| | |
|---|
| | | time2 = time.perf_counter() |
|---|
| | | meta_data["load_data"] = f"{time2 - time1:0.3f}" |
|---|
| | | audio_sample_list = sample_list[0] |
|---|
| | | ocr_sample_list = sample_list[1] |
|---|
| | | if len(sample_list) >1: |
|---|
| | | ocr_sample_list = sample_list[1] |
|---|
| | | else: |
|---|
| | | ocr_sample_list = [[294, 0]] |
|---|
| | | speech, speech_lengths = extract_fbank(audio_sample_list, data_type=kwargs.get("data_type", "sound"), |
|---|
| | | frontend=frontend) |
|---|
| | | time3 = time.perf_counter() |
|---|
| | |
|---|
| | | encoder_out, encoder_out_lens = self.encode(speech, speech_lengths) |
|---|
| | | if isinstance(encoder_out, tuple): |
|---|
| | | encoder_out = encoder_out[0] |
|---|
| | | |
|---|
| | | |
|---|
| | | ocr_list_new = [[x + 1 if x != 0 else x for x in sublist] for sublist in ocr_sample_list] |
|---|
| | | ocr = torch.tensor(ocr_list_new).to(device=kwargs["device"]) |
|---|
| | | ocr_lengths = ocr.new_full([1], dtype=torch.long, fill_value=ocr.size(1)).to(device=kwargs["device"]) |
|---|
