Speech compression and synthesis

This report describes our work for the past year on speech compression and synthesis. A real-time variable-frame-rate LPC vocoder operating was implemented at an average rate of 2000 bits/s. Mixed-source model as part of the vocoder was also tested. To improve the reliability of the extraction of LPC parameters, a range of adaptive lattice and autocorrelation algorithms was implemented and tested. For data rates above 5000 bits/s, a new high-frequency regeneration techinque, spectral duplication, was developed and tested which reduces the roughness in the synthetic speech. As part of an overall very-low-rate speech compression system, a program was designed and implemented for phonetic synthesis from diphone templates. The lengthy process of extracting the 3000 needed diphone templates has been half completed.