Science & Technology20 October 20261 min read
BHASHINI and IISc Bengaluru develop text-to-speech models for Sourashtra
The Digital India BHASHINI Division under the Ministry of Electronics and Information Technology (MeitY) has developed artificial intelligence-based text-to-speech (TTS) models for Sourashtra, a minority Indo-Aryan language spoken mainly in Tamil Nadu. The work was done with the Indian Institute of Science (IISc) Bengaluru.
The models turn written Sourashtra text into spoken audio, bringing speech technology to a language that has had little of it.
The language and the models
Sourashtra is spoken predominantly by the Sourashtra community in Tamil Nadu. According to Census 2011, it has about 2.48 lakh speakers.
The models use the VITS architecture (Variational Inference with adversarial learning for end-to-end TTS), which converts written text directly into speech. They support male and female voices and can process text written in either the Tamil script or the Sourashtra script.
The data resource
BHASHINI's AIKOSH platform hosts the SPICOR TTS 3.0 Sourashtra Corpus, which is meant to support AI research in Indian languages.
Intended uses
The work supports the use of low-resource languages in digital applications, language learning, accessibility solutions, documentation and research.