A case study on the four spectral moments and the peak intensity of /s/ in a natural voice and a voice cloned by AI Elevenlabs; [Um estudo de caso sobre os quatro momentos espectrais e o pico máximo de intensidade de /s/ em uma voz natural e uma voz clonada pela AI Elevenlabs]

Torres Castillo F.A.; Cueva Sanchez O.E.; Jimenez Peña J.; Shicshi Romero E.A. · Lengua y Sociedad (2024) · DOI: 10.15381/lengsoc.v23i2.29209

Resumen

This research compares the four spectral moments and the peak of greatest intensity of the alveolar fricative [s] in coda position between the natural and artificial voice of a speaker. The data from the natural voice were collected in a controlled environment and subsequently segmented with Praat software. For the artificial voice, ElevenLabs’ voice cloning technology was used. The analysis of variance shows significant differences between the means of the acoustic parameters analyzed—center of gravity, standard deviation, kurtosis, skewness, and the peak of maximum intensity—in the natural and artificial voices. However, when analyzing the data grouped by the vowel preceding the fricative, it is observed that not all contexts show significant differences. The results indicate that the spectral parameters of the alveolar fricative /s/—especially the center of gravity—are useful to distinguish between the natural voice and its artificial counterpart. © Los autores.

Instituciones: Universidad Nacional Mayor de San Marcos