = vad_model
            vad_kwargs["model_revision"] = kwargs.get("vad_model_revision", None)
            vad_kwargs["model_revision"] = kwargs.get("vad_model_revision", "master")
            vad_kwargs["device"] = kwargs["device"]
            vad_model, vad_kwargs = self.build_model(**vad_kwargs)

@@ -120,7 +121,7 @@
        if punc_model is not None:
            logging.info("Building punc model.")
            punc_kwargs["model"] = punc_model
            punc_kwargs["model_revision"] = kwargs.get("punc_model_revision", None)
            punc_kwargs["model_revision"] = kwargs.get("punc_model_revision", "master")
            punc_kwargs["device"] = kwargs["device"]
            punc_model, punc_kwargs = self.build_model(**punc_kwargs)

@@ -130,7 +131,7 @@
        if spk_model is not None:
            logging.info("Building SPK model.")
            spk_kwargs["model"] = spk_model
            spk_kwargs["model_revision"] = kwargs.get("spk_model_revision", None)
            spk_kwargs["model_revision"] = kwargs.get("spk_model_revision", "master")
            spk_kwargs["device"] = kwargs["device"]
            spk_model, spk_kwargs = self.build_model(**spk_kwargs)
            self.cb_model = ClusterBackend().to(kwargs["device"])
@@ -312,7 +313,8 @@
            key = res[i]["key"]
            vadsegments = res[i]["value"]
            input_i = data_list[i]
            speech = load_audio_text_image_video(input_i, fs=kwargs["frontend"].fs, audio_fs=kwargs.get("fs", 16000))
            fs = kwargs["frontend"].fs if hasattr(kwargs["frontend"], "fs") else 16000
            speech = load_audio_text_image_video(input_i, fs=fs, audio_fs=kwargs.get("fs", 16000))
            speech_lengths = len(speech)
            n = len(vadsegments)
            data_with_index = [(vadsegments[i], i) for i in range(n)]