| | |
|---|
| | | ): |
|---|
| | | ncpu = kwargs.get("ncpu", 1) |
|---|
| | | torch.set_num_threads(ncpu) |
|---|
| | | language = kwargs.get("model_lang", None) |
|---|
| | | |
|---|
| | | if word_lm_train_config is not None: |
|---|
| | | raise NotImplementedError("Word LM is not implemented") |
|---|
| | |
|---|
| | | text, token, token_int = result[0], result[1], result[2] |
|---|
| | | time_stamp = result[4] if len(result[4]) > 0 else None |
|---|
| | | |
|---|
| | | if use_timestamp and time_stamp is not None and len(time_stamp): |
|---|
| | | postprocessed_result = postprocess_utils.sentence_postprocess(token, time_stamp) |
|---|
| | | if language == "en-bpe": |
|---|
| | | postprocessed_result = postprocess_utils.sentence_postprocess_sentencepiece(token) |
|---|
| | | else: |
|---|
| | | postprocessed_result = postprocess_utils.sentence_postprocess(token) |
|---|
| | | if use_timestamp and time_stamp is not None and len(time_stamp): |
|---|
| | | postprocessed_result = postprocess_utils.sentence_postprocess(token, time_stamp) |
|---|
| | | else: |
|---|
| | | postprocessed_result = postprocess_utils.sentence_postprocess(token) |
|---|
| | | text_postprocessed = "" |
|---|
| | | time_stamp_postprocessed = "" |
|---|
| | | text_postprocessed_punc = postprocessed_result |
|---|
| | |
|---|
| | | time_stamp_writer: bool = True, |
|---|
| | | punc_infer_config: Optional[str] = None, |
|---|
| | | punc_model_file: Optional[str] = None, |
|---|
| | | sv_model_file: Optional[str] = None, |
|---|
| | | sv_model_file: Optional[str] = "~/.cache/modelscope/hub/damo/speech_paraformer-large-vad-punc-spk_asr_nat-zh-cn/campplus_cn_common.bin", |
|---|
| | | streaming: bool = False, |
|---|
| | | embedding_node: str = "resnet1_dense", |
|---|
| | | sv_threshold: float = 0.9465, |
|---|
| | |
|---|
| | | ##### speaker_verification ##### |
|---|
| | | ################################## |
|---|
| | | # load sv model |
|---|
| | | sv_model_dict = torch.load(sv_model_file, map_location=torch.device('cpu')) |
|---|
| | | sv_model_dict = torch.load(sv_model_file.replace("~", os.environ['HOME']), map_location=torch.device('cpu')) |
|---|
| | | sv_model = CAMPPlus() |
|---|
| | | sv_model.load_state_dict(sv_model_dict) |
|---|
| | | sv_model.eval() |
|---|
| | |
|---|
| | | logging.info("decoding, utt: {}, predictions: {}".format(key, text_postprocessed_punc)) |
|---|
| | | torch.cuda.empty_cache() |
|---|
| | | distribute_spk(asr_result_list[0]['sentences'], sv_output) |
|---|
| | | import pdb; pdb.set_trace() |
|---|
| | | return asr_result_list |
|---|
| | | |
|---|
| | | return _forward |
|---|
| | |
|---|
| | | return inference_paraformer(**kwargs) |
|---|
| | | elif mode == "paraformer_streaming": |
|---|
| | | return inference_paraformer_online(**kwargs) |
|---|
| | | elif mode == "paraformer_vad_speaker": |
|---|
| | | elif mode.startswith("paraformer_vad_speaker"): |
|---|
| | | return inference_paraformer_vad_speaker(**kwargs) |
|---|
| | | elif mode.startswith("paraformer_vad"): |
|---|
| | | return inference_paraformer_vad_punc(**kwargs) |
|---|
