NRI leads text-to-speech book project

Kannada Pustaka is using speech computers to produce audio textbooks for visually impaired students

Voluntary group Kannada Pustaka is producing audiobooks for visually challenged students. The audiobooks are uploaded on to a website (www.kannadapustaka.org) and are freely accessible.

Conceived as a library for Kannada books, Kannada Pustaka is the brainchild of Rakesh Tiwari, a PhD scholar in the UK. He shares the story of how it all came about.

“Visually challenged students, especially from rural schools, don’t have a good grip on English. They usually depend on Braille embossed books which are not the best medium of learning,” he says.

Schools he visited had volunteers recording textbook chapters on MP3 voice recorders, and the quality of recording was poor. The audio chapters were not shared with other schools either. Says Rakesh, “I wanted to address this gap.”

He hit upon the idea of voice synthesis, or the use of a computer to convert text to voice. Singer M D Pallavi, who liked the idea, endorsed it on Facebook and called for proofreading volunteers.

The group first created synthesised versions of 10th standard textbooks, cut them into CDs, and mailed them to every single special school in Karnataka.

Though the audio sounds robotic, visually-challenged students were satisfied with the pronunciation. “But there was still the issue of textbook revision. In no time, new versions and completely new textbooks were published, rendering the old audio files useless. Sending CDs is expensive, and not sustainable. They also end up in cupboards inaccessible to students,” he says. As a solution, the audiobooks were uploaded on to a website, and the recordings can be streamed.

Voice command

The audiobooks can be played on Google Assistant and Google Home. You can listen to the chapters with just a voice command. Audiobooks are produced and added to the open library regularly.

The team

The open library has been functioning for three years. A group of volunteers make this happen under the supervision of Rakesh.

Volunteers are trained in proofreading. The project has about 300 registered volunteers, 40 active and 15 involved in daily proofreading.

The core team

*Lavanya Hebbalmath coordinates the proofreading process.

*Yogesh Siddhananjaiah, proofreader, has clocked the most hours.

*Jyothi Venkatasubramanya, proofreader and in-house language expert, is an Ayurvedic doctor based in the US.

*Parimala Anand and Anand Hemmige are proofreaders and fundraisers. They are a couple based in the US and Metrolife had featured them in connection with their Kannada storytelling podcast ‘Kelirondu Katheya’.

Singer Raghu Dixit is also a supporter of the initiative.

How it works: Text to voice

Textbooks are scanned and organised into chapters.
Volunteers make modifications on Google Doc files.
After 4-5 rounds of review, the almost error-free text is loaded on to a ‘speech computer.’
Here, software program ‘Madhuravachaka’ turns text to voice.

Language question

“We are leaning towards English very strongly; even a simple conversation in Kannada is impossible without English words creeping into a sentence! Kannada, its grammar, and literature have all gone out of fashion for a majority of the millennials. But for us, for the visually-challenged students in rural Karnataka, Kannada remains the only stream of knowledge communication. We should come out of the disproportionate emphasis on urban dynamics. Rural parts are still completely dependent and are sustaining the language,” says Rakesh Tiwari.

Counting followers

The website has about 1.5 lakh recorded unique visits. About 3,000 people listen to the audiobooks regularly.

“There are just about 50 schools for the visually challenged in Karnataka that need these audiobooks. I do not know if the subscribers are visually challenged students, but the reach is promising,” says Rakesh. The team has also been learning from and working with several schools in Shivamogga, Rakesh’s hometown, and Bengaluru.

No human voice

None of the audiobooks in Kannada Pustaka is recorded by humans. The recordings are synthesised by a speech synthesis engine called Madhuravachaka, developed by Dr Arun Rajkumar, computer scientist on the faculty of IIT Madras, and his student Shivakumar H R. The process called for some innovation. “Feeding in text formatted for printing was not working for audio file generation; audio files require a completely different format system. This challenge led us to inventing the first distributed proofreading initiative of India,” he says.

Awards and recognition

‘Kannada Pustaka’ was awarded the best innovation award at the Association of the Kannada Kootas of America in 2018, with a cash prize of 5,000 dollars. But the team didn’t get the cash. “The organisers never paid us or the second and third place winners, but they made sure celebrities who did concerts were paid,” says Rakesh.

Work from UK

Rakesh Tiwari (35) is pursuing his PhD at the University of Leeds, UK, on the sensitivity of Amazonian rain forests to extreme temperatures and heatwaves.

Hailing from Shivamogga, he worked at Indian Institute of Science for five years and then for a network of NGOs across three states, implementing climate-smart agriculture projects, before moving to Leeds in 2016.

Lit coming up

The Karnataka government’s Department of Kannada and Culture signed an MoU with the group, sharing all its open source Kannada literature for a project. The team will soon put up Kannada literary works in audio format open to all.

No ads, please

The lack of funds is the only challenge the team faces. “I have spent about Rs 3 lakh over three years to maintain the website. We don’t want to endorse anyone or sell ads,” says Rakesh. “Kannada Pustaka is a huge basket of education, technology and innovation in one place.”