search

rajathkmp / speaker-verification

Implementation of state of the art d-vector approach for speaker verification
127 stars 32 forks source link

What's the meaning of queryWS #1

Open YichiHuang opened 6 years ago

YichiHuang commented 6 years ago

Hi, I saw the code in spk_dnn.py and I don't have idea about the function def queryWS(nameKW) , which is defined as follows: def queryWS(nameKW): a = { 'government': 75, 'company': 71, 'hundred': 59, 'nineteen': 79, 'thousand': 77, 'morning': 69, 'business': 81 } return a[nameKW] Could you please give some comments about this? Thanks.

rajathkmp commented 6 years ago

@YichiHuang The data is from Librispeech, and those are the keywords that I trained my models on, this code is for Text Dependent Speaker Verification

YichiHuang commented 6 years ago

Thanks for your reply. It seems that you used the data that only contains Keyword from Librispeech. As we know, each audio of Librispeech is long duration and contains dozens of words. So, you cut the long duration audio into a smaller one that only contains one Keyword, right?

rajathkmp commented 6 years ago

@YichiHuang Yes, used Kaldi ASR to do so

meixitu commented 6 years ago

@rajathkmp , Thanks for you very good project. Could you share the step to use the Kaldi ASR to split the keyword from setence? I don't have any idea, it is not friendly to use

Thanks Jinhong

meixitu commented 6 years ago

@YichiHuang , could you share the step to use the Kaldi ASR to split the keyword from Librispeech? Now I downloaded Librispeech and latest Kaldi, But Kaldi has many code, I don't know how to use it.

Thanks Jinhong