| | |
|---|
| | | assert text_lengths.dim() == 1, text_lengths.shape |
|---|
| | | # Check that batch_size is unified |
|---|
| | | assert ( |
|---|
| | | speech.shape[0] |
|---|
| | | == speech_lengths.shape[0] |
|---|
| | | == text.shape[0] |
|---|
| | | == text_lengths.shape[0] |
|---|
| | | speech.shape[0] == speech_lengths.shape[0] == text.shape[0] == text_lengths.shape[0] |
|---|
| | | ), (speech.shape, speech_lengths.shape, text.shape, text_lengths.shape) |
|---|
| | | batch_size = speech.shape[0] |
|---|
| | | |
|---|
| | |
|---|
| | | loss_interctc = loss_interctc / len(intermediate_outs) |
|---|
| | | |
|---|
| | | # calculate whole encoder loss |
|---|
| | | loss_ctc = ( |
|---|
| | | 1 - self.interctc_weight |
|---|
| | | ) * loss_ctc + self.interctc_weight * loss_interctc |
|---|
| | | loss_ctc = (1 - self.interctc_weight) * loss_ctc + self.interctc_weight * loss_interctc |
|---|
| | | |
|---|
| | | if self.use_transducer_decoder: |
|---|
| | | # 2a. Transducer decoder branch |
|---|
| | |
|---|
| | | |
|---|
| | | # Post-encoder, e.g. NLU |
|---|
| | | if self.postencoder is not None: |
|---|
| | | encoder_out, encoder_out_lens = self.postencoder( |
|---|
| | | encoder_out, encoder_out_lens |
|---|
| | | ) |
|---|
| | | encoder_out, encoder_out_lens = self.postencoder(encoder_out, encoder_out_lens) |
|---|
| | | |
|---|
| | | return encoder_out, encoder_out_lens |
|---|
| | | |
|---|
