| | |
|---|
| | | decoder_out_1st = None |
|---|
| | | pre_loss_att = None |
|---|
| | | if self.sampling_ratio > 0.0: |
|---|
| | | if self.step_cur < 2: |
|---|
| | | logging.info("enable sampler in paraformer, sampling_ratio: {}".format(self.sampling_ratio)) |
|---|
| | | |
|---|
| | | if self.use_1st_decoder_loss: |
|---|
| | | sematic_embeds, decoder_out_1st, pre_loss_att = \ |
|---|
| | | self.sampler_with_grad(encoder_out, encoder_out_lens, ys_pad, |
|---|
| | |
|---|
| | | self.sampler(encoder_out, encoder_out_lens, ys_pad, |
|---|
| | | ys_pad_lens, pre_acoustic_embeds, scama_mask) |
|---|
| | | else: |
|---|
| | | if self.step_cur < 2: |
|---|
| | | logging.info("disable sampler in paraformer, sampling_ratio: {}".format(self.sampling_ratio)) |
|---|
| | | sematic_embeds = pre_acoustic_embeds |
|---|
| | | |
|---|
| | | # 1. Forward decoder |
|---|
| | |
|---|
| | | logging.info("enable beam_search") |
|---|
| | | self.init_beam_search(**kwargs) |
|---|
| | | self.nbest = kwargs.get("nbest", 1) |
|---|
| | | |
|---|
| | | |
|---|
| | | |
|---|
| | | if len(cache) == 0: |
|---|
| | | self.init_cache(cache, **kwargs) |
|---|
| | | |
|---|
| | |
|---|
| | | self.init_cache(cache, **kwargs) |
|---|
| | | |
|---|
| | | if kwargs.get("output_dir"): |
|---|
| | | writer = DatadirWriter(kwargs.get("output_dir")) |
|---|
| | | ibest_writer = writer[f"{1}best_recog"] |
|---|
| | | if not hasattr(self, "writer"): |
|---|
| | | self.writer = DatadirWriter(kwargs.get("output_dir")) |
|---|
| | | ibest_writer = self.writer[f"{1}best_recog"] |
|---|
| | | ibest_writer["token"][key[0]] = " ".join(tokens) |
|---|
| | | ibest_writer["text"][key[0]] = text_postprocessed |
|---|
| | | |
|---|
| | | |
|---|
| | | return result, meta_data |
|---|
| | | |
|---|
| | | |
|---|
