search

crouchred / speaker-recognition-py3

Base on MFCC and GMM(基于MFCC和高斯混合模型的语音识别)
Apache License 2.0
245 stars 80 forks source link

i always get 1 for the result #13

Open yitong241 opened 4 years ago

yitong241 commented 4 years ago

hello, thank you very much for your code but there is something I do not understand.

I saved some training .wav file in the following path: "\speaker-recognition-py3-master\tmp\person\1.wav", "\speaker-recognition-py3-master\tmp\person\2.wav", "\speaker-recognition-py3-master\tmp\person\3.wav"

when I tried to predict, this was the output.

Snipaste_2019-08-04_16-43-39

could u help me with the problem? thank you

alonyomtov123 commented 4 years ago

I have the same problem

KeyurGK commented 4 years ago

how to give input to this code in spyder

izhangy commented 4 years ago

请问如何添加音频数据进行训练？

izhangy commented 4 years ago

我也遇到了同样的问题，有人解答一下吗？

RitchieQi commented 4 years ago

Actually, I had this problem too and I fixed it by training all the .wav files in just one command instead of training them one by one. The reason ,which caused this problem, I guess it's because the code will generate a model.out file for every training command. If you train them one by one, the posterior model.out file will replace the prior one. As a consequence, you will always get a score 1.0 for your latest training sample. Hope this could be helpful.

Manav-Motwani commented 4 years ago

Sir what was length of each training file?

Actually, I had this problem too and I fixed it by training all the .wav files in just one command instead of training them one by one. The reason ,which caused this problem, I guess it's because the code will generate a model.out file for every training command. If you train them one by one, the posterior model.out file will replace the prior one. As a consequence, you will always get a score 1.0 for your latest training sample. Hope this could be helpful.

RitchieQi commented 4 years ago

Sir what was length of each training file?

Actually, I had this problem too and I fixed it by training all the .wav files in just one command instead of training them one by one. The reason ,which caused this problem, I guess it's because the code will generate a model.out file for every training command. If you train them one by one, the posterior model.out file will replace the prior one. As a consequence, you will always get a score 1.0 for your latest training sample. Hope this could be helpful.

It up to you. In my case, I recorded three 1-min .wav files for each volunteer.