Rule Based Sinusoidal Encoding of Speech
AIR FORCE INST OF TECH WRIGHT-PATTERSON AFB OH SCHOOL OF ENGINEERING
Pagination or Media Count:
A system was developed to investigate the data rate necessary to transmit speech using a rule based sinusoidal model. The system consists of a speech analyzer and a synthesizer. The analyzer outputs discrete frequencies and quantized amplitudes and phases of selected speech spectral components. The synthesizer reconstructs speech from these components based on a sinusoidal model. The selection of spectral components for voiced speech regions is based on the detection of harmonics of the fundamental frequency. To obtain a specific number of spectral components, a variable amplitude threshold is applied to the detected harmonics and their nearest neighbors. For unvoiced regions only the variable amplitude step is applied. The lowest data rate obtained for toll quality speech was about 18 Kbps. This system was implemented in Fortran 77 on a VAX 11780 computer. Visual analysis of speech was provided by the software package SPIRE Speech and Phonetics Interactive Research Environment. Keywords Theses.
- Voice Communications