| | |
|---|
| | | elif isinstance(data_or_path_or_list, str) and data_type == "text" and tokenizer is not None: |
|---|
| | | data_or_path_or_list = tokenizer.encode(data_or_path_or_list) |
|---|
| | | elif isinstance(data_or_path_or_list, np.ndarray): # audio sample point |
|---|
| | | data_or_path_or_list = np.squeeze(data_or_path_or_list) # [n_samples,] |
|---|
| | | data_or_path_or_list = torch.from_numpy(data_or_path_or_list).squeeze() # [n_samples,] |
|---|
| | | else: |
|---|
| | | pass |
|---|
| | | # print(f"unsupport data type: {data_or_path_or_list}, return raw data") |
|---|
| | |
|---|
| | | array = np.frombuffer((middle_data.astype(dtype) - offset) / abs_max, dtype=np.float32) |
|---|
| | | return array |
|---|
| | | |
|---|
| | | def extract_fbank(data, data_len = None, data_type: str="sound", frontend=None): |
|---|
| | | def extract_fbank(data, data_len = None, data_type: str="sound", frontend=None, **kwargs): |
|---|
| | | # import pdb; |
|---|
| | | # pdb.set_trace() |
|---|
| | | if isinstance(data, np.ndarray): |
|---|
| | |
|---|
| | | elif isinstance(data, (list, tuple)): |
|---|
| | | data_list, data_len = [], [] |
|---|
| | | for data_i in data: |
|---|
| | | if isinstance(data, np.ndarray): |
|---|
| | | if isinstance(data_i, np.ndarray): |
|---|
| | | data_i = torch.from_numpy(data_i) |
|---|
| | | data_list.append(data_i) |
|---|
| | | data_len.append(data_i.shape[0]) |
|---|
| | |
|---|
| | | # import pdb; |
|---|
| | | # pdb.set_trace() |
|---|
| | | # if data_type == "sound": |
|---|
| | | data, data_len = frontend(data, data_len) |
|---|
| | | data, data_len = frontend(data, data_len, **kwargs) |
|---|
| | | |
|---|
| | | if isinstance(data_len, (list, tuple)): |
|---|
| | | data_len = torch.tensor([data_len]) |
|---|
