| --- |
| license: apache-2.0 |
| base_model: Qwen/Qwen3-ASR-0.6B |
| base_model_relation: quantized |
| language: |
| - en |
| library_name: mlx |
| pipeline_tag: automatic-speech-recognition |
| tags: |
| - mlx |
| - apple-silicon |
| - speech-to-text |
| - asr |
| - stt |
| - low-bit |
| - ternary |
| - quantization-aware-training |
| - on-device |
| - streaming |
| metrics: |
| - wer |
| --- |
| |
| # Phonon-1 Micro |
|
|
| Phonon-1 Micro is an open speech recognition model for English, the smallest |
| build of the Phonon-1 family. It downloads in 285 MB, runs on a laptop or a |
| datacenter GPU, and beats Moonshine base on all eight benchmarks below. It was |
| trained at 2.4 bits per weight from the start. |
|
|
| ## Benchmarks |
|
|
| | Benchmark | Phonon-1 (415 MB) | **Phonon-1 Micro (285 MB)** | Parakeet-0.6B 4-bit (637 MB) | Moonshine base (248 MB) | Whisper large-v3-turbo (1,619 MB) | Whisper small (967 MB) | wav2vec2-large (1,262 MB) | Qwen3-ASR teacher (1,569 MB) | |
| |---|---:|---:|---:|---:|---:|---:|---:|---:| |
| | LibriSpeech test-clean | 2.640 | **3.002** | 2.186 | 3.417 | 2.10 | 3.4† | 2.8† | 2.235 | |
| | LibriSpeech test-other | 5.699 | **6.511** | 3.937 | 8.262 | 4.07 | 7.6† | 6.3† | 4.618 | |
| | TED-LIUM | 3.421 | **3.878** | 2.829 | 5.272 | — | — | — | 2.889 | |
| | SPGISpeech | 4.163 | **4.858** | 4.104 | 5.731 | 2.79† | — | 13.31† | 3.074 | |
| | VoxPopuli | 8.394 | **9.177** | 6.345 | 10.470 | 11.22† | — | — | 7.151 | |
| | GigaSpeech | 11.396 | **11.882** | 9.614 | 12.114 | 8.52† | — | — | 9.321 | |
| | Earnings-22 | 12.571 | **14.771** | 11.190 | 17.872 | 11.07† | — | 36.28† | 11.188 | |
| | AMI | 13.084 | **14.094** | 12.723 | 17.790 | 15.16† | — | — | 12.560 | |
| | Macro (eight benchmarks) | 7.67 | **8.52** | 6.62 | 10.1 | — | — | — | 6.63 | |
|
|
| 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. |
|
|
| ## Run it |
|
|
| ```bash |
| pip install fermion-research |
| |
| fermion transcribe --model FermionResearch/Phonon-1-Micro recording.wav |
| ``` |
|
|
| Or serve an OpenAI-compatible endpoint: |
|
|
| ```bash |
| fermion serve --model FermionResearch/Phonon-1-Micro |
| |
| curl -s http://127.0.0.1:8000/v1/audio/transcriptions \ |
| -F "file=@recording.wav" \ |
| -F "model=FermionResearch/Phonon-1-Micro" |
| ``` |
|
|
| The same weights run on a Mac (via MLX) or an NVIDIA GPU; the CUDA runtime and |
| Docker image are in the [GitHub repo](https://github.com/fermionresearch/phonon). |
|
|
| ## License |
|
|
| **Apache License 2.0** for the weights and the [command line](https://pypi.org/project/fermion-research/). Base model: [`Qwen/Qwen3-ASR-0.6B`](https://huggingface.co/Qwen/Qwen3-ASR-0.6B), Apache-2.0. |
|
|
