Efficient Coding of the Prediction Residual.
Abstract
This thesis presents an efficient method of coding the prediction residual using the technique of sub-band coding designed at the bit rate of 9600 bits/second. The energy of the prediction residual is used to distribute the bit allocation by sub-bands such that the perceptual criteria is preserved. The perception is enhanced by transitional information within the phoneme connections of speech by a technique that weights the energy based on a normalization factor. A three-tier phoneme classification is derived from an energy study of the phonemes for the prediction residual. With this it is shown that speech intelligibility is enhanced in the coding scheme. The prediction residual is compared with the glottal waveform. In association with these results, a new technique for pitch extraction is presented using the prediction as the input signal to calculate pitch. An adequate indication of coder quality is described using various types of signal-to-noise ratios. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 27, 1979
- Accession Number
- ADA092198
Entities
People
- Legand L. Burge Jr
Organizations
- Air Force Institute of Technology