Abstract
Working in the context of residual speech coders, the proposed technique performs a decomposition of the DFT (discrete Fourier transform) of the target vector (the input speech after the subtraction of the zero input response signal) as the product of the DFT of the impulse response of the LPC (linear predictive coding) synthesis filter and a modified residual signal. At the receiver, the DFT of the target vector is reconstructed by multiplying the quantized modified residual signal by the DFT of the impulse response of the synthesis filter. The inverse DFT is then calculated, and the resulting time-domain quantized target vector is filtered by the inverse synthesis filter in order to get the excitation to the LPC synthesis filter. Magnitude-phase and real-part-imaginary-part encoding of the modified residual signal has been studied, and several quantization schemes have been used. Specifically, scalar quantization and trellis coded quantization have been used in the development of 32-kbit/s coders with a very high quality of the synthesized speech. A 24-kbit/s coder has also been developed by using vector quantization and trellis-coded quantization in a magnitude-phase representation of the modified residual. >
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have