ICASSP 2023论文模型开源|语音分离Mossformer( 四 )

kwargs = dict(model='damo/speech_mossformer_separation_temporal_8k',train_dataset=train_dataset,eval_dataset=eval_dataset,work_dir=work_dir)trainer = build_trainer(Trainers.speech_separation, default_args=kwargs)trainer.model.load_check_point(device=trainer.device)print(trainer.evaluate(None))
相关链接
语音分离模型地址:
2人混合8K音频数据集:
WSJ0数据集:
WSJ0-mix脚本:
Zhao and Bin Ma, “:theLimit ofusing Gated -headwith - Joint Self-”,by2023.