Identification of Costa Rican speaker’s age group based on acoustic vowel analysis

Abstract

The identification of the speaker’s age is a fundamental task in human-computer interaction. In particular, the recognition of children has drawn attention from the research community due to the vulnerability of se persons on the Internet, as well as the potential benefits in digital educational systems. The identification of the age can be determined from the dynamics of the speech information, which has a dependency on language and accent. This paper presents a study on the identification of Costa Rican adults and children based on acoustic vowel analysis. For this purpose, we chose the five basic vowels of the Castilian Spanish language, and extract a set of acoustic features from segmented speech to explore the accuracy of three classifiers (k-NN, Random Forest and SVM) in distinguishing between adults and children. The results show promising results in the classification using a single vowel, with the best results from the vowel /u/. This means that an automatic system could be built to identify age as long as enough speech information is received and transcribed, but the accuracy of such a system based on vowels should not be expected to be very accurate in short interactions.