Wideband speech coding in 7.2 kbit/s

Abstract

A novel method for coding wideband speech at medium bit rates is proposed. A subband approach is adopted, using sharp cut-off filters to split the speech into lower and upper bands. This allows the upper-band to be very coarsely quantized without introducing significant distortion at low frequencies in the reconstructed speech, which would normally occur using short QMF (quadrature mirror filter). A standard CELP (code excited linear prediction) coder is used for the low-frequency band, and the upper band is quantized using a second-order predictor and gain shape vector quantization at 0.05 bit sample, yielding an overall bit rate of 7.2 kbit/s. The overall effect is to produce a wideband coder that has almost the same computational requirements as a narrowband CELP coder and only a slight increase in the bit rate.<>