Open Sara-Sa-Uk opened 4 years ago
Hello,
For your first question: as the trace-back shows, the problem seems to be a permission error, therefore try to run the code as an admin aka using sudo
. Also make sure that you have installed ffmpeg.
As for the second question: the silence elimination part is needed so that the models are trained only on speech data and not speech+silence data. Eliminating the silence should help speed the system and improve its precision.
Thank you for your help, sorry I have one more question regarding SVM-Super vectors it's based on GMM-UBM right? dose it based on HMM, why we are saving the models as .hmm not .gmm? Thank you again.
you are welcome @Sara-Sa-Uk . You don't have to thank me in every message, this is a place to share knowledge so I am here to help and learn too. As for your questions:
According to point 2, you mentioned that your SVM code is based on GMM Super vectors, but the code seems to be based on HMM Hidden Markov Model, I have changed every thing from HMM to GMM and it works fine so is what I did correct?
I think this is a a mistake from my side, I was convinced that I used GMM with SVM but it seems that I am using HMM. I think either one should work fine but yes changing the HMM call to GMM should be it, so you did well :clap: . I will review the code soon and publish the fix for this along with some improvements I have been working on. Thank you for raising this point :)
To Help you understand super-vectors and GMM better; here what the system looks like when modeled(Think of the UBM here as a GMM, after all a UBM is just a one big GMM.): source: The present and future of voiceprint based security.pdf
Hello Mr.Ayoub, Thank you for your great work.
I got some issues while running the SVM-code could you help me please.
I have one question regarding the SVM-code, why we are training the data on .hmm.
And why we you used this?
Thank you in advance.