Is Accentus state of the art?Pending benchmarks.Announced Nov 5, 2026; no independent results yet.
Luel's model for detecting American English in real-world speech used for speech-data QA. It returns a band (Pass / Review / Reject), not a percentage, and on Reject it names a non-American group. Built for conversation, phone lines and room noise rather than studio read-aloud. Public model page and demos on Luel Lab; Luel's speech docs list accent identification as coming soon with no API endpoint yet. Weights not published.
The details so far
- Lab
- Luel
- Announced
- Nov 5, 2026
- API
- Not yet
- Open weights
- Closed · None published
- Role
- American-English accent detection for speech-data quality assurance; outputs Pass, Review or Reject (with a named group on Reject)
- Input
- Real conversation, phone-line and room-noise audio; under 10 seconds of detected speech returns no score
- Output
- A band, not a confidence percentage; Review means a person should listen and is not treated as a reject
- Access
- Public lab page and held-out demos at luel.work/lab/models/accentus; no public API endpoint yet (Luel docs: accent identification coming soon on the same key/auth)
- Related
- Separate from Luel Accent Identification v1 (closed variety checklist) and from Luel Language ID
Tags marked unsure are not stated on the lab's own pages.
What would let us call it
Verdicts use only independent boards we already track. Two consecutive snapshots are required before any crown.
- Publicly announced with a source
- Usable by the public (API or weights)
- Listed on an independent board we track
- Enough results for a rank in a category
- Holds across two snapshots (crowns only)
What the lab claims
Self-reported. Shown for context; not used in any verdict.
Lab claim: in real conversation it rejects under 2% of Americans and names the accent it heard correctly 96% of the time (Luel's announcement).
Lab claim: on held-out real conversation, 5 of 233 Americans rejected; on phone-line audio 2 of 217; with room noise 6 of 218; Americans rejected stay under 6% on every channel in the lab's table.
Lab claim: on Luel's US Accent board (233 Americans / 719 others, original recording), Accentus stops 97.9% of non-Americans while keeping 95.0% of Americans (lab board).
Sources
LABAccentus model page (Luel Lab)luel.work/lab/models/accentus · Nov 5, 2026LABUS Accent benchmark (Luel Lab)luel.work/lab/benchmarks/us-accent · Nov 5, 2026LABLuel docs: Formats & limits (accent identification coming soon)luel.work/docs/speech/formats · Nov 5, 2026LABLuelCompany on X: Accentusx.com/LuelCompany/status/2118402973244215309 · Nov 5, 2026Tracking ends Feb 3, 2027 unless the API or weights become available.