Phonon-1

Phonon-1

Phonon-1 is an open speech recognition model for English. It downloads in 415 MB, runs on a laptop or a datacenter GPU, and transcribes an hour of audio in about two and a half minutes. It was trained at 2.4 bits per weight from the start, and it is the second model in the lab's low-bit lane after Neutrino-1.

Benchmarks

Word error rate, lower is better. Unmarked cells: measured by us — full test sets, Whisper English text normalizer, greedy decoding. † = published figure (model card, paper, or the Open ASR Leaderboard). Dash = no comparable measurement.

Median 23.9× realtime across nine corpora on a base M5 MacBook Air. That figure is decode time only, measured over utterances of up to 30 s with the model already loaded; the first run in a new environment compiles the runtime once (about 20 s), after which runs are fast.

Run it

pip install fermion-research
fermion transcribe phonon-1 recording.wav

fermion serve phonon-1
curl -s http://127.0.0.1:8000/v1/audio/transcriptions \
  -F "[email protected]" \
  -F "model=FermionResearch/Phonon-1"

The same weights run on a Mac (via MLX), on an NVIDIA GPU, or on a plain CPU; the runtimes and Docker images are in the GitHub repo.

License

Apache License 2.0 for the weights and the command line. Base model: Qwen/Qwen3-ASR-0.6B, Apache-2.0.