update raw_text

shixian.shi

2024-01-22 0ee4565c8ec3deffde8362370aee306ae7029ebb

 funasr/auto/auto_model.py

@@ -6,6 +6,7 @@
import string
import logging
import os.path
import numpy as np
from tqdm import tqdm
from omegaconf import DictConfig, OmegaConf, ListConfig

@@ -96,7 +97,7 @@
        vad_kwargs = kwargs.get("vad_model_revision", None)
        if vad_model is not None:
            logging.info("Building VAD model.")
            vad_kwargs = {"model": vad_model, "model_revision": vad_kwargs}
            vad_kwargs = {"model": vad_model, "model_revision": vad_kwargs, "device": kwargs["device"]}
            vad_model, vad_kwargs = self.build_model(**vad_kwargs)

        # if punc_model is not None, build punc model else None
@@ -104,7 +105,7 @@
        punc_kwargs = kwargs.get("punc_model_revision", None)
        if punc_model is not None:
            logging.info("Building punc model.")
            punc_kwargs = {"model": punc_model, "model_revision": punc_kwargs}
            punc_kwargs = {"model": punc_model, "model_revision": punc_kwargs, "device": kwargs["device"]}
            punc_model, punc_kwargs = self.build_model(**punc_kwargs)

        # if spk_model is not None, build spk model else None
@@ -112,9 +113,9 @@
        spk_kwargs = kwargs.get("spk_model_revision", None)
        if spk_model is not None:
            logging.info("Building SPK model.")
            spk_kwargs = {"model": spk_model, "model_revision": spk_kwargs}
            spk_kwargs = {"model": spk_model, "model_revision": spk_kwargs, "device": kwargs["device"]}
            spk_model, spk_kwargs = self.build_model(**spk_kwargs)
            self.cb_model = ClusterBackend()
            self.cb_model = ClusterBackend().to(kwargs["device"])
            spk_mode = kwargs.get("spk_mode", 'punc_segment')
            if spk_mode not in ["default", "vad_segment", "punc_segment"]:
                logging.error("spk_mode should be one of default, vad_segment and punc_segment.")
@@ -132,7 +133,8 @@
        self.punc_kwargs = punc_kwargs
        self.spk_model = spk_model
        self.spk_kwargs = spk_kwargs
        self.model_path = kwargs.get("model_path", "./")
        self.model_path = kwargs.get("model_path")

  
        
    def build_model(self, **kwargs):
@@ -333,7 +335,7 @@
                    for _b in range(len(speech_j)):
                        vad_segments = [[sorted_data[beg_idx:end_idx][_b][0][0]/1000.0,
                                        sorted_data[beg_idx:end_idx][_b][0][1]/1000.0,
                                        speech_j[_b]]]
                                        np.array(speech_j[_b])]]
                        segments = sv_chunk(vad_segments)
                        all_segments.extend(segments)
                        speech_b = [i[2] for i in segments]
@@ -390,13 +392,13 @@
            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, **cfg)
                result["text_with_punc"] = punc_res[0]["text"]
                result["text"] = punc_res[0]["text"]
                     
            # speaker embedding cluster after resorted
            if self.spk_model is not None:
                all_segments = sorted(all_segments, key=lambda x: x[0])
                spk_embedding = result['spk_embedding']
                labels = self.cb_model(spk_embedding, oracle_num=self.preset_spk_num)
                labels = self.cb_model(spk_embedding.cpu(), oracle_num=self.preset_spk_num)
                del result['spk_embedding']
                sv_output = postprocess(all_segments, None, labels, spk_embedding.cpu())
                if self.spk_mode == 'vad_segment':
@@ -404,7 +406,7 @@
                    for res, vadsegment in zip(restored_data, vadsegments):
                        sentence_list.append({"start": vadsegment[0],\
                                                "end": vadsegment[1],
                                                "sentence": res['text'],
                                                "sentence": res['raw_text'],
                                                "timestamp": res['timestamp']})
                else: # punc_segment
                    sentence_list = timestamp_sentence(punc_res[0]['punc_array'], \