IISc Develops Text-To-Speech Models, Speech Corpus For Sourashtra language

Published at:

The Indian Institute of Science (IISc), in collaboration with MeitY's Digital India BHASHINI Division, has developed AI-based text-to-speech (TTS) models and an open-access speech corpus for the low-resource Sourashtra language, supporting male and female voices across both Sourashtra and Tamil scripts

The Indian Institute of Science (IISc)
Indian Institute of Science (IISc) in Bangalore
Summary of this article
  • IISc Bengaluru and MeitY's Digital India BHASHINI Division have developed AI text-to-speech (TTS) models for the low-resource Sourashtra language.

  • Built on the end-to-end VITS architecture, the models generate male and female voices and support text input in both Sourashtra and Tamil scripts.

  • Trained on IISc SPIRE Lab's 28-plus-hour SPICOR TTS 3.0 Sourashtra Corpus, the dataset is now openly available on BHASHINI's AIKOSH platform.

The IISc and the Digital India BHASHINI Division under the MeitY have developed and released openly available text-to-speech models and a speech corpus for Sourashtra, an Indo-Aryan language spoken predominantly by the Sourashtra community.

The release is the first of its kind for the language and marks an important step towards building digital speech resources for languages with limited representation in India’s AI ecosystem, the Indian Institute of Science said in a statement.

Developed at IISc’s SPIRE Lab, the resource comprises four TTS models built on over 28 hours of gender-balanced speech recordings.

The models convert written Sourashtra text into speech, with male and female voices in both Tamil and Sourashtra scripts.

The underlying SPICOR TTS 3.0 Sourashtra Corpus has been made openly accessible to researchers, developers and institutions on AIKOSH, India’s national platform for artificial intelligence datasets and models, the statement said.

Spanning domains such as finance, politics, food, arts, literature, weather and science, the corpus has sentence-level data for developing and evaluating speech technologies.

This provides a diverse range of contexts and vocabulary for developing and evaluating speech technologies.

Prasanta Kumar Ghosh, Professor, IISc, said, “AI can play a significant role in making digital content more accessible to communities whose languages have limited representation in existing technologies. With these Sourashtra text-to-speech models, we are taking a step towards enabling written content in the language to be converted into natural-sounding speech across two scripts.”

Sourashtra is spoken by about 2.48 lakh people, according to the 2011 Census, IISc said.

Read all the latest breaking news on Outlook India and stay updated with top stories from India, Entertainment, Education, and around the world.

×