Distortion measures for speech processing

摘要: Several properties, interrelations, and interpretations are developed for various speech spectral distortion measures. The principle results 1) the development of notions relative strength equivalence measures both in a mathematical sense corresponding to subjective coding when used minimum or nearest neighbor processing systems; 2) demonstration that Itakura-Saito related possess property similar triangle inequality systems such as quantization cluster analysis; 3) normalized model yield efficient computation algorithms generalized centroids points groups clusters frames, an important classical analysis techniques optimal quantizer design. We also argue distortions well-suited computationally, mathematically, intuitively applications.