Voice AI

Voice AI.Models we train, benchmarks we built.

We fine-tune and train our own voice models, Hush and Cue, which run on the edge, and we built proprietary benchmarks to measure voice models properly.

A person putting on a headset microphone

Models we fine-tune and train

Saolabs fine-tunes and trains voice models that run on the edge: Hush for speech enhancement and Cue for turn detection. Doing that work ourselves means we know where voice models break, and what data fixes it.

Benchmarks we built

Public benchmarks miss a lot of what matters in real speech. We built proprietary benchmarks for voice models, and use them to evaluate our own models and others'.

What we offer

  • Voice model evaluation on our proprietary benchmarks
  • Expert speech data for training and fine-tuning
  • Expert review of voice model output
  • Custom benchmarks for your use case

Questions, answered.

Does Saolabs build its own AI models?

Saolabs fine-tunes and trains voice models and builds proprietary benchmarks for voice models. It does not build a general-purpose foundation model.

Can you evaluate our voice model?

Yes. We can run your model on our proprietary benchmarks, or build a custom benchmark for your use case, and report the results with examples.

Every safe model
has an expert behind it.