Language classification using prosodic features: Comparing intensity and pitch

The performance of Human Language Technology systems is known to be severely impacted by language variations. Distance measures have been used in several applications of speech processing to analyze different varying speech attributes. However, not much work has been done on language distance measures, and even less work has been done involving South African languages. Prosodic information-such as intonation and stress which are primarily encoded in pitch and intensity-have been known to lend language-specific variations to human language. We explore the use of intensity and pitch characteristics of languages for measuring the linguistic distance of six South African languages. Both methods are acoustic based and compare accumulative mean differences between pairs of languages to generate difference matrices. It is shown that cluster analysis resulting from the pitch distance matrix correlates closely with human perceptual distances and existing literature about the six languages whereas intensity distances show no distinct patterns of correlation.