An Attempt to Persuade the IBM 704 Computer to Identify Vowel Types

In a classical study of the vowel sounds of English, G. E. Peterson and H. L. Barney collected a large body of experimental data which related perceived vowel quality to measurements of the first three formant frequencies and the voice pitch. If these data are used to make scatter diagrams of vowel types plotted on coordinates representing the physical measurements, it is seen that a particular vowel tends to cluster in a region whose boundary has a complicated shape. The complicated form of these vowel regions makes it difficult for a computer to use the raw data to interpret vowel quality. This paper describes a coordinate transformation which simplifies the boundaries of the vowel regions into simple rectangular blocks. The technique used to find the transformation involved programming the computer to plot scatter diagrams directly on the printer. These snapshots then were used by a human observer in a guided search for a simple transformation. The coordinate system found has been used in speech recognition studies.