Update auto_model.py

liugz18

2024-06-12 648c03fdbba045acf8e25c5e6f7df0f27937c3cf

 funasr/auto/auto_model.py

@@ -234,6 +234,8 @@
        # fp16
        if kwargs.get("fp16", False):
            model.to(torch.float16)
        elif kwargs.get("bf16", False):
            model.to(torch.bfloat16)
        return model, kwargs

    def __call__(self, *args, **cfg):
@@ -428,6 +430,10 @@
            #                      f"time_speech_total_per_sample: {time_speech_total_per_sample: 0.3f}, "
            #                      f"time_escape_total_per_sample: {time_escape_total_per_sample:0.3f}")

            if len(results_sorted) != n:
                results_ret_list.append({"key": key, "text": "", "timestamp": []})
                logging.info("decoding, utt: {}, empty result".format(key))
                continue
            restored_data = [0] * n
            for j in range(n):
                index = sorted_data[j][1]
@@ -463,10 +469,11 @@

            return_raw_text = kwargs.get("return_raw_text", False)
            # step.3 compute punc model
            raw_text = None
            if self.punc_model is not None:
                if not len(result["text"].strip()):
                    if return_raw_text:
                        result["raw_text"] = ""
                        result["raw_text"] = raw_text = ""
                else:
                    deep_update(self.punc_kwargs, cfg)
                    punc_res = self.inference(
@@ -476,9 +483,7 @@
                    if return_raw_text:
                        result["raw_text"] = raw_text
                    result["text"] = punc_res[0]["text"]
            else:
                raw_text = None

                
            # speaker embedding cluster after resorted
            if self.spk_model is not None and kwargs.get("return_spk_res", True):
                if raw_text is None: