Skip to content
AntorLet's Talk
Ventures /AI Startup

Voxly.

Real-time voice-to-text for Bangla and other under-served South Asian languages.

BetaFounded 2024AI Startup
Voxly logo
What it does

The product, in plain language.

Voxly takes voice in and produces text out in real time, for languages that GPT-class APIs handle poorly today. Bangla first, with Hindi and Urdu in queue. The product is a developer API plus a consumer transcription app.

The thesis is simple: 230 million Bangla speakers should not be waiting for OpenAI to prioritize their language. The model quality available right now is good enough to ship; what was missing was the productization.

The technical architecture combines fine-tuned acoustic models with a careful evaluation pipeline that catches drift in production. Every shipped feature has measurable accuracy and latency targets — eval harness first, UI second.

How I built it

The story so far.

Started as an internal tool in 2024 to transcribe Bangla podcasts faster. The accuracy was good enough to ship within six weeks of starting. The harder problem was productization — turning the internal tool into something a developer could integrate or a consumer could use without context.

Two lessons that still apply: (1) eval harness before model selection — without it you're picking models on vibes, and (2) cost-per-conversation modeling before scaling — the unit economics of voice processing collapse fast if you're not watching them.

Where we are now

Today's state, with numbers where I have them.

Voxly is in closed beta with developer API access and a consumer-facing transcription app. Accuracy on Bangla speech is competitive with English-on-Whisper. Latency is under 800ms for streaming use.

Early users are journalists, podcasters, and accessibility teams. The product roadmap is being shaped by real usage rather than imagined personas.

Traction

<800msStreaming latency
92%Bangla word accuracy (beta)
230MSpeakers in addressable market
What's next

Roadmap.

Public launch in Q3 2026. Hindi support in beta by Q4. The longer roadmap is Urdu, Tamil, Telugu — the South Asian language coverage that the big labs continue to under-serve.

Strategic conversation with potential acquirers in the AI infrastructure space is open but not driving the product. We're building toward a sustainable revenue base first.

Related ventures

More from the ai startup list.

ORBIX logo

ORBIX

AI Startup
Active

AI-powered Business OS, live across BD, UK, and Luxembourg with 500+ business users.

Details
NOBBYO logo

NOBBYO

AI Startup
Beta

AI knowledge ops — copilot for non-technical operators.

Details
Pannakhata logo

Pannakhata

AI Startup
Beta

AI-assisted bookkeeping and accounting for Bangladeshi SMBs.

Details
Inspired by this build?

Let's build yours next.

If Voxly sparked an idea for your own AI product, the next step is a 15-minute discovery call.