Classification of Speech Data
We have seen in previous chapters that a given acoustic parameter, such as zero-crossing-rate, the first formant frequency, or root-mean-square energy, can be used to differentiate between classes of speech sounds. In many cases, it takes a combination of parameters to separate different classes of speech sounds. For example, vowels of different phonetic quality require at the very least not just the first formant frequency but a combination of the first two formants for their separation; similarly, classes of fricatives such as [s], [∫], and [θ] can be quite effectively separated in a plane of two parameters based on the first two cepstral coefficients, as we saw in Chapter 6.
KeywordsDiscriminant Function Parameter Vector Formant Frequency Decision Boundary Speech Sound
Unable to display preview. Download preview PDF.