Continuous Speech Processing



 I started working on MATLAB mainly in my final year when we had entirely new subject introduced for first time in semester i.e continuous speech processing...

We tried lot of things starting with Rabinar & Shaffer  speech recognition  ebook , Sadoaki Furui 's speech synthesis & lot of information from web. It is broadly divided as

  • Speaker recognition
  • Speech recognition
  • Language Recognition 

Our aim at semester end was  speaker independant voice recognition. We stored database of various students speaking one to ten. We applied Dynamic Time Warping based analysis with comparison of Mel Frequency Based Cepstral Coefficients (MFCC)...

We can convert speech of one second into various frames whrer we can assume it to be stationary for short interval of say 32 ms with overlap of 20ms . Frame analysis can convert 256 samples frame into just 13 dimensional cepstral coefficients. Thus 8000 samples with 8KHz sampling Frequnecy for 1 second gets converted into 79 frames of 13 vectors each.(256 samples & overlap of 156 samples.)

Here is list of complete software codes in matlab:- 

 

  • Spectrogram of Recorded Speech 
  •  Short Time Energy Calculation
  • Feature Analysis and Extraction for Audio Automatic  Classification (IEEE Paper)
  • Computation of Pitch Frequency Using Cepstrum Analysis on Quefreqency axis 
  • Mel frequency cepstrum coefficient analysis.
  • HMM Based Maximum Likelihood Probability
  • HMM Based Maximum Likelihood Probability with Forward Algorithm
  • Isolated Word Recognition DTW Based (Not Reliable as we didnt implemented K-means clustering algorithm. )

Here is complete soft copy in pdf Format :-

MatlabPrograms.pdf

complete set of programs with G711 paper & IEEE paper :-

CSPFINAL.rar 

Some Good Links for Voicebox tool & HMM toolbox :-

Matlab Introduction to Speech Processing 

DTW introduction 

HMM Toolbox 

Voicebox Toolbox 

 

Thanks to Hrishikesh sabale as we worked together on all programs & he completed feature    analysis code to differentiate music & speech...IEEE paper

             Feature Analysis and Extraction for Audio Automatic Classification  
                Bai Liang, Hu Yanli, Lao Songyang, Chen Jianyun, Wu Lingda


Thanks to Prof. S.P. Mahajan for his Guidance & help...