Niti Aayog released a draft paper stating –Artificial Intelligence expecting to boost India’s annual growth rate by 1.3% by 2035. The Niti Aayog draft paper titled ‘Towards Responsible #AIForAll’ also said there is a potential of large-scale adoption of AI in a variety of social sectors. With regard to a possible opportunity, Vernacular.ai is one of India’s most innovative voice automation platforms.
Here, Sourabh Gupta, co-founder and CEO, Vernacular.ai, tells us more. Excerpts from an interview:
DQ: What is the key role that AI and other emerging technologies play in India’s growth?
Sourabh Gupta: According to a recent IDC survey, more than 2,000 IT and line of business decision-makers confirm the adoption of Artificial Intelligence (AI) is growing worldwide. AI is expected to transform the way we humans live and work and this could be by helping with repetitive tasks or customizing products/services. AI systems can minimize occurrences of ‘human error’, assuming that they are programmed correctly and can help in making faster decisions using cognitive technologies.
According to Research and Markets, Artificial Intelligence for speech recognition market in India is anticipated to expand at a CAGR of ~65.17% during the forecast period (2019-2024), and is expected to reach a value of INR 14.61 billion by 2024.
Today, since global economies are trying to recover from the crisis, we are seeing rising demand from enterprises across diverse sectors, who are looking at automating support operations to reduce the costs, and are, at the same time, looking to maintain and deliver better customer engagement and experience,
AI is evolving as a key driver in India’s economic growth. Today, leading companies and industries can boost profitability to transform their business with the use of AI.
AI has the potential to markedly increase industry growth. We believe that voice combined with AI is the future of human interface with machines, and our technology is the most advanced and accurate voice AI platform for 10+ Indian language speech and 160+ dialects.
DQ: How does Vernacular.ai fit into this? Elaborate.
Sourabh Gupta: Vernacular.ai as the name suggests is unique in the depth that we are able to bring to local languages. The dialects and accents in the local languages are extensive. What Vernacular.ai offers is the ability to understand speech nuances and idiolect like age, vocal, gender, eagerness, dialect, pace, etc. of a user and talk to them accordingly.
Our technology is capable of providing hyper-personalized conversation and understanding various aspects of a caller’s speech habits peculiar to a particular person in real-time and latent factors, along with the regular lexical components needed for regular Spoken Language Understanding (SLU).
In a country like India, only a little more than 10% of Indians are reported to speak some English, and even the dialects and accents in the local languages are quite comprehensive, with a significant portion of customers also known to be bi-lingual.
This is a peculiar problem that companies like we at Vernacular.ai have solved, recognizing every minute detail, including, the accent, speech rate, dialect, sentiment, among other nuances.
Convenience is a major factor we bring to our customers because we understand that the next billion people coming online are finding it inconvenient to communicate through texts. It is also seen that most of the text based platforms do not support local languages and in that way Vernacular.ai makes it a way better option to choose.
Never before has voice automation been more in focus than in the post-COVID-19 world, where physical distancing, work from home, and contactless interactions have become the new norm. Vernacular.ai –which helps enterprises automate contact Centre queries using its multilingual voice automation platform, VIVA –is seeing a pronounced increase in demand and new client signups, as the ongoing pandemic accelerates adoption of automation across sectors.
Vernacular Intelligent Voice Assistant, or VIVA is currently being used by our clients across sectors, but especially banking and financial services, food and beverage, and ISP to automate up to 70-80 percent of their contact Centre operations. VIVA, which uses cutting-edge natural language understanding (NLU) and speech recognition technology, with the focus on improving customer engagement and experience.
Sourabh Gupta: With AI increasingly making its way into almost all the product designs and sectors, it is quite crucial to make sure we are leading to a safe future with it. Fairness is one of the major factors that need to be looked into because unfairness in any of the systems will lead to a wide scale impact.
Although there is no set path to measure this fairness, maximum action needs to be taken to ensure any discriminative situations.
One of the best implemented ways to avoid these biases is to build a diverse team giving way to more inclusive experiences. Diversity on the basis of gender, race, education, disability and skill set. For a responsible AI, the team also needs to conduct constant user research and testing, and training the system accordingly.
DQ: What is Vernacular.ai doing in this area?
Sourabh Gupta: Since we build and are moving towards building hyper personalized user experience in voice conversations, we work on many algorithms and datasets that try to model users in some form.
While we at Vernacular.ai try to remove all possible bias through an active area of research, we also strive to reduce the known impacts which might be representative of a discriminative situation. We do this using the following:
• While evaluation, we make sure the dataset penalizes errors equally, irrespective of the labels since the label distribution might reflect real world biases. This is also important from a performance perspective since a model that has really learned a problem, like gender classification, should work irrespective of how many females or malesdata points come while serving.
• While modelling, we pick features so that the model doesn’t depend on things it should not use to make final decisions. These are specifically driven by approaches to work with more interpretable models wherever possible.
• The serving pipeline makes sure to state which representations are good to rely on for making important decisions. For example, a bad speaker language classification (fairness sense) doesn’t get used to make decisions about serving or not serving a customer. If used, they only provide soft signals to help us handle errors gracefully.
Additionally, we actively try to keep up with industry standards like this to make sure we are in sync with recent research and techniques. Since presence or absence of bias impacts how well an ML model generalizes too, we consider these parts of the process while developing and working with models.
DQ: What sort of sector-specific laws for AI will need to be created to protect human interests?
Sourabh Gupta: While responsible use of Artificial Intelligence continues to be a hot topic for discussion, there need to be specific laws that can be introduced. Privacy factor to the consumers of technology. The protection of personal or sensitive data based on respect of freedom of expression, opinions, health records, bank account numbers, etc.