India may not have an English-learning problem as much as it has an English-speaking problem, according to Arpit Mittal, founder of SpeakX. The entrepreneur believes millions of Indians can understand English but struggle to use it confidently in real-world situations, creating a gap between language knowledge and communication.

SpeakX is attempting to address that gap through a voice-first, AI-powered learning platform that enables users to practise spoken English through real-time conversations, personalised feedback and roleplay-based learning.

The company says it has now crossed 10 million learners and 200,000 paying subscribers, while reaching $7.5 million in annual recurring revenue (ARR). It is also looking to expand its language capabilities and take its technology to international markets.

In an exclusive conversation with Indian Startup Times, Mittal discussed the origins of SpeakX, its AI-led learning model, the challenges of building for India’s diverse linguistic landscape, the company’s evolution from Yellow Class and its larger ambition of helping 100 million Indians build confidence in English.

From Yellow Class to SpeakX

Mittal’s journey with SpeakX began with a different business.

In 2020, he started Yellow Class, a platform offering children’s hobby classes across more than 40 categories. While building the company, his team noticed a recurring pattern among parents, particularly mothers. Many could understand English and communicate through reading and writing but lacked confidence when it came to speaking.

The observation led to a fundamental shift in the company’s direction.

As generative AI matured in 2023, Mittal and his team saw an opportunity to use voice technology to provide personalised speaking practice at scale. The company subsequently pivoted towards building SpeakX specifically for learners who understood English but lacked opportunities to practise speaking.

“We don’t have an English-learning problem; we have an English-speaking problem,” Mittal said, describing the insight that shaped the company’s product philosophy.

The focus moved away from traditional grammar and vocabulary lessons towards real-life conversations, roleplays and AI-powered speaking practice.

Making AI the Learning Environment

At SpeakX, AI is not positioned as an additional feature but as the core of the product.

The platform’s voice-first AI tutor, Sia, works with a nine-level adaptive learning ladder that takes learners from beginner to advanced levels. As users speak, the system evaluates areas such as pronunciation, fluency, grammar, vocabulary and sentence formation.

The learning experience then adapts based on the individual’s performance.

For example, a learner preparing for a job interview may receive different practice from someone trying to become more confident during workplace meetings or customer conversations.

SpeakX’s AI also retains information across sessions, allowing it to understand what a learner has struggled with, where improvement has taken place and which areas still require attention.

The company has invested in speech recognition designed for Indian accents and streaming voice technology, with conversational latency brought down to less than two seconds, according to Mittal.

The objective is to make the technology feel almost invisible to the learner.

“The learner should feel like they’re simply having a conversation, while AI is quietly personalising the learning behind it,” Mittal explained.

Building for India’s Diverse Learners

India’s language-learning market is characterised by differences in geography, income, education and linguistic backgrounds. SpeakX says it has found a common thread among many of its users: aspiration combined with limited access to an English-speaking environment.

Around 60% of SpeakX’s active learners come from Tier-2 and Tier-3 India, according to the company, with a large proportion being first-generation English speakers.

The platform therefore allows learners to begin in the language they are most comfortable thinking in before gradually transitioning into English.

The product is also designed around practical constraints faced by Indian consumers, including multilingual interfaces, lower-end Android smartphones and inconsistent connectivity.

According to Mittal, the ambition remains broadly similar across geographies, even though the learner’s starting point may differ.

Voice First, Confidence Before Perfection

One of SpeakX’s key product decisions was to make voice the primary input instead of typing.

Typing allows users to pause, edit and rewrite their responses. Real conversations do not offer the same luxury.

SpeakX therefore encourages learners to practise speaking spontaneously through AI conversations and real-world roleplays covering situations such as job interviews, customer calls, workplace discussions and everyday interactions.

The company also follows a “confidence before perfection” approach.

Rather than correcting every mistake, the AI is designed to identify the feedback that matters most at a particular moment. The objective is to prevent learners from becoming overwhelmed and losing the confidence to continue speaking.

The company says paid users currently average around 13.7 minutes of active speaking practice per day, with 61% practising for more than 20 minutes daily.

Its first-month retention is around 67%, according to the company.

Competing on Access, Relevance and Outcomes

SpeakX operates in a crowded English-learning market, but Mittal believes its differentiation comes down to three factors: access, relevance and outcomes.

Traditional one-on-one English tutoring can cost between ₹500 and ₹3,000 per hour, according to the founder. SpeakX aims to provide one-on-one AI-powered speaking practice at around ₹299 per month.

The company also positions itself differently from global language-learning applications by building specifically for Indian users.

This includes speech recognition tuned for Indian accents, multilingual entry points and technology designed to function on affordable Android devices and relatively limited connectivity.

The company also sees motivation as a key distinction between SpeakX and general-purpose AI models.

While platforms such as GPT or Gemini can conduct conversations, SpeakX is designed specifically around the learner’s history, progress and educational journey.

The company consequently measures success differently from traditional EdTech platforms.

Instead of focusing primarily on lessons completed or time spent watching content, SpeakX looks at whether learners can actually hold conversations and express complete thoughts independently.

Turning Passive Knowledge Into Real-World Communication

For Mittal, one of the biggest gaps in language learning is the transition from knowing English to actually using it.

This is why SpeakX’s learning experience is centred on active speaking rather than passive consumption.

A learner enters the platform without a lengthy onboarding process and starts speaking with the AI within seconds. The initial interaction also helps determine the user’s starting proficiency.

The learner is then placed on SpeakX’s nine-level learning ladder, ranging from Starter to Master.

Progress is connected to practical milestones such as becoming “Interview Ready”, rather than simply collecting points or completing lessons.

The company tracks active speaking minutes, retention, engagement and changes in speaking ability, including whether learners progress from short, prompted responses to independently expressing complete thoughts.

The Challenge of Building AI for Indian Accents

Building an AI-driven language-learning platform for India comes with technological challenges of its own.

According to Mittal, one of the company’s biggest challenges has been developing speech recognition and feedback systems that work across India’s diverse accents, dialects and code-mixed speech.

The company has invested heavily in improving its ability to understand Indian learners, including users who switch between English and Hindi or another Indian language when they get stuck.

Personalisation has also required balancing accuracy with motivation.

An AI system may be capable of identifying numerous mistakes during a conversation, but correcting every error can make learners uncomfortable and discourage them from speaking.

SpeakX therefore focuses on delivering the most relevant feedback at the right moment.

For the company, the objective is to make the AI feel like a supportive coach rather than an evaluator.

From a Children’s EdTech Platform to 10 Million Learners

The evolution from Yellow Class to SpeakX has also changed the company’s business model.

SpeakX began monetising in November 2023, with performance marketing playing an important role in customer acquisition.

However, Mittal says the company learned that acquiring an installation is only the beginning. Retention and sustained engagement ultimately determine whether the business can grow.

Today, SpeakX says it has surpassed 10 million learners, with 200,000 paying subscribers and $7.5 million in ARR.

The company also says it became EBITDA-positive before raising capital, allowing it to approach funding as a means of accelerating growth rather than sustaining operations.

The founder points to the economic impact of improved communication as one reason for the product’s adoption.

For workers and professionals in lower-income segments, even a modest increase in earning potential following improved communication skills can have a meaningful impact on their lives, he said.

What Building SpeakX Taught the Founder

Building SpeakX has also required Mittal to rethink assumptions formed over his previous 15 years of entrepreneurship.

The rapid development of generative AI, according to the founder, has changed not only what technology companies can build but also how products need to be built.

One of the biggest lessons has been the need to challenge established assumptions.

SpeakX itself emerged from that process. The team had an existing business with users and backing but chose to pursue what it believed was a larger problem.

Mittal also believes language learning is fundamentally less about acquiring more content and more about creating opportunities to practise.

The experience of building for Tier-2 and Tier-3 India further challenged assumptions about the aspirations of these consumers.

According to Mittal, many users are not looking for certificates. They are looking for better employment opportunities, greater workplace confidence and the ability to participate more comfortably in conversations.

AI Could Make Language Practice Continuous

Looking ahead, Mittal believes AI will fundamentally change how people learn languages by making practice more personal, continuous and accessible.

One of the biggest limitations of conventional language education is the lack of opportunities to practise speaking outside scheduled classes.

AI can potentially remove that limitation by allowing learners to practise conversations repeatedly, make mistakes, receive feedback and try again.

Mittal expects AI to evolve from being a feature within education products to becoming the learning environment itself.

The technology could understand a learner’s speaking patterns, identify areas of difficulty and adapt learning based on the specific situations they are preparing for.

SpeakX is also exploring VR-based learning experiences designed to recreate real-world communication situations.

However, Mittal believes technology will not eliminate the importance of human interaction.

As AI handles more routine tasks, interpersonal skills, judgement and relationships could become increasingly important.

English as a Door to Opportunity

For SpeakX, the long-term vision extends beyond English education.

Mittal describes English as “the door, not the destination.”

The company’s larger objective is to help people access opportunities that may previously have seemed out of reach, whether that involves performing better in an interview, speaking in a workplace meeting, handling a customer interaction or simply expressing an idea without hesitation.

SpeakX’s stated ambition is to help 100 million Indians find their voice in English, particularly across Tier-2 and Tier-3 India.

The company believes stronger communication skills can contribute to greater employability, economic mobility and participation in education, entrepreneurship and professional life.

The next phase of the company’s journey will involve expanding its language capabilities and taking its technology into international markets.

For Mittal, the ultimate measure of success is not the number of lessons completed or certificates issued. It is whether a learner becomes more confident using their voice in the real world.

The company now aims to build a language-learning platform from India that can adapt to different languages, accents, cultural contexts and learner goals across global markets.

As SpeakX expands, its larger proposition remains centred on the same gap that prompted its creation: helping people move from knowing a language to confidently using it.

Interview By: Arushi Agarwal