Lokutor

Small voice AI models. We are a company based in Madrid building speech recognition, speech synthesis, turn-taking and noise suppression models small enough to run where the user is: in our cloud, on your own servers, or on the device.

Small models make voice agents cheaper to run, and they are the only kind that can go on premises, on robots and on connected devices.

What we build

• Versa, text to speech. Real-time speech synthesis in nine languages, including Spanish, Catalan, Galician and Basque. Try it in our Space: Lokutor Versa 2.0 TTS on CPU.
• Oído, speech recognition on a microcontroller. Open-vocabulary English recognition running on an ESP32-S3, with no cloud and no accelerator. Open source: oido-ctc-small-int8, oido-ctc-small-int4 and the code on GitHub.
• Ito, text to speech for the ESP32-S3. Streaming English speech on a microcontroller, 4.9 MB, no cloud. Verified in emulation; board timings are coming. Weights on Hugging Face (gated: accept the licence to download) and code on GitHub.
• Voice agents. Speech recognition, language model and voice in one hosted API, with turn-taking and noise suppression built in.

Run it where you need it

Cloud API, on-premises deployment, or on-device. The same small models cover all three.

Work with us

Free API key and docs at lokutor.com and docs.lokutor.com. Questions: contact@lokutor.com.

Website | Docs | GitHub | LinkedIn | X