A new dual-channel speech enhancement technique with application to CELP coding in noise