Principal component amplitude compression
Diane K. Bustamante, Louis D. Braida · The Journal of the Acoustical Society of America · 1983
Multiband amplitude compression systems which process each frequency band independently of the others have yet to prove advantageous for sensorineural hearing impaired listeners over good linear amplification systems. Although audibility is improved, certain speech cues are consistently degraded by multiband compression. Acoustic analysis indicates that multiband systems whiten the short-term speech spectrum, reducing the spectral detail. Analysis of speech reception errors suggests that this degrades speech intelligibility. Spectral detail can be preserved by utilizing the correlation of speech levels in different frequency bands. Rather than treating each band independently, the compression in each channel is a function of the levels in all the channels. One implementation of such a system is based on a principal component analysis of short-term speech spectra. Compression of the first component, which corresponds roughly to overall level, equalizes the energy in consonants and vowels much as does a wideband compressor. Compression of the second component, which corresponds roughly to spectral tilt, acts as a variable frequency gain characteristic. The principal component system implementation and measurements of the properties of processed speech materials will be presented.