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 124 forks source link

测试结果不理想 #52

Open qq274943639 opened 4 years ago

qq274943639 commented 4 years ago

你好我用这个网络在0db情况下进行了训练，训练集能到到0.66的loss，但是测试集一直是0.90，我用的是timi库的训练集和测试集，请问是哪里出了问题吗

qiuqiangkong commented 4 years ago

您好！可以检查下训练样本数量是否正确。

Best wishes,

Qiuqiang

From: qq274943639 notifications@github.com Sent: 23 February 2020 13:46 To: yongxuUSTC/sednn sednn@noreply.github.com Cc: Subscribed subscribed@noreply.github.com Subject: [yongxuUSTC/sednn] 测试结果不理想 (#52)

你好我用这个网络在0db情况下进行了训练，训练集能到到0.66的loss，但是测试集一直是0.90，我用的是timi库的训练集和测试集，请问是哪里出了问题吗

― You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHubhttps://eur02.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2FyongxuUSTC%2Fsednn%2Fissues%2F52%3Femail_source%3Dnotifications%26email_token%3DADFXTSJQ6TMQEDCE3UPW2S3REJ44XA5CNFSM4KZ2W63KYY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4IPR3BGA&data=02%7C01%7Cq.kong%40surrey.ac.uk%7Ca2af47aa1a044093b21208d7b866d6f6%7C6b902693107440aa9e21d89446a2ebb5%7C0%7C0%7C637180624144748532&sdata=bDoQ3%2BFydn4ykX2r1F30i4CKTNbVP8NZs%2B8hpV8cJcg%3D&reserved=0, or unsubscribehttps://eur02.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fnotifications%2Funsubscribe-auth%2FADFXTSLQ3CG2AR3YDKJHF2TREJ44XANCNFSM4KZ2W63A&data=02%7C01%7Cq.kong%40surrey.ac.uk%7Ca2af47aa1a044093b21208d7b866d6f6%7C6b902693107440aa9e21d89446a2ebb5%7C0%7C0%7C637180624144748532&sdata=Xudx6z43kZJN3RkRp1ZdzQUvAG5%2B7wElJwAO82TBimo%3D&reserved=0.

qq274943639 commented 4 years ago

你好，我看了下你们使用的训练集和测试集，训练集Training speech: TIMIT 4620 training sentences. 测试集Testing: Speech: TIMIT 168 testing sentences (selected 10% from 1680 testing sentences)，是不是测试集使用太少了？一般不是按照训练集：测试集3:1的数量设置的吗