python runtime

游雁

2024-07-22 aa3fe1a353bde71d106755d030d9e5300fbde328

 funasr/auto/auto_model.py

@@ -114,7 +114,7 @@
        try:
            from funasr.utils.version_checker import check_for_update

            check_for_update()
            check_for_update(disable=kwargs.get("disable_update", False))
        except:
            pass

@@ -337,9 +337,11 @@
        end_vad = time.time()

        #  FIX(gcf): concat the vad clips for sense vocie model for better aed
        if kwargs.get("merge_vad", False):
        if cfg.get("merge_vad", False):
            for i in range(len(res)):
                res[i]["value"] = merge_vad(res[i]["value"], kwargs.get("merge_length", 15000))
                res[i]["value"] = merge_vad(
                    res[i]["value"], kwargs.get("merge_length_s", 15) * 1000
                )

        # step.2 compute asr model
        model = self.model
@@ -379,6 +381,9 @@

            if len(sorted_data) > 0 and len(sorted_data[0]) > 0:
                batch_size = max(batch_size, sorted_data[0][0][1] - sorted_data[0][0][0])

            if kwargs["device"] == "cpu":
                batch_size = 0

            beg_idx = 0
            beg_asr_total = time.time()
@@ -506,8 +511,8 @@
                sv_output = postprocess(all_segments, None, labels, spk_embedding.cpu())
                if self.spk_mode == "vad_segment":  # recover sentence_list
                    sentence_list = []
                    for res, vadsegment in zip(restored_data, vadsegments):
                        if "timestamp" not in res:
                    for rest, vadsegment in zip(restored_data, vadsegments):
                        if "timestamp" not in rest:
                            logging.error(
                                "Only 'iic/speech_paraformer-large-vad-punc_asr_nat-zh-cn-16k-common-vocab8404-pytorch' \
                                           and 'iic/speech_seaco_paraformer_large_asr_nat-zh-cn-16k-common-vocab8404-pytorch'\
@@ -517,8 +522,8 @@
                            {
                                "start": vadsegment[0],
                                "end": vadsegment[1],
                                "sentence": res["text"],
                                "timestamp": res["timestamp"],
                                "sentence": rest["text"],
                                "timestamp": rest["timestamp"],
                            }
                        )
                elif self.spk_mode == "punc_segment":