From asking Siri existential questions, to ordering Alexa to give us the weather report, India’s gadget aficionados have come a long way in using voice assistants.
Indian startup Reverie Technologies has created a voice assistant meant exclusively for Indians in seven major local languages.
Subodh Kumar, a tech entrepreneur in his early 30s, has spent almost half a decade of his career working in artificial intelligence (AI) and analytics in New York, Hong Kong and Singapore. Yet, Hindi, his first language, is his preferred language of communication in both professional and personal settings.
The Bengaluru resident hails from Gaya, the second largest city in Bihar. People can speak three different languages in Gaya and none of them is English. When the internet came to his town, his parents got a data plan but hardly ever used basic apps like WhatsApp and Facebook, let alone browse the web. “All they did was read messages and forward them at best,” Kumar recalls. Hindi keyboards were available but typing in Hindi was cumbersome. “What if we could talk to our devices in our local languages,” he thought.
Four years ago, he along with his IIT Kharagpur batchmates Kishore Mundra and Sanjeev Kumar set up a speech-recognition software company called Liv.ai. Currently, the startup has over 500 business-to-business (B2B) and business-to-consumer (B2C) developers using its speech-recognition application program interface (API) for their apps and products. Liv.ai supports 10 regional languages and plans to add many more.
“You can instruct the device to buy groceries, pay your bills, book tickets, all within the comfort of your language. Imagine the amount of time it’ll save and the efficiencies it’ll bring across industries,” says Subodh.
Regional languages is a topic of frequent discussion for tech companies. Subodh’s passion for voice, however, is not an outcome of the trend. “As a child, we never knew the difference between living and non-living things. It defined how we communicated with everything and everyone. That’s what drives me to put life in every device. Hence the name Liv.ai,” he adds. Subodh’s ultimate ambition is to make every device talk to people in their preferred language.
India Finds Its Voice From asking Apple’s Siri existential questions, to ordering Amazon’s Alexa to give us the weather report, India’s smartphone users and gadget aficionados have come a long way in using voice assistants in the last three years. Com-Score, a data analytics firm, predicts that half of all online search will be voice-based by 2020. Accenture’s 2018 Digital Consumer Survey says 39% of India’s online population is likely to own some kind of digital voice assistant by the end of this year. Experts are betting big on voice becoming the preferred mode of transaction for ecommerce, banking, and payments in the near future.
No wonder then that tech majors like Apple, Microsoft (whose digital voice assistant is named Cortana), Amazon, Google, and even Samsung (Bixby) are battling it out to capture India’s 300 million smartphone users (projected to rise to over 440 million by 2022) through their voice assistants - available as a service on smartphones or on smart speakers. But here’s the thing: Over 90% of the current digital voice assistants in India support only English and most Indians are like Subodh’s parents: they may be comfortable in one or more Indian languages, but do not speak English.
So, if the Rs 100 crore voice-and-speechrecognition market in India has to grow, it will have to be on the back of software that supports multiple local languages. Apple has been talking about launching Siri in Hindi for quite a while now. Google has officially announced that a Hindi version of Google Assistant for its Google Home smart speaker will be out by end of this year. Amazon has expanded Alexa’s natural language understanding to comprehend context and intent, even if the sentences include Hindi, Tamil, Telugu, Kannada, Malayalam or Punjabi proper nouns, according to Puneesh Kumar, country manager for Alexa experience & devices. Amazon, however, is noncommittal on whether Alexa, which powers its Echo speakers, will communicate in Hindi.
Indian startup Reverie Technologies has created a voice assistant meant exclusively for Indians in seven major local languages. The team, led by Arvind Pani, Vivek Pani and SK Mohanty, has named the software ‘Gopal’. They wanted to steer clear of international-sounding names so as not to alienate people. Currently in a beta-testing stage, Gopal is finetuning its speech-to-text capabilities for all the possible accents in those seven languages. Cofounder Vivek recently had his parents in Bhubaneswar successfully order groceries through Gopal.
Besides Liv.ai and Reverie, another Bengaluru-based startup called Vokal is making strides in the field of voice recognition and assistance. Founded by Aprameya Radhakrishna (a former cofounder of TaxiForSure) and Mayank Bidawatka, Vokal is essentially a voicebased Quora but in regional languages. It aims to build an online repository of regional content while getting experts to answer user questions. Currently, it supports only Hindi but the team is working on replicating the speech-to-text tech in 10 Indian languages.
Not in English, Please “The real India doesn’t feel the necessity to learn English but unfortunately, over 50% of the content on the internet is in English. The internet needs to be built for them, in their language,” says Radhakrishna. He points out how the queries of an English-speaking online user differ from that of a non-English-speaking one.
“They don’t Google how to get into an IIT or an IIM. They want to find out how to crack the railways exam, what to do after 12th standard, how to become a bank probationary officer.” These answers are not readily available on the internet, not in their preferred language of communication at least. Vokal currently has a voicebased content library of 50,000 questions answered by a select set of experts. For any question outside of the library, the average response time is less than half an hour.
Vokal does not pay the experts and Radhakrishna thinks eventually he will be able to provide them alternate monetising opportunities, like live sessions or special paid requests, as opposed to enrolling them on a fee model. He admits Vokal does not have a robust revenue model in place at present, even though the platform has been adding 2,00,000 active users every month for the past four months. But he has investors like Accel Partners and Blume Ventures backing his venture. “It’s too early to have a concrete monetisation strategy. We are confident voice will drive new discovery models and change consumer habits in the near future,” says Prashanth Prakash, partner at Accel Partners.
For the homegrown players, the need to grow the market is quite personal since each one of them has grown up speaking an Indian language that they want to inculcate in their voice tech. For the major tech players, an understanding of the huge business opportunity and the availability of ample resources drive them to expand the scope of voice assistant market in India. But it is not as simple as it sounds.
Clarity in Cacophony Collecting speech data in various accents, and from diverse groups is no mean feat, especially in a country like India which has 22 official languages. ET Magazine asked Daan van Esch, Google’s technical program manager, about the challenges of collecting data in different Indian languages, and then analysing it to create a fool-proof automatic speech recognition model. “The background noise is very different in India as compared to say in Europe, so your acoustic model that converts sound to phonetics has to be robust,” he says.
Creating a training model for each language takes weeks on end, he adds. “The written text reference for languages like Santali, spoken by 6 million people across India, Bangladesh, Nepal and Bhutan, dates back to decades-old books.” Van Esch is a Dutch national residing in California who knows more about Hindi maatras (vowel signs) than many Hindi scholars put together.
“You don’t have to speak the language to understand its technicalities,” he says. He knows that Wagdi is a language spoken by over 2 million people of southern Rajasthan. He knows Punjabi is written in two scripts: Gurmukhi (in India) and Shahmukhi (resembling Urdu, in Pakistan), a fact that many native speakers of the language are often oblivious to. He has to know these things because they add to the list of challenges in recording and analysing these languages.
Indian government organisations like the Mysore-based Central Institute of Indian Languages (CIIL) have been trying to fill the country’s linguistic gap since 2008. “A team of 40 people started collecting speech data in 2012. We have managed to get data in 19 Indian languages so far. This data will be released in public soon,” says Narayan Chaudhary, officer-in-charge at CIIL. For tech players, the challenge is to collect and process enough data to be able to accurately read all the different accents and dialects of one language For CIIL, finding linguists to collect and analyse speech data in some languages like Sindhi, Gujarati and Kashmiri is also a struggle. A lot of homegrown companies are working with each other and the tech giants to combat the challenges together.
Grand View Research pegs the global voice-recognition market at $127.6 billion by 2024. If India is to make a sizeable contribution to this figure, multiple-language-supporting voice assistants might not be enough. As Karthik Reddy, cofounder and managing partner, Blume Ventures, points out, “Indians don’t pay well. Even the likes of Google are building voice-tech as an infrastructural element right now. Unless you find a consumer hook for your software, you’ll remain in the B2B play and scaling up will remain an issue.” It is thus clear sweet-talking alone won’t win Indians over.
Get Unlimited Access to The Economic Times
