| | |
|---|
| | | 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) |
|---|
| | | cmvn = torch.as_tensor(cmvn, dtype=torch.float32) |
|---|
| | | return cmvn |
|---|
| | | |
|---|
| | | |
|---|
| | | def apply_cmvn(inputs, cmvn_file): # noqa |
|---|
| | | def apply_cmvn(inputs, cmvn): # noqa |
|---|
| | | """ |
|---|
| | | Apply CMVN with mvn data |
|---|
| | | """ |
|---|
| | |
|---|
| | | dtype = inputs.dtype |
|---|
| | | frame, dim = inputs.shape |
|---|
| | | |
|---|
| | | cmvn = load_cmvn(cmvn_file) |
|---|
| | | means = np.tile(cmvn[0:1, :dim], (frame, 1)) |
|---|
| | | vars = np.tile(cmvn[1:2, :dim], (frame, 1)) |
|---|
| | | inputs += torch.from_numpy(means).type(dtype).to(device) |
|---|
| | | inputs *= torch.from_numpy(vars).type(dtype).to(device) |
|---|
| | | means = cmvn[0:1, :dim] |
|---|
| | | vars = cmvn[1:2, :dim] |
|---|
| | | inputs += means.to(device) |
|---|
| | | inputs *= vars.to(device) |
|---|
| | | |
|---|
| | | return inputs.type(torch.float32) |
|---|
| | | |
|---|
| | |
|---|
| | | self.dither = dither |
|---|
| | | self.snip_edges = snip_edges |
|---|
| | | self.upsacle_samples = upsacle_samples |
|---|
| | | self.cmvn = None if self.cmvn_file is None else load_cmvn(self.cmvn_file) |
|---|
| | | |
|---|
| | | def output_size(self) -> int: |
|---|
| | | return self.n_mels * self.lfr_m |
|---|
| | |
|---|
| | | |
|---|
| | | if self.lfr_m != 1 or self.lfr_n != 1: |
|---|
| | | mat = apply_lfr(mat, self.lfr_m, self.lfr_n) |
|---|
| | | if self.cmvn_file is not None: |
|---|
| | | mat = apply_cmvn(mat, self.cmvn_file) |
|---|
| | | if self.cmvn is not None: |
|---|
| | | mat = apply_cmvn(mat, self.cmvn) |
|---|
| | | feat_length = mat.size(0) |
|---|
| | | feats.append(mat) |
|---|
| | | feats_lens.append(feat_length) |
|---|
| | |
|---|
| | | mat = input[i, :input_lengths[i], :] |
|---|
| | | if self.lfr_m != 1 or self.lfr_n != 1: |
|---|
| | | mat = apply_lfr(mat, self.lfr_m, self.lfr_n) |
|---|
| | | if self.cmvn_file is not None: |
|---|
| | | mat = apply_cmvn(mat, self.cmvn_file) |
|---|
| | | if self.cmvn is not None: |
|---|
| | | mat = apply_cmvn(mat, self.cmvn) |
|---|
| | | feat_length = mat.size(0) |
|---|
| | | feats.append(mat) |
|---|
| | | feats_lens.append(feat_length) |
|---|
| | |
|---|
| | | return feats_pad, feats_lens, lfr_splice_frame_idxs |
|---|
| | | |
|---|
| | | def forward( |
|---|
| | | self, input: torch.Tensor, input_lengths: torch.Tensor, is_final: bool = False |
|---|
| | | self, input: torch.Tensor, input_lengths: torch.Tensor, is_final: bool = False, reset: bool = False |
|---|
| | | ) -> Tuple[torch.Tensor, torch.Tensor]: |
|---|
| | | if reset: |
|---|
| | | self.cache_reset() |
|---|
| | | batch_size = input.shape[0] |
|---|
| | | assert batch_size == 1, 'we support to extract feature online only when the batch size is equal to 1 now' |
|---|
| | | waveforms, feats, feats_lengths = self.forward_fbank(input, input_lengths) # input shape: B T D |
|---|
