Title: Distributed Speech Recognition Using Codec Parameters

Authors: Bhiksha Raj, Joshua Migdal, Rita Singh

Abstract: Communication devices that perform distributed speech recognition currently transmit coded parameters of speech signals. Recognition features are extracted from signals that are decoded on remote servers. As reconstruction losses degrade recognition accuracy, proposals are being considered to standardize DSR-codecs which derive recognition features, to be transmitted and used directly for recognition. Such codecs must be embedded on the transmitting device alongwith its standard codec. Performing recognition using codec bitstreams avoids these complications: no additional feature-extraction mechanism is required on the device and there are no reconstruction losses. In this paper we propose an LDA-based method for extracting optimal feature sets from codec bitstreams and show that features so derived result in improved recognition accuracies for the LPC, GSM and CELP codecs. For GSM and CELP we show that the performance is comparable to that with uncoded speech and DSR-codec features.

a01br139.ps a01br139.pdf