Development of a Configuration Concept of a Speech Digitizer Based on Adaptive Estimation Techniques.
SOUTHERN METHODIST UNIV DALLAS TEX INFORMATION AND CONTROL SCIENCES CENTER
Pagination or Media Count:
The report describes the results of a fifteen month contract to study the feasibility of using sequential estimation techniques for speech digitization. The original goals of this contract were to analyze several sequential estimation algorithms for possible use in speech digitization and then to select an algorithm for detailed study. In addition, a new residual encoding structure was developed as a by-product of this study and it has been given preliminary investigation for concept feasibility. All of the stated goals of this study have been met and it has been shown that the sequential linear Kalman filter algorithm can Yield high quality synthetic speech in a pitch synchronous mode Operate in real time with the use of realistic parallel hardware Be used in a residual encoding structure with good results without the use of pitch information. The conclusion is that the Kalman filter algorithm offers a meaningful alternative approach to the existing batch process algorithms. The output speech produced during this study compares very favorably with the speech obtained by other methods. Author
- Non-Radio Communications