On-device voice models that let devices understand and speak the languages of 4.6 billion people
We open new markets for device makers with on-device voice AI. A royalty on every device shipped.
Request the deck- 4.6B
- people in our target regions
- 2.5B
- devices sold every year
- $3
- royalty per device
- $7.5B
- TAM per year
Voice AI still can’t understand or speak to more than half of humanity
Whisper (OpenAI) recognizes 100 of the 7,000+ languages spoken. Most of them are spoken rather than written, so their data isn’t online. For device makers, that means slower launches and delayed revenue.
Meanwhile voice is becoming the interface between people and machines: Gemini talks to drivers, ChatGPT to 150M+ users every week. Voice AI no longer needs the cloud, and each language a device learns opens a new market for its maker.
- South Asia 2.04B
- Sub-Saharan Africa 1.26B
- Southeast Asia ~690M
- Middle East & North Africa ~500M
- Other low-resource regions(Central Asia, Andes, Pacific) ~150M
Voice models ready to run inside everyday devices
The device understands users and answers them in their own language. Everything works offline, so the voice stays on the device: 75 MB models that run on a phone under $200.
Labari Tiny makes 22.4% fewer errors
A per-device royalty, like Dolby, Arm or Cerence AI
We license our models to device makers and earn a royalty on every device shipped with them.
- Price
- $3 per device
- TAM
- $7.5B per year
2.5Bdevices sold every year
- Smartphones1.26B
- Wearables538M
- PCs280M
- TVs200M
- Tablets152M
- Cars90M
We produce the data nobody can download
- 12+ monthsof in-house development saved for the device maker.
- 3 monthsto launch a new language.
- 2 studiosup and running, with a network of narrators and linguists: we control data collection on the ground.
- Know-howturning rarely written languages into AI training data.
The full deck, on request
Market, business model, roadmap, team and the raise: we share the deck on request.
Request the deck