| | |
|---|
| | | # Copyright FunASR (https://github.com/FunAudioLLM/SenseVoice). All Rights Reserved. |
|---|
| | | # MIT License (https://opensource.org/licenses/MIT) |
|---|
| | | |
|---|
| | | |
|---|
| | | import torch |
|---|
| | | import os.path |
|---|
| | | import librosa |
|---|
| | | import numpy as np |
|---|
| | |
|---|
| | | return self.lid_dict[lid] |
|---|
| | | else: |
|---|
| | | raise ValueError( |
|---|
| | | f"The language {l} is not in {list(self.lid_dict.keys())}" |
|---|
| | | f"The language {lid} is not in {list(self.lid_dict.keys())}" |
|---|
| | | ) |
|---|
| | | |
|---|
| | | def _get_tnid(self, tnid): |
|---|
| | |
|---|
| | | ) |
|---|
| | | for b in range(feats.shape[0]): |
|---|
| | | # back to torch.Tensor |
|---|
| | | if isinstance(ctc_logits, np.ndarray): |
|---|
| | | ctc_logits = torch.from_numpy(ctc_logits).float() |
|---|
| | | # if isinstance(ctc_logits, np.ndarray): |
|---|
| | | # ctc_logits = torch.from_numpy(ctc_logits).float() |
|---|
| | | # support batch_size=1 only currently |
|---|
| | | x = ctc_logits[b, : encoder_out_lens[b].item(), :] |
|---|
| | | yseq = x.argmax(dim=-1) |
|---|
| | | yseq = torch.unique_consecutive(yseq, dim=-1) |
|---|
| | | yseq = np.argmax(x, axis=-1) |
|---|
| | | # Use np.diff and np.where instead of torch.unique_consecutive. |
|---|
| | | mask = np.concatenate(([True], np.diff(yseq) != 0)) |
|---|
| | | yseq = yseq[mask] |
|---|
| | | |
|---|
| | | mask = yseq != self.blank_id |
|---|
| | | token_int = yseq[mask].tolist() |
|---|
