Phonon-2

Phonon-2 is the most accurate open speech recognition model for English under 900 MB. Across the Open ASR Leaderboard's seven English sets it averages 5.21 % word error, and every open model that scores better is at least 5.8 times its size. Set for set it holds the accuracy of its 2.5 GB full-precision teacher, reaching 100.8 % of the teacher's word accuracy on parliamentary speech and beating it on meetings, from a download 15 times smaller. Its encoder holds each weight at one of five learned levels in about 2.1 bits.

It transcribes an hour of audio in about 20 seconds on an M5 MacBook Air (174x realtime), at 143x on eight Zen 5 cores (16 vCPU) and at 6,680x on one H100 in batches of 128.

Benchmarks

† Open ASR Leaderboard's published row; the other rows use its code on the full test sets. * Size from the parameter count at 16 bits.

Run it

Phonon-2 is the model inside Detta, the dictation app for the Mac.

On Apple silicon, from the command line:

pip install fermion-research
pip install mlx mlx-audio mlx-lm soundfile scipy zstandard
fermion transcribe recording.wav

The same package runs on Linux (x86-64 and Arm) and Windows CPUs; the engines are at github.com/fermionresearch/phonon. With Docker, on a CPU or a GPU:

docker run --rm -v "$PWD":/audio ghcr.io/fermionresearch/phonon-cpu:2.0.2 transcribe /audio/recording.wav --model phonon-2
docker run --rm --gpus all -v "$PWD":/audio ghcr.io/fermionresearch/phonon-cuda:1.0.3 transcribe /audio/recording.wav --model phonon-2

Notes

Based on parakeet-tdt-0.6b-v3 by NVIDIA; the tokenizer and output conventions (punctuation, casing, numerals) are the original's. Licence CC-BY-4.0, same as the original; NOTICE lists the changes. The command line and this repository's code are Apache 2.0.