A frequency domain waveform speech compression system based on product vector quantizers

The discrete short-time Fourier transform (DS TFT) has been widely used to study and analyze several speech signal characteristics. However, this Scheme has not been used as successfully in speech compression based on scalar quantization. On the other hand, most speech perception concepts have very interesting frequency-domain interpretations, which suggest the design of compression schemes based on frequency domain analysis. In this paper we apply vector quantization techniques for designing and simulating discrete short-time Fourier transform based speech compression systems. The main conclusion we can draw is that a low to medium rate compressor, in which the reproduction signal has at least communications quality, and in which the complexity is moderate, is at least in principle possible.