New encoding algorithm for distributed speech recognition based on DTFS transform

Abstract

The paper presents a new algorithm for efficient compression of front-end feature extracted parameters used in distributed speech recognition systems (DSR). In the proposed method the source encoder is mainly based on discrete time Fourier series (DTFS) by interpolation using Fourier coefficients with conventional vector quantization. The system provides a compression bit rate as low as 4 kbps; the experiments were carried out on the TIDigits Aurora2 database [1]. The simulation results show good recognition performance without dramatic change comparing with ETSI STQ-AURORA standard front-end feature compression algorithm with quantized features at 4.4 kbps [2]. © 2012 Springer-Verlag.