A first approach to Acoustic Characterization of Costa Rican Children’s Speech

Abstract

As human interaction with computers becomes more pervasive, the value of developing automatic speech recognition, text-to-speech synthesis, and related speech technologies become more important for people of all ages, accents, and conditions.

One of the groups that represent bigger challenges is children, due to the difficulties in recording enough speech, and the lack of characterization of their speech, which is particular of every language and accent. This paper presents the first approach to acoustic analyses of Costa Rican children aged from six to twelve years. These analyses aimed to achieve a better understanding of the characteristics of speech produced by this group, in terms of providing future development and enhancement of automatic speech recognizers and speaker identification systems.

For this purpose, we record the speech consisting of isolated words of three children, and compare the results with three adults, in terms of the vowel’s formants. The formants give information about the vocal track of the speaker, and it is an important method to provide the first analysis of these signals. Results show noticeable differences between the children and adults and may provide useful information about future trends to adapt and develop the current speech technologies for this population.