Natural-to-Shouted Voice Transformation for Distance Cues of Monosyllabic Consonant-Vowel-Consonant Words
Résumé
This paper deals with voice transformation techniques in the case of natural-to-shouted voice transformation. A database is first recorded and analyzed in order to find relevant cues relative to the speaker's vocal effort, especially concerning the fundamental frequency. Then, a new experiment aiming at transforming natural voices into shouted voices is presented. It is based upon an observed repetitive f0-pattern. The transformation also takes into account intensity variations, spectral tilt variations and duration variations. A perceptual test was carried out in order to evaluate the quality and the perceived sensation of the transformations. The results show that about 80% of the proposed transformed voices are perceived as shouted voices.