ADPCM compression of speech signals using a frequency domain LMS algorithm

This work introduces the use of an improved version of the popular stochastic gradient LMS algorithm, the frequency domain LMS algorithm, for the predictor adaptation, in ADPCM speech compression systems. Tests made indicate that the FLMS version, which is shown to converge at a faster rate than the traditional LMS algorithm, does not improve ADPCM performance, thus demanding a more careful interpretation of the predictors operation. This leads to a series of conclusions involving the speech generation model that extend the understanding of the ADPCM speech coding schemes and unveils new strategies to improve these systems