Home SEEE Student Project Page
Speech Enhancement

Speech enhancement is the process of obtaining clear speech from the noisy speech. The noise coming from the surrounding background will distort the clear speech signal. If the noise is large enough then the speech signal will be distorted and will not be recognizable. So to make the speech recognizable we reduce the estimated noise from the noisy signal and will thus obtain a clear speech signal. For the process, noise is estimated from the non-voice signal or the pause and then subtracted from the noisy signal. The speech will be divided into number of small frames and then Fourier transform will be taken. The noise is estimated from the minimum noise power estimate, which is obtained by frame-wise comparisons of the actual smoothed signal power estimate. The estimated noise is then subtracted from the noisy speech signal. The pause is detected using the voice activity detector; the entropy of voice signal will be large compared to the unvoiced signal. Thus the calculated minimum noise power will be subtracted from the speech signal. Hence clear signal is obtained by spectral subtraction process.


  •     Isha Shrestha
  •     Sujan Sunder Shrestha
 

Course of Study

Course of Study

Project Page

Project Page

Feedback

Feedback

Click HERE to give feedbacks and comments on the site!

Encipher

Encipher 2009Download the technical journal, Encipher, published annually by SEEE.