Segmentation-based noise suppression for speech coders using auxiliary sensors

Despite the significant progress, low perceptual quality of encoded noisy speech is still an unsolved problem. The quality problem at noisy environments is addressed for MELP speech encoder by using a novel speech enhancement algorithm at the front-end. The speech signal is segmented into broad phonetic classes using auxiliary sensors in addition to the acoustic microphone. Each phoneme class is enhanced by suppressing maximum noise while minimally distorting perceptually important cues using the acoustic-phonetic knowledge about the class. The A/B quality test shows significant improvement over the MELPe coder that uses MMSE algorithm for enhancement.