We have been developing an objective measure of how similar various consonants and vowels are to each other. In this model, phonetic similarity is based on articulatory data from ultrasound, electroglottograph, and aerodynamic measurement, acoustic measures, and perceptual judgments (SSHRC grant #410-2007-0735, "Measuring the Phonetic Similarity of Speech Sounds").
Here are some visualizations of the similarity data from an article to appear in the upcoming Lingua special issue on phonological similarity.
| Acoustic distance PCA | Vocal tract shape distance PCA |
| Airflow measurements | Laryngeal measurements |
| Neighbour-joining tree for acoustic distance | |
The data from the project these figures are based on is in this file, and some relevant R code is found here. This version of the phonetic distance data has a row for each of 51 segments, and columns for different kinds of phonetic data, and a file with R code that should be useful for getting oriented. You can also open up the csv file in Excel and calculate a distance between segments by subtracting their values for any of the columns. To get an overall acoustic distance between two segments, you can take the square root of the sum of the squares of all 16 acoustic distances. These are things the R code does.
This kind of data underestimates the difference between consonants and vowels, since these segments were all produced in the same contexts. Differences between consonants and vowels can be easily extracted from information about their distribution.
Watch this space for more data. This version has articulatory and acoustic data for 51 segments based on the speech of three people. Soon I expect to be adding data from a fourth speaker, perception data, and data for another 74 segments.