QoriTalk: An Adapted Neural Text-to-Speech System Tailored for Peruvian Spanish

Párraga J.A.F.; Arroyo P.A.F. · Proceedings of the 2025 IEEE 32nd International Conference on Electronics, Electrical Engineering and Computing, INTERCON 2025 (2025) · DOI: 10.1109/intercon67304.2025.11244657

Resumen

Most existing Text-to-Speech Systems (TTS) targeting the Spanish language are based on neutral Latin American or Iberian data, often overlooking the importance and variety of accents across the Spanish-speaking countries. This work presents QoriTalk as an initial system-level implementation of a voice synthesizer tailored to generate natural-sounding and culturally resonant speech with unique Peruvian accents and prosody. Built upon the Spanish-adapted version of FS-TTS - an open-source, MIT-licensed model - QoriTalk was trained with publicly available datasets containing several hours of high-quality recordings from native Peruvian speakers. While no formal perceptual evaluation has yet been conducted, this preliminary contribution demonstrates a promising path toward accessible, culturally adapted TTS systems for applications in education, assistive tools and healthcare interfaces.

Instituciones: Universidad Peruana Cayetano Heredia