Arpit Mittal on How SpeakX Is Cracking India’s English-Speaking Puzzle with AI

‘Communication works for those who work at it,’ said composer John Powell — yet, speaking confidently remains a challenge for many. Arpit Mittal, Founder & CEO of SpeakX, is working to change that with an AI-powered platform designed to help learners across India improve their spoken English.

In this interaction with StartupTalky, Mittal shares how SpeakX adapts in real time to users' regional accents and fluency levels. He explains how the team uses speech recognition, user data, and AI to boost learners' confidence. He also talks about the company’s subscription-based model, its growth through both paid and organic channels, and future plans involving regional expansion and strategic partnerships with edtech and telecom players.

StartupTalky: How does SpeakX’s Generative AI adapt lessons in real time for each learner, and what unique advancements in speech recognition make this possible?

Mr. Mittal: At SpeakX, we are deeply aware that English learners in India span a wide range from complete beginners who struggle to put together even a basic sentence, to advanced users aiming to polish their fluency to near-native levels. To meet every learner where they are, our Generative AI constantly adapts in real time. It listens closely, analyses how a user is speaking, right down to fluency, vocabulary usage, and pronunciation, and adjusts the complexity of the practice session accordingly.

For example, if a learner is getting comfortable forming basic five-word sentences, the AI gently nudges them to try more complex constructions using conjunctions or idiomatic expressions.

What makes this seamless adaptability possible is our investment in highly refined speech recognition technology, specifically tuned for the Indian context. India has a multitude of accents, even varying within a single state. Our system has been trained on massive datasets featuring real voices from across regions, North, South, East, and West, so that it can understand and provide accurate feedback, regardless of whether a learner speaks with a Punjabi, Bengali, Tamil, or Marathi accent. That inclusivity and precision allow us to create a learning journey that feels personal and motivating.

StartupTalky: How has SpeakX's selection for the Google for Startups Accelerator Apps program impacted its growth and innovation?

Mr. Mittal: Getting selected for the Google for Startups Accelerator was more than just a badge of honour, it was a transformational moment for SpeakX. The program brought us into close contact with world-class mentors, product specialists, and engineers who helped us zoom in on the right growth levers. The structured sessions helped us identify user drop-off points and refine engagement strategies, which significantly improved retention.

Perhaps more importantly, it forced us to take a step back and see the bigger picture. With Google's guidance, we were able to fine-tune our roadmap to balance short-term user acquisition wins with our longer-term vision—building a scalable, AI-powered platform that truly changes how Indians learn to speak English. It also helped us think more globally, even while staying hyper-focused on our local strengths.

StartupTalky: With 10 million+ downloads, what has been your most effective growth strategy—organic, paid marketing, or partnerships?

Mr. Mittal: Reaching 10 million downloads did not happen overnight, and it certainly was not driven by one single channel. That said, our most effective strategy has been performance-driven paid marketing, particularly on platforms like Google and Facebook.

What worked in our favour is that we understand Indian learners extremely well. We communicate in the languages and dialects they are most comfortable with, and we speak to their real needs: getting a better job, speaking with confidence, or helping their kids learn English.

What is exciting is that as we scaled our campaigns, our customer acquisition cost dropped, which is something rare in edtech. That tells us two things: first, that the demand for spoken English tools in India is massive; and second, that there’s still a largely untapped audience looking for exactly what we offer.

Organic growth and word-of-mouth also played a role, especially once learners started seeing progress within weeks and began recommending us to their friends and family.

StartupTalky: What role does AI play in monetisation? Do you see potential for alternative revenue models beyond subscriptions?

Mr. Mittal: AI plays a crucial role not just in product delivery but in monetisation as well. Thanks to AI, we can offer highly personalised learning at scale without the need for live human tutors, which keeps our costs low and our pricing affordable. Our monthly subscription model has emerged as the most natural fit, especially in India, where families are accustomed to paying recurring fees for tuition or coaching classes. It aligns well with how people already invest in education.

That said, we are exploring other revenue models too. Freemium models, tiered plans, and even B2B offerings for schools or coaching centres are on the table. But for now, subscriptions give us both sustainability and predictability. More importantly, they allow us to keep improving the experience for every learner without needing to rely on ads or upsells.

StartupTalky: What distinguishes SpeakX from competitors like Duolingo and ELSA Speak beyond AI-driven learning?

Mr. Mittal: Duolingo is fantastic when it comes to introducing people to a new language in a gamified way, and ELSA Speak does a great job with test-prep-focused pronunciation feedback. But SpeakX was built with a different mission altogether—helping Indians speak English confidently in everyday life. That focus shapes everything we do.

We are not trying to teach 30 languages or prepare people for IELTS alone. We focus on real-life communication, like how to speak to your manager at work, how to explain symptoms to a doctor, or how to introduce yourself confidently in a college interview. Our curriculum is rooted in real scenarios our users face daily. And because we train users in an Indian context, with culturally relevant prompts and accents, our learners feel immediately at home, and that leads to faster progress.

StartupTalky: How does SpeakX leverage user data and AI to personalise learning experiences, and what key insights from user behaviour have influenced its product evolution and roadmap?

Mr. Mittal: One of the biggest things we have learned from user data is that the biggest barrier for most learners is not lack of knowledge, it is lack of confidence. Many Indians can read and understand English, and even write well. But when it comes to speaking, they hesitate. They fear making mistakes, being judged, or simply freezing up.

Our AI is designed to gently coach them past those fears. It picks up on where they are struggling, be it forming full sentences, using the right tense, or pronouncing certain sounds, and tailors the practice accordingly. If someone is repeating errors in pronunciation, they will get more focused practice. If someone is holding back from speaking, we reduce the pressure and create easier prompts to build momentum.

This insight has shaped our product philosophy. We have moved away from textbook-style lessons and grammar-heavy drills and instead focused on delivering a more intuitive, confidence-building experience that gets people talking.

StartupTalky: What were the most significant technical and ethical challenges in integrating real-time AI coaching, and how did you overcome them?

Mr. Mittal: Technically, the biggest challenge was affordability. Building a real-time AI tutor with speech feedback requires significant computing power and processing. We knew that to reach India’s vast lower and middle-income populations, we had to keep costs low without compromising quality. Thankfully, the rise of efficient open-source AI models has made that possible—we are now able to deliver world-class speech analysis at a fraction of the cost it would have taken just a few years ago.

Ethically, a major challenge is ensuring fairness and inclusivity, especially in a country like India with so many languages, dialects, and accents. Traditional speech recognition models often perform poorly when faced with code-mixed language (like Hinglish) or strong regional accents. We have had to train and retrain our models to ensure they understand and fairly assess all users. We have also taken care to design the feedback in a way that is encouraging rather than critical, because for learners to open up, they need to feel safe making mistakes.

StartupTalky: What’s next for SpeakX? Are there plans for expansion, new product features, or strategic partnerships?

Mr. Mittal: We are entering an exciting phase of growth. Right now, we are laser-focused on keeping learners motivated over time. Speaking a new language is not a weekend project; it takes weeks of steady effort, and keeping users engaged through that journey is our top priority. 

From a business perspective, we are currently strongest in the Hindi-speaking belt, where we have built a solid user base of 125,000+ paying monthly users, with $4 million in annual revenue and over $1 million in profit.

Now, we are preparing to expand to other linguistic regions across India, including Tamil Nadu, Maharashtra, and West Bengal, etc. We are also exploring strategic partnerships with edtech players and telecom providers to deepen our reach. Our ambition is clear: to become India’s leading English-speaking brand and to empower millions more with the confidence to speak up.

Mastering Communication Skills: Essential Books for Effective Communication

Explore a curated list of books to develop your communication skills, from navigating difficult conversations to mastering the art of persuasion.

StartupTalkyRia Sahni