nyra-FA Ranks First on FA-Bench Forced Alignment
nyra labs' forced aligner scores a word timestamp F1 of 0.688 on FA-Bench, ahead of Montreal-FA at 0.658, and the two intervals do not overlap. Its largest margin is on noisy conversational speech.
FA-Bench Leaderboard · github.com/olewave/fa-bench · Paper on arXiv
nyra-FA, the forced aligner from nyra labs, the team behind CrisperWhisper, joined the FA-Bench forced alignment word board on November 10, 2026, at rank 1. It scores a word timestamp F1 at 20 ms of 0.688, with a 95% interval of 0.677 to 0.698. Montreal-FA comes second at 0.658, from 0.650 to 0.667. The two intervals do not overlap, and nyra-FA's rank range is 1 to 1 among 16 forced aligners.
The board runs nyra_forced_aligner_en with nyra-forced-aligner 0.2.4. Its
weights are published under the
nyra health Non-Commercial Research License.
Research and other non-commercial use is free. Commercial use of the model, or
of the timestamps it produces, needs a commercial license from nyra health.
Where the gap is
The largest gap is on Buckeye conversation with noise added, 0.674 against 0.594. On clean Buckeye it is 0.731 against 0.706. On clean TIMIT, Montreal-FA keeps a slightly higher F1, 0.685 against 0.683. It also keeps a lower word boundary error there, the average distance from the hand-labelled boundaries, at 21.8 ms against 27.1 ms.
All the numbers
| TIMIT clean | TIMIT noisy | Buckeye clean | Buckeye noisy | Mean of 4 | |
|---|---|---|---|---|---|
| Word timestamp F1, nyra-FA | 0.683 | 0.663 | 0.731 | 0.674 | 0.688 |
| Word timestamp F1, Montreal-FA | 0.685 | 0.649 | 0.706 | 0.594 | 0.658 |
| Word boundary error, nyra-FA | 27.1 ms | 30.0 ms | 20.2 ms | 31.4 ms | |
| Word boundary error, Montreal-FA | 21.8 ms | 30.2 ms | 21.3 ms | 35.6 ms |
How FA-Bench scores
Each aligner gets the audio and its reference transcript, so only timing is scored. A word's start and its end each count as a hit when they land within 20 ms of the boundary a linguist labelled by hand. The test sets are TIMIT core test and Buckeye test, each clean and with reverb, noise, music and babble added.
FA-Bench ran nyra-FA on all 20 test cells. Thank you to Laurin Wagner and the nyra labs team for testing in the open.
Have an aligner or an ASR API? Send a pull request with an adapter. FA-Bench runs every test cell again and publishes the result.
Olewave makes both FA-Bench and Olign, so Olign is listed unranked at the bottom of the table.