The Use of Formants’ Correlation in Assessing the Sadness State of the Speakers

Abstract

The purpose of this study is to further the comparative analysis of the variations in the correlation coefficients of the formants for the Romanian vowels during emotional speech. We created an annotated speech database composed of recordings of speakers pronouncing the following sentences: /vine mama/ (/mother is coming/), /aseară/ (/last night/), /cine a făcut asta/ (/who did this/), first on a neutral tone of voice, and then expressing sadness. The analysis focuses on the influence that sadness has on the vocal signal. The formants and F0 (pitch) of each vowel were extracted. Statistical analysis techniques were applied in order to verify whether the variation of the correlation coefficients between F0 and F1-F4 presents significant variations or whether it tends to be homogenous under the following conditions: (1) same speaker – same vowel – different sentences; (2) same speaker – same vowel – emotional neutrality vs. sadness; (3) different speakers – same sentence – same vowel.