fix progress bar for batch_size (#1917)

彭震东

2024-07-15 f2ed4b3856eaed8abe568e6904ffd8dc3a799f5f

 funasr/auto/auto_model.py

@@ -20,7 +20,7 @@
from funasr.download.file import download_from_url
from funasr.utils.timestamp_tools import timestamp_sentence
from funasr.utils.timestamp_tools import timestamp_sentence_en
from funasr.download.download_from_hub import download_model
from funasr.download.download_model_from_hub import download_model
from funasr.utils.vad_utils import slice_padding_audio_samples
from funasr.utils.vad_utils import merge_vad
from funasr.utils.load_utils import load_audio_text_image_video
@@ -92,7 +92,8 @@
                if isinstance(data_i, str) and os.path.exists(data_i):
                    key = misc.extract_filename_without_extension(data_i)
                else:
                    key = "rand_key_" + "".join(random.choice(chars) for _ in range(13))
                    if key is None:
                        key = "rand_key_" + "".join(random.choice(chars) for _ in range(13))
                key_list.append(key)

    else:  # raw text; audio sample point, fbank; bytes
@@ -110,11 +111,15 @@

    def __init__(self, **kwargs):

        try:
            from funasr.utils.version_checker import check_for_update

            check_for_update()
        except:
            pass

        log_level = getattr(logging, kwargs.get("log_level", "INFO").upper())
        logging.basicConfig(level=log_level)

        if not kwargs.get("disable_log", True):
            tables.print()

        model, kwargs = self.build_model(**kwargs)

@@ -163,7 +168,8 @@
        self.spk_kwargs = spk_kwargs
        self.model_path = kwargs.get("model_path")

    def build_model(self, **kwargs):
    @staticmethod
    def build_model(**kwargs):
        assert "model" in kwargs
        if "model_conf" not in kwargs:
            logging.info("download models from model hub: {}".format(kwargs.get("hub", "ms")))
@@ -209,11 +215,11 @@
        kwargs["frontend"] = frontend
        # build model
        model_class = tables.model_classes.get(kwargs["model"])
        assert model_class is not None, f'{kwargs["model"]} is not registered'
        model_conf = {}
        deep_update(model_conf, kwargs.get("model_conf", {}))
        deep_update(model_conf, kwargs)
        model = model_class(**model_conf, vocab_size=vocab_size)
        model.to(device)

        # init_param
        init_param = kwargs.get("init_param", None)
@@ -236,6 +242,11 @@
            model.to(torch.float16)
        elif kwargs.get("bf16", False):
            model.to(torch.bfloat16)
        model.to(device)

        if not kwargs.get("disable_log", True):
            tables.print()

        return model, kwargs

    def __call__(self, *args, **cfg):
@@ -304,7 +315,7 @@
            speed_stats["rtf"] = f"{(time_escape) / batch_data_time:0.3f}"
            description = f"{speed_stats}, "
            if pbar:
                pbar.update(1)
                pbar.update(end_idx - beg_idx)
                pbar.set_description(description)
            time_speech_total += batch_data_time
            time_escape_total += time_escape
@@ -324,7 +335,7 @@
            input, input_len=input_len, model=self.vad_model, kwargs=self.vad_kwargs, **cfg
        )
        end_vad = time.time()
            

        #  FIX(gcf): concat the vad clips for sense vocie model for better aed
        if kwargs.get("merge_vad", False):
            for i in range(len(res)):
@@ -466,7 +477,7 @@
                            result[k] = restored_data[j][k]
                        else:
                            result[k] += restored_data[j][k]
                            

            if not len(result["text"].strip()):
                continue
            return_raw_text = kwargs.get("return_raw_text", False)
@@ -481,7 +492,7 @@
                if return_raw_text:
                    result["raw_text"] = raw_text
                result["text"] = punc_res[0]["text"]
                

            # speaker embedding cluster after resorted
            if self.spk_model is not None and kwargs.get("return_spk_res", True):
                if raw_text is None:
@@ -602,12 +613,6 @@
        )

        with torch.no_grad():

            if type == "onnx":
                export_dir = export_utils.export_onnx(model=model, data_in=data_list, **kwargs)
            else:
                export_dir = export_utils.export_torchscripts(
                    model=model, data_in=data_list, **kwargs
                )
            export_dir = export_utils.export(model=model, data_in=data_list, **kwargs)

        return export_dir