Early Access to the FFASR Leaderboard

See how your ASR model performs in the real world

Treble and Hugging Face are launching the FFASR Leaderboard — a new standard for evaluating speech recognition in far field conditions.

Distance. Reverberation. Environmental noise.
Not simulated shortcuts, but physically accurate acoustic scenarios.

If you are building voice AI, this is where performance differences become visible.

What is FFASR

A leaderboard built for real deployment conditions

Most ASR benchmarks measure performance in clean or near field setups.

The FFASR leaderboard evaluates models across:

  • Distance variation
  • Room acoustics and reverberation
  • Realistic environmental noise

All test data is generated using Treble’s wave based acoustic simulation engine, enabling controlled, repeatable, and physically accurate evaluation at scale.

Why it matters

Your model does not ship into a lab

Performance gaps often appear only after deployment.

The FFASR leaderboard exposes:

  • Degradation across distance
  • Sensitivity to reverberation
  • Robustness in noisy environments

This gives you a clearer signal than traditional benchmarks.

Launch webinar on June 11th

We will cover:

  • How the leaderboard is structured
  • What the evaluation scenarios look like
  • Early results and insights
  • How to submit and benchmark your model