Wispr Flow bets on voice AI in India with Hinglish dictation, defying market difficulties
US-based voice AI startup Wispr Flow says its India user base has doubled in the past seven months — growth that came after the company shipped a dictation model customised for Hinglish, the natural mix of Hindi and English. Wispr Flow, founded in 2022 in San Francisco by former Apple engineer Tanay Kothari and Brown alumnus Sahaj Garg, positions itself as an app for fast voice-to-document transcription. The company says the growth pace in India has been distinct from its other markets.
Wispr Flow's India share of annual recurring revenue (ARR) has grown 180 per cent since the Hinglish launch last October. With roughly 100,000 global subscribers, the India portion has risen to about 18 per cent of the total — up from 7 per cent. Kothari told TechCrunch: "Voice AI in India is a different game from the US or Europe. Users do not just want to type in English; in their actual lived language they switch to English mid-sentence and then back to Hindi. That phenomenon is Hinglish, and we have to get it right."
Hinglish is what linguists call code-mixing — a phenomenon in which two languages meet within a single sentence and the grammatical rules derive from both. About 250 million people in India speak Hinglish; that is roughly 75 per cent the population of the entire United States. A standard English-Hindi translation system handles Hinglish in fragmented form and produces errors; Wispr Flow's approach is to retrain the model from scratch to recognise both languages in a single pass.
The technical approach starts from a Whisper-style open-source speech-modelling base but is trained on a dataset built specifically for India. The dataset comprises 1,500 hours of Hinglish speech, gathered from users in Bengaluru, Mumbai, Delhi and Hyderabad. Kothari says the model achieves a word error rate (WER) of 4.7 per cent on Hinglish; for standard English, Wispr Flow targets 2.1 per cent. "What we've done for Hinglish is technically harder than what we did for English in the US — but we have an early window in this market."
Wispr Flow's India strategy diverges from a wider trend in Indian voice AI. Most companies have steered clear of localisation gaps left by big platforms like Apple's Siri, because Indian linguistic groups have wide-ranging needs — Hindi, Tamil, Telugu, Bengali, Marathi and dozens of smaller languages. Wispr Flow chose Hinglish as a single target, representing a homogeneous segment of India's educated, urbanised user base.
The competitive landscape in India is challenging. ElevenLabs, Deepgram and Speechly all offer voice AI from the US but none has shipped a Hinglish-specific model. Among India's local players, Bengaluru-based Krutrim AI and Sarvam AI have built Hindi-focused models, but neither has gone after the specific dimension of Hinglish. Wispr Flow's advantage comes from adapting early to language hybridity.
For revenue, Wispr Flow runs a different pricing in India from its global model. The US monthly subscription, $12, sells in India for $4 — an adjustment to the realities of India's consumer electronics market. The company offsets the lower margin with higher volume. India ARR is about $1.8 million; for the whole company, that is a small share of total ARR, which is around $36 million a year, but strategically important because of the growth rate.
The true durability of Wispr Flow's India growth is in question with sector analysts. Bengaluru-based AI analyst Pranay Bhardwaj said: "Wispr Flow's lead may last at least two years; but once Krutrim AI or OpenAI ship Hinglish models, Wispr Flow's pricing power weakens. The window for leadership is two years." Wispr Flow's own projection is that India ARR can reach $5 million in the next 18 months.
Wispr Flow's total investment, completed as a $30 million Series A, included Google's CapitalG, Conviction Partners and Ribbit Capital. The company is preparing to seek $50 million more in a follow-on round to expand operations in India; the strategy is to grow the India user base to about $5 million in ARR and to fund the expansion of the AI infrastructure.
Kothari closed with a forward-looking note: "India is currently one of the fastest-growing voice AI markets in the world, but it's also one of the hardest. Hinglish is not just a technical feature; it's a mirror of India's digital culture. If you want to enter this market, you have to build a model that carries the real complexity of the language. Otherwise users won't answer back."