Academic Journals Database
Disseminating quality controlled scientific knowledge

Fundamental Frequency Estimation in Speech Signals using Minimum Deviation Method

ADD TO MY LIST
 
Author(s): N. Siddiah

Journal: Journal of Current Engineering Research
ISSN 2250-2637

Volume: 2;
Issue: 1;
Date: 2012;
Original page

Keywords: Fundamental | Frequency |

ABSTRACT
The pitch extraction problem is of importance because the information about pitch is important in several applications like voiced/unvoiced classification, speaker recognition, and speech enhancement. The voiced regions of speech are clearly cyclic, while the unvoiced regions are much more noise-like. The current pitch extraction algorithms work pretty well for clean speech cases. If this clean speech is corrupted by background noise, then these pitch extraction algorithms give a degraded performance. The crux of this project is to develop certain algorithms that perform well for speech effected by background noise as compared to the existing algorithms. The Simplified Inverse Filtering Technique (SIFT) Algorithm uses autocorrelation of the LP (Linear predictive) Residual of the speech signal to find the pitch value. The Cepstrum Algorithm uses the frequency domain to find the pitch of the speech signal. In another standard method autocorrelation analysis is performed on the Hilbert envelope of the LP residual. Although these algorithms work well for clean speech, the performance degrades significantly for noisy speech signal. Three methods are proposed that take evidences from the above three methods and use them together to predict the pitch of the noisy speech signal. The proposed algorithms give a superior performance for both clean as well as noisy speech cases. In this project it is demonstrated that using the evidences from all these three algorithms it is indeed possible to improve the performance compared to the performance of each of them individually.
Affiliate Program      Why do you need a reservation system?