search

yongxuUSTC / sednn

deep learning based speech enhancement using keras or pytorch, make it easy to use
http://staff.ustc.edu.cn/~jundu/The%20team/yongxu/demo/SE_DNN_taslp.html
334 stars 125 forks source link

为啥降噪后的语音能量比原始语音还大 #28

Open bailiangze opened 5 years ago

bailiangze commented 5 years ago

TIM图片20190313154158 TIM图片20190313154235

bailiangze commented 5 years ago

上面是降噪后的语音，后面是纯净语音，我直接用的源代码跑出来，为啥能量增大了

qiuqiangkong commented 5 years ago

您好，原因是stft变换后，反变换时需要乘以一个系数，这个系数和np.hamming窗有关，我们没有深入研究。

Best wishes,

Qiuqiang

From: bailiangze notifications@github.com Sent: 13 March 2019 07:45 To: yongxuUSTC/sednn Cc: Subscribed Subject: Re: [yongxuUSTC/sednn] 为啥降噪后的语音能量比原始语音还大 (#28)

上面是降噪后的语音，后面是纯净语音，我直接用的源代码跑出来，为啥能量增大了

― You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHubhttps://emea01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2FyongxuUSTC%2Fsednn%2Fissues%2F28%23issuecomment-472314338&data=02%7C01%7Cq.kong%40surrey.ac.uk%7Ceb2842ab507049f5043c08d6a787d76c%7C6b902693107440aa9e21d89446a2ebb5%7C0%7C1%7C636880599203234760&sdata=ETQU6%2BM%2FrETQQlh397Tu83bqz%2FSSGDiEzNMcKxdCuWI%3D&reserved=0, or mute the threadhttps://emea01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fnotifications%2Funsubscribe-auth%2FAMt5ydZw9mgC0X0Y5XiljY4rP2O8vNHnks5vWKyOgaJpZM4bsliS&data=02%7C01%7Cq.kong%40surrey.ac.uk%7Ceb2842ab507049f5043c08d6a787d76c%7C6b902693107440aa9e21d89446a2ebb5%7C0%7C1%7C636880599203234760&sdata=SZDnEqURmFnosshoYyGAODkjrVFo2lrjY9JQwk2Il%2BU%3D&reserved=0.

bailiangze commented 5 years ago

.........................这个就尴尬了。。。我发现如果在层加上了正则化之后，能量就正常了，但是效果不好了

qiuqiangkong commented 5 years ago

可以检查还原后的音频是否介于[-1, 1]之间，如果超过了这个范围会导致削顶失真。另外PESQ评价指标对幅度变化不敏感。

Best wishes,

Qiuqiang

From: bailiangze notifications@github.com Sent: 15 March 2019 09:09 To: yongxuUSTC/sednn Cc: Kong, Qiuqiang (PG/R - Elec Electronic Eng); Comment Subject: Re: [yongxuUSTC/sednn] 为啥降噪后的语音能量比原始语音还大 (#28)

.........................这个就尴尬了。。。我发现如果在层加上了正则化之后，能量就正常了，但是效果不好了

― You are receiving this because you commented. Reply to this email directly, view it on GitHubhttps://eur02.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2FyongxuUSTC%2Fsednn%2Fissues%2F28%23issuecomment-473210967&data=02%7C01%7Cq.kong%40surrey.ac.uk%7C81e7b23b779140c68f5708d6a925f2dd%7C6b902693107440aa9e21d89446a2ebb5%7C0%7C0%7C636882377786446288&sdata=4yzrudwUGkxvwHYHH0NGk3sZ1Btxqw45T7sPxzR9teY%3D&reserved=0, or mute the threadhttps://eur02.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fnotifications%2Funsubscribe-auth%2FAMt5ySx4ZaLqj6RsSEl4iXEwFS94niyqks5vW2NQgaJpZM4bsliS&data=02%7C01%7Cq.kong%40surrey.ac.uk%7C81e7b23b779140c68f5708d6a925f2dd%7C6b902693107440aa9e21d89446a2ebb5%7C0%7C0%7C636882377786446288&sdata=KTGAKGeB6jPcY0teteBi3DthqHQPgmPJFUgasfyLXWA%3D&reserved=0.