Tandem Interconnections of LPC and CVSD Digital Speech Coders.

摘要

The study described in this report was concerned with the limitations on the performance of tandem interconnections of LPC and CVSD digital speech coders. Such interconnections were systematically studied so as to identify sources of speech quality degradation and schemes for improving the performance of such tandem interconnections were investigated. In the LPC-to-CVSD connection, the major source of speech quality degradation appears to be the minimum phase nature of the output waveform of conventional LPC synthesizers. The resulting 'peaky' waveform causes increased slope overload distortion in the CVSD coder, thereby degrading the overall performance. To alleviate this problem, a flexible approach to modifying the phase of LPC synthetic speech was developed. Both objective and subjective tests of the phase modification system show significant but not dramatic improvements in quality for the overall LPC-to-CVSD tandem. In the CVSD-to-LPC connection, the major source of degradation is the quantization error introduced by the CVSD coder. This quantization noise distorts the spectrum estimate obtained in the LPC analysis by broadening the bandwidths of the formant resonances. To improve the LPC spectrum estimate, the LPC analysis was modified as follows: Individual pitch periods in voiced frames were located and averaged to reduce the noise relative to the signal.