BHASHINI and IISc Develop AI Text-to-Speech Models for Sourashtra Language 2026 | भाषिणी और IISc ने सौराष्ट्र भाषा के लिए AI वाक् संश्लेषण मॉडल विकसित किए

1. Why in News?

On 16 October 2026, the Ministry of Electronics and Information Technology announced that the Digital India BHASHINI Division, in collaboration with the Indian Institute of Science, Bengaluru, had developed artificial intelligence-based Text-to-Speech models for the Sourashtra language.

Sourashtra is a minority Indo-Aryan language spoken predominantly by the Sourashtra community in Tamil Nadu. The initiative aims to improve digital access, language preservation, educational applications and accessibility for speakers of a language with relatively limited digital resources.

2. Key Highlights

• Announcement date: 16 October 2026.
• Ministry: Ministry of Electronics and Information Technology.
• Government initiative: Digital India BHASHINI Division.
• Research partner: Indian Institute of Science, Bengaluru.
• Language: Sourashtra.
• Technology: AI-based Text-to-Speech.
• Architecture: VITS.
• Number of models: Four.
• Voices: One male and one female.
• Supported writing systems: Tamil and Sourashtra scripts.
• Speech corpus: More than 28 hours.
• Research laboratory: SPIRE Lab, IISc Bengaluru.
• Dataset: SPICOR TTS 3.0 Sourashtra Corpus.
• Dataset platform: AIKOSH.
• Audio sampling frequency: 22,050 Hz.
• 2011 Census speaker population: Approximately 2.48 lakh.
• PIB Release ID: 2324289.

3. What Is Text-to-Speech Technology?

Text-to-Speech, abbreviated TTS, is a technology that converts written text into spoken audio. Modern TTS systems commonly use machine learning to produce speech that resembles a human voice.

For example, an educational application can use TTS to read written lessons aloud. This may help language learners and people who prefer or require spoken content.

TTS should not be confused with speech recognition, which performs the opposite task of converting spoken language into text.

4. What Is Sourashtra?

Sourashtra is an Indo-Aryan language associated with a community living predominantly in Tamil Nadu.

The name of the language should not be confused with the Saurashtra geographical region of Gujarat. Although the names are historically related, the official announcement concerns the Sourashtra-speaking community and its language technology requirements.

According to the 2011 Census figure cited by the ministry, approximately 2.48 lakh people speak Sourashtra.

5. Why Is the Initiative Important?

Major languages often have substantial collections of digitised text, recorded speech and software tools. Smaller languages may lack sufficient datasets for developing comparable digital services.

Without language-specific resources, speakers may find fewer opportunities to use their language in educational software, accessibility tools and other digital applications.

The new models address this gap by providing speech generation capabilities and reusable research resources for Sourashtra.

6. Which Organisations Developed the Models?

The initiative was undertaken by the Digital India BHASHINI Division, under Digital India Corporation and the Ministry of Electronics and Information Technology, in collaboration with the Indian Institute of Science, Bengaluru.

The speech corpus used for model development was prepared at IISc Bengaluru's SPIRE Lab.

The project illustrates cooperation between a national digital-language initiative and an academic research institution.

7. What Is BHASHINI?

BHASHINI is a Government of India initiative focused on creating an ecosystem for AI-powered multilingual digital inclusion.

It supports language technologies such as translation, speech recognition, speech synthesis and transliteration.

The broader objective is to make digital information and services more accessible across India's linguistic diversity.

8. Institutional Structure of BHASHINI

The Digital India BHASHINI Division operates under Digital India Corporation, which functions within the Ministry of Electronics and Information Technology.

For examinations, the important institutional sequence is Ministry of Electronics and Information Technology, Digital India Corporation and Digital India BHASHINI Division.

9. What Is the Indian Institute of Science?

The Indian Institute of Science, commonly called IISc, is a leading research and higher education institution located in Bengaluru, Karnataka.

Its involvement in the Sourashtra TTS initiative highlights the importance of academic research in building language datasets and developing AI systems for languages with limited digital resources.

10. What Is VITS?

VITS stands for Variational Inference with adversarial learning for end-to-end Text-to-Speech.

It is a neural speech synthesis architecture that can generate speech from written text.

The Sourashtra models use this architecture to produce speech directly from character-level input.

11. What Does End-to-End Mean?

In machine learning, an end-to-end system is designed to learn a transformation from input to output with fewer separately engineered intermediate stages.

For this initiative, the relevant transformation is from written Sourashtra text to speech audio.

The official announcement states that the models operate on character-level text without requiring a separate phonemizer.

12. What Is a Phonemizer?

A phonemizer converts written text into a representation of speech sounds, often called phonemes.

Some speech synthesis systems use a separate phonemizer as a preprocessing component.

The new Sourashtra models can operate directly on character-level text, eliminating the need for a separate phonemizer in the reported model arrangement.

13. Four Speech Models

The project includes four TTS models formed from two speaker voices and two writing systems.

There is a male-voice model for Tamil-script input and another for Sourashtra-script input. Similarly, there is a female-voice model for each of the two scripts.

Thus, four models does not mean four different human voice artists. The official release identifies one male and one female voice artist.

14. Why Are Two Scripts Supported?

The models accept Sourashtra language text written in Tamil script and in Sourashtra script.

This is significant because the language is used in different writing practices. Supporting both scripts may make the technology useful to a wider range of readers and developers.

The output remains Sourashtra speech; the two scripts refer to the writing systems used for the input text.

15. What Is a Speech Corpus?

A speech corpus is an organised collection of recorded speech and associated information, often including written transcriptions.

Such datasets help researchers train and evaluate speech-processing models.

The quality, diversity and accuracy of recordings and transcriptions can strongly influence the performance of a TTS system.

16. Sourashtra Speech Corpus

The corpus developed at the SPIRE Lab of IISc Bengaluru contains more than 28 hours of gender-balanced Sourashtra speech recordings.

The recordings were made using one male and one female voice artist.

The dataset contains specially composed sentences as well as sentences drawn from literature.

17. Transcriptions in Two Scripts

The corpus includes transcriptions in both Tamil and Sourashtra scripts.

This allows the recorded speech to be associated with written representations used by the language community.

Providing aligned audio and text resources can support future research in speech synthesis and other language-processing tasks.

18. Number of Training Utterances

The male-voice model was trained on approximately 10,000 utterances, while the female-voice model used approximately 11,400 utterances.

An utterance is a segment of spoken language, such as a word, phrase or sentence, used as a unit in a speech dataset.

The ministry stated that the speech and text data were carefully prepared and cleaned for model development.

19. Audio Sampling Frequency

The models generate audio at a sampling frequency of 22,050 hertz.

In digital audio, the sampling frequency describes how many samples of an audio signal are represented per second.

A sampling rate of 22,050 Hz means 22,050 samples per second. It is a technical audio parameter and should not be interpreted as the speed at which the model speaks.

20. What Is AIKOSH?

AIKOSH is a platform within the BHASHINI ecosystem that provides access to language datasets and resources supporting the development of Indian-language technologies.

Making datasets discoverable and available to researchers can encourage additional experimentation and application development.

The official announcement identifies AIKOSH as the platform hosting the Sourashtra corpus.

21. What Is SPICOR TTS 3.0?

SPICOR TTS 3.0 Sourashtra Corpus is the dataset made available through AIKOSH for this initiative.

It contains more than 28 hours of gender-balanced Sourashtra speech data developed at IISc Bengaluru.

The corpus includes diverse and literary content represented in both Tamil and Sourashtra scripts.

22. Low-Resource Languages

A low-resource language is a language for which relatively limited digital datasets, computational tools or annotated resources are available.

The term describes the availability of technology resources. It does not imply that the language is culturally unimportant or linguistically inferior.

Developing datasets for such languages is an important step towards more inclusive AI systems.

23. Digital Language Preservation

Digital preservation can involve recording speech, documenting written forms and creating resources that future generations can study or use.

The Sourashtra corpus includes recorded speech and written transcriptions, making it relevant to both technological development and language documentation.

However, a speech synthesis model alone cannot guarantee the long-term preservation of a language. Continued community use and documentation remain important.

24. Benefits for Education

TTS technology can help students hear written material pronounced aloud.

Potential applications include reading practice, pronunciation support and language-learning software.

For Sourashtra learners, the availability of speech models may create opportunities to develop educational resources in the language.

25. Benefits for Accessibility

Spoken digital content can assist users who have difficulty reading text or who prefer audio-based interaction.

TTS can support reading applications, assistive interfaces and audio delivery of selected information.

Actual accessibility depends on how developers integrate the technology into usable services.

26. Benefits for Developers

Developers can use language models and datasets as foundations for creating new digital applications.

Potential products include text-reading tools, educational applications and research prototypes.

The ministry presents the Sourashtra resources as a foundation for further development rather than announcing that all such applications have already been launched.

27. Benefits for Researchers

Speech researchers need reliable datasets to compare techniques, train models and study language-specific challenges.

The availability of recorded speech and transcriptions can support research on speech synthesis for underrepresented languages.

It may also encourage collaboration between universities, public institutions and technology developers.

28. BHASHINI Language Coverage

According to the 16 October 2026 PIB announcement, BHASHINI supports 36 Indian text languages, 23 Indian voice languages and 35 international languages.

These figures describe the platform's reported coverage at the time of the announcement.

Students should distinguish text-language coverage from voice-language coverage because the two figures are different.

29. Speech Synthesis Versus Speech Recognition

Speech synthesis converts written text into audio. Speech recognition converts spoken audio into written text.

Both are important branches of language technology, but they perform different tasks.

The newly announced Sourashtra models are Text-to-Speech systems, not a newly announced speech-to-text recognition service.

30. Speech Synthesis Versus Translation

Translation changes content from one language into another language.

Text-to-Speech produces audio from written text, generally without changing the language of the content.

The Sourashtra initiative concerns generating Sourashtra speech from Sourashtra text written in either of two scripts.

31. Artificial Intelligence and Linguistic Diversity

AI services are more inclusive when they can support languages used by different communities.

Without targeted dataset development, smaller languages may remain underrepresented in commercial AI products.

The Sourashtra initiative demonstrates how collaboration between public institutions and researchers can help address this imbalance.

32. Responsible Use and Limitations

AI-generated speech may contain pronunciation errors or other imperfections. Performance can vary depending on the input and the training data.

Developers should evaluate accuracy, user experience and appropriate safeguards before relying on a model in important applications.

The official announcement describes the models and their potential applications but does not publish a comprehensive independent performance comparison against every alternative TTS system.

33. Relevance for Digital India

Digital India seeks to strengthen digital services and technology adoption.

Multilingual language technologies can make digital interfaces more accessible to people who prefer using regional or community languages.

The Sourashtra project contributes to this objective by expanding speech technology resources for a relatively underrepresented Indian language.

34. UPSC GS-I Relevance

The topic is relevant to Indian linguistic diversity, cultural preservation and the relationship between language and identity.

It also illustrates how digital documentation can support the continued use and study of community languages.

35. UPSC GS-III Relevance

The initiative is relevant to artificial intelligence, digital infrastructure, research and development, inclusive technology and the application of science to social needs.

A mains answer may discuss the importance of public datasets, academic collaboration and language technology in reducing digital exclusion.

36. SSC, Railway and State PCS Relevance

Important examination facts include the Sourashtra language, IISc Bengaluru, BHASHINI, VITS, AIKOSH, the number of models, speech corpus duration and the ministry responsible for the initiative.

37. Important Examination Facts

• Initiative: AI-based Text-to-Speech models for Sourashtra.
• Announcement date: 16 October 2026.
• Organisations: Digital India BHASHINI Division and IISc Bengaluru.
• Ministry: Ministry of Electronics and Information Technology.
• Language family: Indo-Aryan.
• Predominant speaker community location: Tamil Nadu.
• Architecture: VITS.
• Full form: Variational Inference with adversarial learning for end-to-end Text-to-Speech.
• Models: Four.
• Voices: Male and female.
• Scripts: Tamil and Sourashtra.
• Corpus duration: More than 28 hours.
• Research laboratory: SPIRE Lab, IISc Bengaluru.
• Dataset: SPICOR TTS 3.0 Sourashtra Corpus.
• Dataset platform: AIKOSH.
• Male training utterances: Approximately 10,000.
• Female training utterances: Approximately 11,400.
• Audio sampling frequency: 22,050 Hz.
• Separate phonemizer: Not required by the reported character-level models.
• Sourashtra speakers: Approximately 2.48 lakh according to the 2011 Census.
• BHASHINI text-language coverage: 36 Indian languages.
• BHASHINI voice-language coverage: 23 Indian languages.
• BHASHINI international-language coverage: 35 languages.
• PIB Release ID: 2324289.

Official Source

Ministry of Electronics and Information Technology, Press Information Bureau, 16 October 2026, Release ID 2324289.