Vowel Lab II: Vowel Features

In Lab #7, we looked at designing low-order digital filters to model human vowel production. Why is this a useful technique? We argued that such models were useful for studying speech production and the like. However, there are more practical, applied reasons for doing this. Suppose that we want to perform speech recognition and determine which vowel is being said. A system would be hard pressed to make this determination from the time-domain waveform. Instead, we need some other description of the signal, some set of features that describe important aspects of the signal. Digital filter modeling, such as what we did in Lab #7, is one way of extracting features. In this lab, we will investigate a number of types of vowel features and explore what makes a set of “good” features for classification. In the next lab (Lab #9), we will design and test a classifier for vowels.