Improved Speech Recognition Using Perceptual Linear Prediction

摘要

In this study, two speech analysis methods Perceptual Linear Prediction (PLP) and Linear Prediction (LPC) are compared. For PLP the spectral scale is the non-linear Bark scale and the spectral features are smoothed within frequency bands. In contrast for LPC the spectral scale is linear and no smoothing is done. Another approach is from the computational requirements: the implementation of both methods on Motorola StarCore SC140 DSP is presented, and the number of cycles for optimized and non-optimized code is then compared.