Harmonic Analysis and Sinusoidal Speech Coding.

This memorandum describes a new technique to perform speech coding in which voiced frames are modeled as a sum of sinusoids that depict the harmonic structure of the original speech segment. Unvoiced frames are synthesized using a conventional linear predictive coding filter with a noise or multipulse excitation. Improved synthetic speech quality has been obtained from this system. A real-time speech coder based upon this method and operating at a data rate of 4800 bits/second has been implemented in hardware to be used in the U.S. Government's Future Secure Voice System being developed by AT&T.