IIT Madras working on AI solution for text to speech translation of Indian languages: V Kamakoti
- 16th Mar 2023
-
Money Control
IIT Madras director V Kamakoti said that researchers working at the Nilekani Centre at AI4Bharat have developed a 9-language text to text translator, and that they are working on converting translated text into speech.
Researchers at Indian Institute of Technology, Madras are developing a nine-language enabled text-to-text automatic translator with an aim to bridge the language gap between various parts of the country as part of Ministry of Electronics and Information Technology (Meity)'s Bhashini initiative, IIT Madras director V Kamakoti told Moneycontrol in an interview.
Ultimately, the aim is to develop a text-to-speech translator, Kamakoti added.
He said the researchers have developed algorithms that can translate a piece of text to text in Tamil, Kannada, Telugu, Malayalam, Hindi, Bengali, Gujarati, Marathi, and English automatically.
The work is being done at the 'Nilekani Centre at AI4Bharat' in the IIT Madras campus, which was launched in 2022 with a grant of Rs 36 crore through Nilekanti Philanthropies.
"Today what is happening at Bhashini is translations are being made from English to multiple languages and from multiple languages to multiple languages. This translation is happening where we give a text and then it translates and give us back," Kamakoti said.
MeitY's Bhashini aims to build a National Public Digital Platform for languages to develop services and products for citizens by leveraging the power of artificial intelligence and other emerging technologies.
"Over a period of time the algorithm (for text-to-text) learns. There is artificial intelligence -- a heavy level of NATURAL language processing going on that affects this translation," he said.
Work has also been completed at the Nilekani Centre, Kamakoti said, on translating from speech to text.
However, Kamakoti said that this feature may not support as many languages as text-to-text translations do. He explained that speech to text translations would have immense benefits in the education sector.
"I teach an advanced course in Physics, and say I want to translate it into some Indian language. The speech will get extracted by the AI as text, the text will get translated by the engine, and you will get the translated text as subtitles," he said.
But that's not where the research at the Nilekani Centre at AI4Bharat ends.
Researchers are also working on getting the translated text converted into speech.
"For instance, say there is a video where I am talking in English. The first step is to extract the text using AI, and then translate it. So, now either the translated version can be sent as text, or now, we are working on converting the translated text into voice," he said.
"So I can talk in English, and the next video would be as if I am talking in Hindi," he added.
The Nilekani Centre at AI4Bharat has been working on translation, transliteration, speech recognition, language understanding, language generation, and sign language.
"With increasing digital penetration and the preference for regional language content on the web, a good translation system for Indian languages is a necessity to provide equitable access to information and content," the AI4Bharat website says.