paraformer streaming bugfix

游雁

2024-03-18 cbe2ea7e07cbf364827bd89cefc42b3f643ea3be

paraformer streaming bugfix

 funasr/models/paraformer_streaming/model.py

@@ -532,8 +532,10 @@
            kwargs["is_final"] = _is_final and i == n -1
            audio_sample_i = audio_sample[i*chunk_stride_samples:(i+1)*chunk_stride_samples]
            if kwargs["is_final"] and len(audio_sample_i) < 960:
                continue

                cache["encoder"]["tail_chunk"] = True
                speech = cache["encoder"]["feats"]
                speech_lengths = torch.tensor([speech.shape[1]], dtype=torch.int64).to(speech.device)
            else:
            # extract fbank feats
            speech, speech_lengths = extract_fbank([audio_sample_i], data_type=kwargs.get("data_type", "sound"),
                                                   frontend=frontend, cache=cache["frontend"], is_final=kwargs["is_final"])