Merge pull request #398 from alibaba-damo-academy/dev_zly2

zhifu gao

2023-04-21 49909f904c1382fd3c23ee64ff6737689216373f

Merge pull request #398 from alibaba-damo-academy/dev_zly2

vad 

 funasr/bin/vad_inference.py

@@ -109,7 +109,7 @@
            fbanks, fbanks_len = self.frontend.forward_fbank(speech, speech_lengths)
            feats, feats_len = self.frontend.forward_lfr_cmvn(fbanks, fbanks_len)
            fbanks = to_device(fbanks, device=self.device)
            # feats = to_device(feats, device=self.device)
            feats = to_device(feats, device=self.device)
            feats_len = feats_len.int()
        else:
            raise Exception("Need to extract feats first, please configure frontend configuration")
@@ -131,7 +131,7 @@
                "in_cache": in_cache
            }
            # a. To device
            batch = to_device(batch, device=self.device)
            #batch = to_device(batch, device=self.device)
            segments_part, in_cache = self.vad_model(**batch)
            if segments_part:
                for batch_num in range(0, self.batch_size):

 funasr/models/frontend/wav_frontend.py

@@ -34,7 +34,7 @@
    means = np.array(means_list).astype(np.float)
    vars = np.array(vars_list).astype(np.float)
    cmvn = np.array([means, vars])
    cmvn = torch.as_tensor(cmvn, dype=torch.float32)
    cmvn = torch.as_tensor(cmvn, dtype=torch.float32)
    return cmvn