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 124 forks source link

关于增强后的语音问题 #57

Open lymiou opened 4 years ago

lymiou commented 4 years ago

您好,我想问下为什么增强后的语音发出来的只有吱吱吱的声音呢? 下面是它对应的语谱图,请问下是出了什么问题呢?期待您的解答,谢谢! image

yongxuUSTC commented 4 years ago

0dB的时候 首先 任务比较难,确实 有吱吱的声音,主要是模型不够强大,带来的非线性失真

On Fri, 15 May 2020 at 19:37, lymiou notifications@github.com wrote:

您好,我想问下为什么增强后的语音发出来的只有吱吱吱的声音呢? 下面是它对应的语谱图,请问下是出了什么问题呢?期待您的解答,谢谢! [image: image] https://user-images.githubusercontent.com/46339102/82108425-e69e2580-9760-11ea-9ed4-871079d63e87.png

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/yongxuUSTC/sednn/issues/57, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABJGHUXL7VRVOZSQ3G2OFRLRRX35HANCNFSM4NCWY3CQ .

lymioumm commented 4 years ago

0dB的时候 首先 任务比较难,确实 有吱吱的声音,主要是模型不够强大,带来的非线性失真 On Fri, 15 May 2020 at 19:37, lymiou @.***> wrote: 您好,我想问下为什么增强后的语音发出来的只有吱吱吱的声音呢? 下面是它对应的语谱图,请问下是出了什么问题呢?期待您的解答,谢谢! [image: image] https://user-images.githubusercontent.com/46339102/82108425-e69e2580-9760-11ea-9ed4-871079d63e87.png — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#57>, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABJGHUXL7VRVOZSQ3G2OFRLRRX35HANCNFSM4NCWY3CQ .

谢谢您的解答,还有一点想麻烦您解答一下,项目里的特征提取采用的是什么方法呢?好像不是MFCC,再次感谢您!

qiuqiangkong commented 4 years ago

您好,

分离的结果是对的,是否使用了全部数据训练?、

On Sat, 16 May 2020 at 10:37, lymiou notifications@github.com wrote:

您好,我想问下为什么增强后的语音发出来的只有吱吱吱的声音呢? 下面是它对应的语谱图,请问下是出了什么问题呢?期待您的解答,谢谢! [image: image] https://user-images.githubusercontent.com/46339102/82108425-e69e2580-9760-11ea-9ed4-871079d63e87.png

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/yongxuUSTC/sednn/issues/57, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADFXTSMCYXPOHXIGFN3PMQDRRX35HANCNFSM4NCWY3CQ .

lymioumm commented 4 years ago

您好, 分离的结果是对的,是否使用了全部数据训练?、 On Sat, 16 May 2020 at 10:37, lymiou @.***> wrote: 您好,我想问下为什么增强后的语音发出来的只有吱吱吱的声音呢? 下面是它对应的语谱图,请问下是出了什么问题呢?期待您的解答,谢谢! [image: image] https://user-images.githubusercontent.com/46339102/82108425-e69e2580-9760-11ea-9ed4-871079d63e87.png — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#57>, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADFXTSMCYXPOHXIGFN3PMQDRRX35HANCNFSM4NCWY3CQ .

您好,是的,使用了mini_data文件夹下的数据,另外train_speech中还另加了一些语音数据

qiuqiangkong commented 4 years ago

数据太少了,需要使用全部timit数据训练

On Sun, 17 May 2020 at 09:56, lymioumm notifications@github.com wrote:

您好, 分离的结果是对的,是否使用了全部数据训练?、 … <#m-3999892883733078355> On Sat, 16 May 2020 at 10:37, lymiou @.***> wrote: 您好,我想问下为什么增强后的语音发出来的只有吱吱吱的声音呢? 下面是它对应的语谱图,请问下是出了什么问题呢?期待您的解答,谢谢! [image: image] https://user-images.githubusercontent.com/46339102/82108425-e69e2580-9760-11ea-9ed4-871079d63e87.png — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#57 https://github.com/yongxuUSTC/sednn/issues/57>, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADFXTSMCYXPOHXIGFN3PMQDRRX35HANCNFSM4NCWY3CQ .

您好,是的,使用了mini_data文件夹下的数据,另外train_speech中还另加了一些语音数据

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/yongxuUSTC/sednn/issues/57#issuecomment-629730966, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADFXTSO5K22PTO3HI6A355LRR474NANCNFSM4NCWY3CQ .

lymioumm commented 4 years ago

数据太少了,需要使用全部timit数据训练 On Sun, 17 May 2020 at 09:56, lymioumm @.> wrote: 您好, 分离的结果是对的,是否使用了全部数据训练?、 … <#m-3999892883733078355> On Sat, 16 May 2020 at 10:37, lymiou @.> wrote: 您好,我想问下为什么增强后的语音发出来的只有吱吱吱的声音呢? 下面是它对应的语谱图,请问下是出了什么问题呢?期待您的解答,谢谢! [image: image] https://user-images.githubusercontent.com/46339102/82108425-e69e2580-9760-11ea-9ed4-871079d63e87.png — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#57 <#57>>, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADFXTSMCYXPOHXIGFN3PMQDRRX35HANCNFSM4NCWY3CQ . 您好,是的,使用了mini_data文件夹下的数据,另外train_speech中还另加了一些语音数据 — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#57 (comment)>, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADFXTSO5K22PTO3HI6A355LRR474NANCNFSM4NCWY3CQ .

好的,明白了,谢谢您!

QianYing1996 commented 4 years ago

您好,看到你跑出了增强后的结果,想请问您是用的python3吗? 有遇到spectrogram_to_wav.py这个文件里real_to_complex这个函数报错吗? 报错是显示相乘的两项维度不一致,具体可以看我提出的issue#58,谢谢您!

qiuqiangkong commented 4 years ago

您好!请检查下spectrogram的维度

On Sun, 21 Jun 2020 at 23:12, QianYing1996 notifications@github.com wrote:

您好,看到你跑出了增强后的结果,想请问您是用的python3吗? 有遇到spectrogram_to_wav.py这个文件里real_to_complex这个函数报错吗? 报错是显示相乘的两项维度不一致,具体可以看我提出的issue#58,谢谢您!

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/yongxuUSTC/sednn/issues/57#issuecomment-647140888, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADFXTSNQHEF7RZ6SZPLZVYLRXYPN7ANCNFSM4NCWY3CQ .

QianYing1996 commented 4 years ago

您好,谢谢您的回复,请问可以具体说下怎么看spectrogram的维度是否正确吗?是对训练语料有什么要求吗? 我是用mini_data跑的,没有修改源码,就遇到了这个问题。 我是刚入门语音的小白,麻烦您指点一下,谢谢!

qiuqiangkong commented 4 years ago

试着print(spectrogram.shape)

On Mon, 22 Jun 2020 at 10:18, QianYing1996 notifications@github.com wrote:

您好,谢谢您的回复,请问可以具体说下怎么看spectrogram的维度是否正确吗?是对训练语料有什么要求吗? 我是用mini_data跑的,没有修改源码,就遇到了这个问题。 我是刚入门语音的小白,麻烦您指点一下,谢谢!

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/yongxuUSTC/sednn/issues/57#issuecomment-647229352, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADFXTSPZPIZIAHWYUPCWZCTRX25QHANCNFSM4NCWY3CQ .

smylab commented 3 years ago

您好,我想问下为什么增强后的语音发出来的只有吱吱吱的声音呢? 下面是它对应的语谱图,请问下是出了什么问题呢?期待您的解答,谢谢! image

smylab commented 3 years ago

您好,我是刚接触语音增强的小白,本来是打算用这份代码学习的,结果运行了好几次minidata里的数据,代码是原封不动运行的,环境也是按照要求去配置的,,,但增强后的结果除了吱吱吱吱吱吱吱吱吱吱吱,基本的混合语音都听不到了,只有吱吱吱的声音,我快疯了,,这个怎么回事,您最终怎么解决这个问题的,,万分感谢,我快被搞疯了。

yongxuUSTC commented 3 years ago

你把增强的wav和mix wav发给我听听,用mini data 训练数据是不够的,需要更大的训练数据来训练,才有比较好的结果。另外你可以试下SNR高的例子 增强看看效果,验证代码是否正确

On Fri, 23 Apr 2021 at 22:09, smylab @.***> wrote:

您好,我是刚接触语音增强的小白,本来是打算用这份代码学习的,结果运行了好几次minidata里的数据,代码是原封不动运行的,环境也是按照要求去配置的,,,但增强后的结果除了吱吱吱吱吱吱吱吱吱吱吱,基本的混合语音都听不到了,只有吱吱吱的声音,我快疯了,,这个怎么回事,您最终怎么解决这个问题的,,万分感谢,我快被搞疯了。

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/yongxuUSTC/sednn/issues/57#issuecomment-826036104, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABJGHUSIN6A2F6W3K3SABN3TKJHAVANCNFSM4NCWY3CQ .

smylab commented 3 years ago

你把增强的wav和mix wav发给我听听,用mini data 训练数据是不够的,需要更大的训练数据来训练,才有比较好的结果。另外你可以试下SNR高的例子 增强看看效果,验证代码是否正确 On Fri, 23 Apr 2021 at 22:09, smylab @.***> wrote: 您好,我是刚接触语音增强的小白,本来是打算用这份代码学习的,结果运行了好几次minidata里的数据,代码是原封不动运行的,环境也是按照要求去配置的,,,但增强后的结果除了吱吱吱吱吱吱吱吱吱吱吱,基本的混合语音都听不到了,只有吱吱吱的声音,我快疯了,,这个怎么回事,您最终怎么解决这个问题的,,万分感谢,我快被搞疯了。 — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#57 (comment)>, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABJGHUSIN6A2F6W3K3SABN3TKJHAVANCNFSM4NCWY3CQ .

感谢您的回复,谢谢。这是0db和5db的minidata测试结果,按道理说如果因为数据集小,没充分的训练模型,增强后的结果最起码不会比原始的混合语音差才对啊。感谢您给出的建议,这个问题困惑了我好几天,我马上用大数据集再训练测试一下。

sednn_minidata_test.zip

smylab commented 3 years ago

您好,我想问下为什么增强后的语音发出来的只有吱吱吱的声音呢? 下面是它对应的语谱图,请问下是出了什么问题呢?期待您的解答,谢谢! image

你把增强的wav和mix wav发给我听听,用mini data 训练数据是不够的,需要更大的训练数据来训练,才有比较好的结果。另外你可以试下SNR高的例子 增强看看效果,验证代码是否正确 On Fri, 23 Apr 2021 at 22:09, smylab @.***> wrote: 您好,我是刚接触语音增强的小白,本来是打算用这份代码学习的,结果运行了好几次minidata里的数据,代码是原封不动运行的,环境也是按照要求去配置的,,,但增强后的结果除了吱吱吱吱吱吱吱吱吱吱吱,基本的混合语音都听不到了,只有吱吱吱的声音,我快疯了,,这个怎么回事,您最终怎么解决这个问题的,,万分感谢,我快被搞疯了。 — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#57 (comment)>, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABJGHUSIN6A2F6W3K3SABN3TKJHAVANCNFSM4NCWY3CQ .

感动哭,感谢作者对我们初学者的一一回复,大数据集训练就没有吱吱吱吱吱的声音了,增强效果很好,之前一直找不到问题所在,感谢。