VARIABLE RATE DIGITAL SPEECH WAVEFORM CODING
Abstract
A new variable rate speech waveform coding algorithm capable of operating on commonly used fixed bit rate modems at either 9.6 or 16 kilobits per second is presented. The algorithm uses a voiced/unvoiced/silence classifier to select one of three possible coding schemes. Half cycle sinusoid replacement is used for voiced speech while a form of one bit quantization is used for unvoiced speech. Intervals of silence are replaced with zero amplitude sinusoids. Much research has been done in the areas of low and medium bit rate speech waveform coding. A trade-off exists among coder complexity, bit rate, and quality of decoded speech. Continuously Variable Slope Delta (CVSD) modulation is used as a benchmark algorithm in this research as it appears to be the best of the simple algorithms operating at the above bit rates. Results of this research indicate a substantial listener preference, based on speech quality tests, for the new variable rate algorithm over CVSD. The new algorithm is preferred 2 to 1 over CVSD at 16 kilobits per second and over 8 to 1 over CVSD at 9.6 kilobits per second. Intelligibility test scores indicate little difference in the two algorithms; both maintain a high degree of decoded speech intelligibility. Intelligibility test scores ranging from 85 to 95 percent intelligible at 16 kilobits per second and from 75 to 85 percent intelligible at 9.6 kilobits per second can be expected with this new algorithm. Coder complexity based on the number of mathematical operations required per output speech sample is similar for the two algorithms. However, more data storage and decision making capability is required by the new variable rate algorithm suggesting a micro-computer based implementation.
Degree
Ph.D.
Subject Area
Electrical engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.