fix bug: 模型初始化可传入参数disable_pbar=True (#1387)

Yuming Zhang

2024-02-23 7759ab5febbccdeb929c34a5d5dd4b5387bfd83c

fix bug: 模型初始化可传入参数disable_pbar=True (#1387)

Co-authored-by: 张玉明 <zhangyuming@wepie.com>

 funasr/auto/auto_model.py

@@ -219,7 +219,7 @@
        speed_stats = {}
        asr_result_list = []
        num_samples = len(data_list)
        disable_pbar = kwargs.get("disable_pbar", False)
        disable_pbar = self.kwargs.get("disable_pbar", False)
        pbar = tqdm(colour="blue", total=num_samples, dynamic_ncols=True) if not disable_pbar else None
        time_speech_total = 0.0
        time_escape_total = 0.0
@@ -263,18 +263,16 @@
        return asr_result_list
    
    def inference_with_vad(self, input, input_len=None, **cfg):
        
        kwargs = self.kwargs
        # step.1: compute the vad model
        self.vad_kwargs.update(cfg)
        beg_vad = time.time()
        res = self.inference(input, input_len=input_len, model=self.vad_model, kwargs=self.vad_kwargs, **cfg)
        end_vad = time.time()
        print(f"time cost vad: {end_vad - beg_vad:0.3f}")


        # step.2 compute asr model
        model = self.model
        kwargs = self.kwargs
        kwargs.update(cfg)
        batch_size = int(kwargs.get("batch_size_s", 300))*1000
        batch_size_threshold_ms = int(kwargs.get("batch_size_threshold_s", 60))*1000
@@ -285,7 +283,7 @@
        time_speech_total_all_samples = 1e-6

        beg_total = time.time()
        pbar_total = tqdm(colour="red", total=len(res), dynamic_ncols=True)
        pbar_total = tqdm(colour="red", total=len(res), dynamic_ncols=True) if not kwargs.get("disable_pbar", False) else None
        for i in range(len(res)):
            key = res[i]["key"]
            vadsegments = res[i]["value"]
@@ -323,7 +321,7 @@
                batch_size_ms_cum = 0
                end_idx = j + 1
                speech_j, speech_lengths_j = slice_padding_audio_samples(speech, speech_lengths, sorted_data[beg_idx:end_idx])       
                results = self.inference(speech_j, input_len=None, model=model, kwargs=kwargs, disable_pbar=True, **cfg)
                results = self.inference(speech_j, input_len=None, model=model, kwargs=kwargs, **cfg)
                if self.spk_model is not None:
                    # compose vad segments: [[start_time_sec, end_time_sec, speech], [...]]
                    for _b in range(len(speech_j)):
@@ -333,7 +331,7 @@
                        segments = sv_chunk(vad_segments)
                        all_segments.extend(segments)
                        speech_b = [i[2] for i in segments]
                        spk_res = self.inference(speech_b, input_len=None, model=self.spk_model, kwargs=kwargs, disable_pbar=True, **cfg)
                        spk_res = self.inference(speech_b, input_len=None, model=self.spk_model, kwargs=kwargs, **cfg)
                        results[_b]['spk_embedding'] = spk_res[0]['spk_embedding']
                beg_idx = end_idx
                if len(results) < 1:
@@ -384,7 +382,7 @@
            # step.3 compute punc model
            if self.punc_model is not None:
                self.punc_kwargs.update(cfg)
                punc_res = self.inference(result["text"], model=self.punc_model, kwargs=self.punc_kwargs, disable_pbar=True, **cfg)
                punc_res = self.inference(result["text"], model=self.punc_model, kwargs=self.punc_kwargs, **cfg)
                raw_text = copy.copy(result["text"])
                if return_raw_text: result['raw_text'] = raw_text
                result["text"] = punc_res[0]["text"]
@@ -434,6 +432,7 @@
            results_ret_list.append(result)
            end_asr_total = time.time()
            time_escape_total_per_sample = end_asr_total - beg_asr_total
            if pbar_total:
            pbar_total.update(1)
            pbar_total.set_description(f"rtf_avg: {time_escape_total_per_sample / time_speech_total_per_sample:0.3f}, "
                                 f"time_speech: {time_speech_total_per_sample: 0.3f}, "