Design and performance of a 4.0 kbit/s speech coder based on frequency-domain interpolation

Abstract

The 4.0 kbit/s speech codec described is based on a frequency domain interpolative (FDI) coding technique, which belongs to the class of prototype waveform interpolation (PWI) coding techniques. The codec also has an integrated voice activity detector (VAD) and a noise reduction capability. The input signal is subjected to LPC analysis and the prediction residual is separated into a slowly evolving waveform (SEW) and a rapidly evolving waveform (REW) component. The SEW magnitude component is quantized using a hierarchical predictive vector quantization approach. The REW magnitude is quantized using a gain and a sub-band based shape. The SEW and REW phases are derived at the decoder using a phase model, based on a transmitted measure of voice periodicity. The spectral (LSP) parameters are quantized using a combination of scalar and vector quantizers. The 4.0 kbits/s coder has an algorithmic delay of 60 ms and an estimated floating point complexity of 21.5 MIPS. The performance of this coder has been evaluated using in-house MOS tests under various conditions such as background noise, channel errors, self-tandem, and DTX mode of operation, and has been shown to be statistically equivalent to ITU-T G.729 8 kbps codec across all conditions tested.