I'm working in a speech recognition project for the university. I'm using the TIMIT dataset and pylearn2 as machine learning library. The project need to be simple, and it doesn't have to follow the state of art at all costs.

I've decided to use the RASTA-PLP technique to extract the features. More precisely I'm using the rastaplp.m function in Matlab. I want a sample for each phoneme, but when i execute the RASTA-PLP function on a single phoneme, i have multiple values for each phoneme. This is normal, considering that the RASTA-PLP function calculate his value over a certain number of temporal windows, and each phoneme has a certain lenght. But i need for each phoneme, a single value. Actually, after having chosen the number of feature (14 in my case), i need an array (1X14), composed of one line and 14 cells.

How can a transform the result to that ? Is it ok just to calculate the average value over the different values of each feature ?

  • $\begingroup$ You need to describe better what do you want from this "single value". In many cases average is ok, but your requirements are not clear. $\endgroup$ – Nikolay Shmyrev Aug 20 '15 at 20:12
  • $\begingroup$ A note that may help: For MFCCs, the transformation is applied to a window of the signal. If you know the signal contains a single phoneme, fitting the window length to that signal length should give an (1xC) set of coefficients (C is the number of coefficients). I am about to use PLP, and I expect to be able to do something similar. $\endgroup$ – Eric Platon May 12 '16 at 1:05

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.