FUNASR训练
parent: 84b4a019 | 补丁 | 提交 | show whitespace
haoneng.lhn
2023-05-30 d9984c2062adaef959b470f5608ca99580ef909f 
update
1个文件已修改
4 ■■■■■ 已修改文件
funasr/models/e2e_asr_paraformer.py 4 ●●●●● 补丁 | 查看 | 原始文档 | blame | 历史 
 funasr/models/e2e_asr_paraformer.py


@@ -1160,8 +1160,8 @@


                                                                                           mask_chunk_predictor=mask_chunk_predictor,


                                                                                           target_label_length=None,


                                                                                           )


        predictor_alignments, predictor_alignments_len = self.predictor.gen_frame_alignments(pre_alphas[:, :-1],


                                                                                             encoder_out_lens)


        predictor_alignments, predictor_alignments_len = self.predictor.gen_frame_alignments(pre_alphas,


                                                                                             encoder_out_lens+1 if self.predictor.tail_threshold > 0.0 else encoder_out_lens)




        scama_mask = None


        if self.encoder.overlap_chunk_cls is not None and self.decoder_attention_chunk_type == 'chunk':