Deploying Voice AI in India’s Multilingual Market

We can't get enough of Alexa or Siri or even Google Echo, can we?

In the last few years, we have seen many changes in the way we work, shop, communicate and access entertainment. With these changes, there are also major innovations in Voice AI. For example, conversational intelligence tools extract insights from conference calls and sales calls to improve efficiency of the teams. Similarly, intelligent voice assistants are being enhanced by various industries to serve and engage better with the consumers. Today, we can also use voice to search on the web and deploy voice notes or recording to search.

While Voice AI has proliferated in Indian homes through Alexa and Google Echo, as well as cars with in-built voice solutions, it is predominantly in English. This narrows access to technology in a multilingual market like India, which is dominated by regional speech communities.

Realizing this need, a Kolkata-based startup, Mihup, has launched its AI-powered vernacular voice interface for vehicles. The startup has developed a software—AVA Auto—for automobiles that allows users to interact with a car system in multiple languages. The system supports various Indian languages, including code-mixed speech like 'Hinglish' (Hindi and English), and aims to introduce other major Indian languages by 2022, as reported by YourStory. Flipkart's 'Supermart', an online grocery store in India, introduced a voice assistant that allows users to buy products in multiple languages, starting with Hindi and English. Similarly, in 2020, Vernacular.ai, an AI-first SaaS business, partnered with Axis Bank to launch AXAA, an AI-powered multilingual voice bot to resolve customer queries.

Power of Voice AI Across Industries

Consumers are looking for seamless experiences, and the capabilities of modern technology are shaping this demand. There are several use cases driving the expansion of Voice AI to support and enhance businesses across sectors. The most notable growth of Voice AI has been in banking, media, retail, and healthcare, as organizations seek to keep pace with the consumer's rising demand.

The main reason for this shift towards voice is convenience, and increased adoption by younger consumers. Moreover, with the number of internet of things (IoT) devices on the rise, such as smart watches, appliances, speakers, smart home systems etc., voice applications have more utility than ever, in the ever-connected lives of the users.

From the industry front, use of voice applications in healthcare is enabling the ecosystem to manage patient care and provide 24×7 assistance, especially in absence of healthcare staff. Similarly for the financial sector, especially insurance companies, voice has played an important role in sharing information and guiding users to the right platforms for their requirements. This has enabled insurance companies to manage large volumes of customer queries and grievances more efficiently and in time.

Challenges for Voice AI in India

Voice AI might be the newest technological innovation in India, but it is also one of the most popular ones. There are certain challenges hindering the adoption of Voice AI across industries:

• Low accuracy/Poor interpretation – Accuracy has been a persisting problem with voice technology. 73% of businesses thus believe that lack of accuracy is the key reason why they don't use voice technology, as stated by an industry report. This holds true for a market like India with multiple languages.
• Expensive and time consuming – Monetary investment has been one of the key reasons preventing the development of technology, especially in developing countries like India. Organizations are hesitant to adopt technologies like Voice AI because of the cost and time involved, as it is rarely homegrown for countries like India.
• Language variation – Voice applications are predominantly in English, which makes it challenging for the software to comprehend Indian regional languages. This further proves to be not so useful for businesses in India to invest in.
• Limited data from low-resource languages – Development of Voice AI needs human intelligence to train and tune automatic speech recognition and natural language understanding models. While multilingual large language models like Whisper from OpenAI are increasingly powerful, model performance on low-resource languages continues to lag behind by comparison to English.

Improving Voice AI Adoption

A report by PWC explains that one of the three main reasons why users are scared to experiment with voice tech is simply a lack of trust. Here are a few features that can help improve Voice AI adoption in India.

• Increase Accuracy – Building high-quality training datasets by sourcing right data, performing robust data annotation, and testing and updating training data regularly will improve voice AI accuracy.
• Multilingual Voice AI – Voice applications are now being used by several industries and sectors. While they are proving beneficial in empowering businesses, a large section of the market remains untapped because of language limitations. There is a huge scope for businesses across sectors to benefit from deploying Voice AI with multilingual capabilities.
• Improve Security – Research shows that worldwide, 45% of smart speaker users are concerned about voice data privacy, and 42% worry about voice data hacking. In another survey, 59% of respondents said privacy is an important factor when using voice control devices. Thus, improving security of voice data and adding stringent measures to protect customers is a vital step in building trust for Voice AI.

For a relatively conservative but ready-to-rave market like India, awareness of and trust in Voice AI are key drivers of adoption. While technology innovators are building voice applications for the world, regions like India with its diverse culture and customer behavior need special measures to be able to deploy Voice AI. While the makers of Voice AI are innovating to bring in newer capabilities of voice applications, it is of utmost importance to fill in the need gap for consumers, especially in a market like India. And multilingual voice applications are just the beginning.

Author:

Teresa O'Neill, PhD, Director, Natural Language Solutions

Teresa O'Neill is the Director of Natural Language Solutions at iMerit, where she leads iMerit's human-in-the-loop data enrichment practice for Speech and Natural Language Understanding use cases. Teresa previously served as Principal Language Architect at OpenCity and helped bring a voice-based conversational AI assistant to market in the hospitality industry. Before moving to the ML/AI space, Teresa worked in academia, where her research and teaching focused on theoretical syntax, grammatical variation in speech, and challenges facing low-resource language communities. She is passionate about embedding human insights in the AI development process to create more useful, accessible, and context-sensitive applications.