zzw922cn

@jeffxusichao

@kmrambo Ok, I will implement your demand soon.

you should change "/home/pony/github" to your own path, because that's my hard coding for path. I am sorry, I will improve this soon.

@Ostnie @cdcfyjh To run the program, you should make sure the mfcc files and label files are in mfcc/ and label/ folders, respectively.

@Ostnie Sorry, the code is based on TensorFlow r1.1, and i will keep updates with latest TensorFlow.

@Ostnie you can also try anaconda.

@njuzrs You should change the feature size parameter, since your dataset's feature vector length is 39, but the code is 60. So you should change 60 to 39.

@njuzrs probably several tens epochs, if set lr to be 0.0001

hi, it is because that a 39 vector is feature of a time window, an audio is composed of many time windows, so the geature should be a matrix, first...

是的，特征矩阵大小就是windows*39。 2017年11月19日 下午9:31，"Ostnie" 写道： > 噢，那意思是说我之前对特征向量的理解一直有错误，不是一整段语音对应一个特征向量，而是每一个窗口截下来的语音数据就是一个长度为39的特征向量， > 这就是说在后面的处理时这整个39*n的矩阵才代表这段语音的完整特征向量，不需要转化为1*39的形式？我的理解对吗？ > > — > You are receiving this because you commented. > Reply to this email directly, view it on GitHub >...