Accuracy Report

What the report will include

When the benchmark is published, it will cover:

• Corpora: real professional recordings from external beta users (therapists, legal professionals, managers) plus standard open-source speech benchmarks
• Device and model version labeled: results will specify the iPhone model, iOS version, and Whisper model used
• Two conditions: clean-room (quiet, 2 speakers, near phone) and adverse (background noise, 3+ speakers, accented speech) — we will not publish only the favorable number
• Speaker attribution accuracy (DER): not just word accuracy, but how reliably Lucy assigns words to the correct speaker
• Explicit limitations: what the benchmark does not cover (highly accented non-English speech, 5+ speakers, phone-quality audio)
• Published date and model version: so you can see when the numbers were measured and whether they still apply to your version

Why we are waiting

Every competitor in this space either publishes no accuracy number or publishes a self-reported round number with no methodology. We think you deserve better than that — particularly at the price point Lucy commands.

We are waiting because the only numbers worth publishing are ones measured on real professional audio by real users, with a test methodology we are willing to defend. That requires external beta testing to complete first.

No accuracy claim appears anywhere in Lucy’s App Store listing or marketing materials. That is intentional. When this page is populated, it will be linked from the app and from the landing page.

When

We are targeting publication within 30 days of the v1.0 launch, once sufficient real-world data from external testers has been collected and reviewed. Subscribe to the mailing list or check back here.

Questions? support@xymex.com