Speaker normalization for automatic speech recognition — An on-line approach

We propose a method to transform the on line speech signal so as to comply with the specifications of an HMM-based automatic speech recognizer. The spectrum of the input signal undergoes a vocal tract length (VTL) normalization based on differences of the average third formant F₃. The high frequency gap which is generated after scaling is estimated by means of an extrapolation scheme. Mel scale cepstral coefficients (MFCC) are used along with delta and delta²-cepstra as well as delta and delta² energy. The method has been tested on the TI digits database which contains adult and kids speech providing substantial gains with respect to non normalized speech.