Merge pull request #473 from alibaba-damo-academy/dev_smohan

jmwang66

2023-05-09 8dab6d184a034ca86eafa644ea0d2100aadfe27d

 funasr/bin/tp_inference.py

@@ -54,7 +54,7 @@
        assert check_argument_types()
        # 1. Build ASR model
        tp_model, tp_train_args = ASRTask.build_model_from_file(
            timestamp_infer_config, timestamp_model_file, device
            timestamp_infer_config, timestamp_model_file, device=device
        )
        if 'cuda' in device:
            tp_model = tp_model.cuda()  # force model to cuda
@@ -116,8 +116,8 @@
            enc = enc[0]

        # c. Forward Predictor
        _, _, us_alphas, us_cif_peak = self.tp_model.calc_predictor_timestamp(enc, enc_len, text_lengths.to(self.device)+1)
        return us_alphas, us_cif_peak
        _, _, us_alphas, us_peaks = self.tp_model.calc_predictor_timestamp(enc, enc_len, text_lengths.to(self.device)+1)
        return us_alphas, us_peaks


def inference(
@@ -179,6 +179,9 @@
        **kwargs,
):
    assert check_argument_types()
    ncpu = kwargs.get("ncpu", 1)
    torch.set_num_threads(ncpu)
    
    if batch_size > 1:
        raise NotImplementedError("batch decoding is not implemented")
    if ngpu > 1:
@@ -219,6 +222,13 @@
        split_with_space=split_with_space,
        seg_dict_file=seg_dict_file,
    )

    if output_dir is not None:
        writer = DatadirWriter(output_dir)
        tp_writer = writer[f"timestamp_prediction"]
        # ibest_writer["token_list"][""] = " ".join(speech2text.asr_train_args.token_list)
    else:
        tp_writer = None
    
    def _forward(
            data_path_and_name_and_type,
@@ -227,7 +237,14 @@
            fs: dict = None,
            param_dict: dict = None,
            **kwargs
    ):
    ):  
        output_path = output_dir_v2 if output_dir_v2 is not None else output_dir
        writer = None
        if output_path is not None:
            writer = DatadirWriter(output_path)
            tp_writer = writer[f"timestamp_prediction"]
        else:
            tp_writer = None
        # 3. Build data-iterator
        if data_path_and_name_and_type is None and raw_inputs is not None:
            if isinstance(raw_inputs, torch.Tensor):
@@ -265,6 +282,9 @@
                ts_str, ts_list = ts_prediction_lfr6_standard(us_alphas[batch_id], us_cif_peak[batch_id], token, force_time_shift=-3.0)
                logging.warning(ts_str)
                item = {'key': key, 'value': ts_str, 'timestamp':ts_list}
                if tp_writer is not None:
                    tp_writer["tp_sync"][key+'#'] = ts_str
                    tp_writer["tp_time"][key+'#'] = str(ts_list)
                tp_result_list.append(item)
        return tp_result_list