Female voice age classification based on acoustic characteristics of speech.
Eun-Min Lim, Jeung-Yoon Choi · The Journal of the Acoustical Society of America · 2011
As there are many changes in vocal apparatus caused by aging, many acoustic differences also can be found in speech. These differences show different tendencies for male and female speakers and are more pronounced for females. Therefore, this paper focuses on female speaker age analysis and classification. To analyze acoustic differences, global measurements related to the vocal source such as band energy, jitter, shimmer, open quotient, and spectral tilt are used. Additionally, local features that are extracted near vowel landmark regions are also used. Appropriate age boundaries are selected based on the results of analysis, and feature selection is carried out according to the results of analysis of variance (ANOVA) test for classification. The proposed age classification system is modeled using Gaussian mixture models (GMMs), and the sequential feature selection method is used. Experiments are performed on the aGender corpus provided in the INTERSPEECH 2010 Paralinguistic Challenge.