10
India’s efforts to bring underrepresented languages into the digital ecosystem have advanced with the development of four artificial
intelligence-powered text-to-speech models for Sourashtra, a language spoken by a community concentrated in Tamil Nadu.
The models were developed by the Digital India BHASHINI Division in collaboration with the Indian Institute of Science (IISc), Bengaluru. The initiative aims to make Sourashtra more accessible through speech technology and strengthen digital resources for a language with a comparatively limited presence in mainstream technology platforms.
The four models use the VITS architecture to convert written Sourashtra text into spoken audio. They support male and female voices, with each voice model accommodating text in both Tamil script and Sourashtra script.This provides flexibility for users who write the language using different scripts.
The development draws on a dedicated speech dataset created by IISc’s SPIRE Lab. The corpus contains more than 28 hours of gender-balanced recordings, including composed sentences and passages from literature. The male voice model was trained on approximately 10,000 utterances, while the female voice model used around 11,400.
The SPICOR TTS 3.0 Sourashtra Corpus is available through BHASHINI’s AIKOSH platform, which hosts datasets and resources intended to support the development of Indian-language artificial intelligence applications. Its availability could help researchers and developers build further speech tools for the language.
According to India’s 2011 Census, Sourashtra had approximately 2.48 lakh speakers. Limited digital language resources can make it harder to develop reliable speech interfaces and other language-based services, particularly for communities whose languages receive less attention from commercial technology developers.
Text-to-speech technology converts written material into audio, potentially supporting digital reading tools, language-learning resources and services for people who prefer listening to text. The new models provide a technical foundation for exploring such applications in Sourashtra, although their real-world reach will depend on further development and adoption.
The project highlights the role of collaboration between public digital infrastructure initiatives and academic research in expanding language technology. As AI-powered speech tools become more widely used, resources for languages such as Sourashtra could help broaden the range of communities represented in India’s digital services.
The models were developed by the Digital India BHASHINI Division in collaboration with the Indian Institute of Science (IISc), Bengaluru. The initiative aims to make Sourashtra more accessible through speech technology and strengthen digital resources for a language with a comparatively limited presence in mainstream technology platforms.
The four models use the VITS architecture to convert written Sourashtra text into spoken audio. They support male and female voices, with each voice model accommodating text in both Tamil script and Sourashtra script.This provides flexibility for users who write the language using different scripts.
The development draws on a dedicated speech dataset created by IISc’s SPIRE Lab. The corpus contains more than 28 hours of gender-balanced recordings, including composed sentences and passages from literature. The male voice model was trained on approximately 10,000 utterances, while the female voice model used around 11,400.
The SPICOR TTS 3.0 Sourashtra Corpus is available through BHASHINI’s AIKOSH platform, which hosts datasets and resources intended to support the development of Indian-language artificial intelligence applications. Its availability could help researchers and developers build further speech tools for the language.
According to India’s 2011 Census, Sourashtra had approximately 2.48 lakh speakers. Limited digital language resources can make it harder to develop reliable speech interfaces and other language-based services, particularly for communities whose languages receive less attention from commercial technology developers.
Text-to-speech technology converts written material into audio, potentially supporting digital reading tools, language-learning resources and services for people who prefer listening to text. The new models provide a technical foundation for exploring such applications in Sourashtra, although their real-world reach will depend on further development and adoption.
The project highlights the role of collaboration between public digital infrastructure initiatives and academic research in expanding language technology. As AI-powered speech tools become more widely used, resources for languages such as Sourashtra could help broaden the range of communities represented in India’s digital services.
