Abstract: Possible applications of quadratic classifier with heuristically decision threshold in robust AR speech analysis and speech compression schemes are considered. A robust recursive procedure based on WRLS algorithm with VFF and the frame-based quadratic classifier with heuristically decision threshold is proposed for identification of nonstationary AR model of speech. Additionally, possible application of the proposed procedure in standard CELP 4800 b/s speech compression algorithm is evaluated. Experimental analysis is done based on the results obtained through analyzing speech signal with voiced and mixed excitation frames. Obtained results show that the proposed robust recursive procedure achieves more accurate AR speech parameter estimation, provides improved tracking performance, and could be used to solve some speech compression tasks more efficiently.
Key words: Speech analysis, quadratic calssifier, linear predicion coding, least squares algorithm.