FluidVoice Labs
Models that run on your device.
We build small speech models for everyday dictation. Explore the weights, benchmarks, and research behind FluidVoice.
Our models
Speech recognitionEnglish · Speech to text
Fluid Speech Mini
An open-weights English speech model. Parakeet-level accuracy in a 253 MB Core ML package, trained for dictation in noisy rooms.
253 MB Core MLEnglish Speech to text
Explore the model 
Why on-device?
Your audio stays on your device, and dictation works without a connection. We publish our weights and evaluations so you can try the models in your own applications.