
Voice AI on speech foundation models we train ourselves
Lokutor is a speech AI company in Madrid. We design and train our own small speech foundation models: Versa (text to speech in 9 languages, incl. Catalan, Galician and Basque), Turno (turn-taking: knows when the caller is done and lets them interrupt) and Psst (noise suppression), with Conv 2.0 (speech recognition) in development. Small models make it fast and affordable: agents answer in about 0.6 s (596 ms median) for about 5¢ a minute, LLM included. Live demo on lokutor.com, no sign-up.
Lokutor is a Madrid-based speech AI company that develops its own small speech foundation models, including Versa for text-to-speech in nine languages, Turno for turn-taking, and Psst for noise suppression. The models enable fast and cost-effective voice interactions, with agents responding in a median of 596 milliseconds at approximately 5 cents per minute.
A source that found nothing is a measurement. A source that has not run is a gap. Neither means the launch lacks the thing.