update data2vec pretrain: add clipping

jmwang66

2023-02-06 9befa9e508d5ca95cb5faa29cd20d23e04e525c9

 funasr/bin/asr_inference_uniasr.py

@@ -96,11 +96,14 @@
        else:
            decoder = asr_model.decoder2

        ctc = CTCPrefixScorer(ctc=asr_model.ctc, eos=asr_model.eos)
        if asr_model.ctc != None:
            ctc = CTCPrefixScorer(ctc=asr_model.ctc, eos=asr_model.eos)
            scorers.update(
                ctc=ctc
            )
        token_list = asr_model.token_list
        scorers.update(
            decoder=decoder,
            ctc=ctc,
            length_bonus=LengthBonus(len(token_list)),
        )

@@ -388,7 +391,7 @@
#         except TooShortUttError as e:
#             logging.warning(f"Utterance {keys} {e}")
#             hyp = Hypothesis(score=0.0, scores={}, states={}, yseq=[])
#             results = [[" ", ["<space>"], [2], hyp]] * nbest
#             results = [[" ", ["sil"], [2], hyp]] * nbest
#
#         # Only supporting batch_size==1
#         key = keys[0]
@@ -613,7 +616,7 @@
            except TooShortUttError as e:
                logging.warning(f"Utterance {keys} {e}")
                hyp = Hypothesis(score=0.0, scores={}, states={}, yseq=[])
                results = [[" ", ["<space>"], [2], hyp]] * nbest
                results = [[" ", ["sil"], [2], hyp]] * nbest
    
            # Only supporting batch_size==1
            key = keys[0]