Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper trains XTTS v2, F5-TTS, and DiFlow-TTS models on a bilingual (Quechua and Spanish) legal corpus to synthesize high-quality speech for the Peruvian Constitution. Cross-lingual transfer learning is used to address the data scarcity in Quechua, improving synthesis quality in both languages. The authors release trained checkpoints, inference code, and synthesized audio, providing a valuable resource for low-resource TTS in legal and political contexts.
Synthesizing speech for marginalized languages like Quechua is now easier, thanks to a new pipeline that leverages cross-lingual transfer learning to produce high-quality audio from limited data.
We present a unified pipeline for synthesizing high-quality Quechua and Spanish speech for the Peruvian Constitution using three state-of-the-art text-to-speech (TTS) architectures: XTTS v2, F5-TTS, and DiFlow-TTS. Our models are trained on independent Spanish and Quechua speech datasets with heterogeneous sizes and recording conditions, and leverage bilingual and multilingual TTS capabilities to improve synthesis quality in both languages. By exploiting cross-lingual transfer, our framework mitigates data scarcity in Quechua while preserving naturalness in Spanish. We release trained checkpoints, inference code, and synthesized audio for each constitutional article, providing a reusable resource for speech technologies in indigenous and multilingual contexts. This work contributes to the development of inclusive TTS systems for political and legal content in low-resource settings.