| --- |
| 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 Big |
|
|
| This is the largest build of the Phonon-1 family, an open speech recognition |
| model for English that downloads in 581 MB. |
|
|
| ## Benchmarks |
|
|
| | Benchmark | Phonon-1 Big (581 MB) | |
| |---|---:| |
| | LibriSpeech test-clean | 2.667 | |
| | LibriSpeech test-other | 5.722 | |
| | TED-LIUM | 3.400 | |
| | SPGISpeech | 4.156 | |
| | VoxPopuli | 8.369 | |
| | GigaSpeech | 11.291 | |
| | Earnings-22 | 12.417 | |
| | AMI | 12.812 | |
| | Macro (eight benchmarks) | 7.604 | |
|
|
| Word error rate, lower is better. Measured by us — full test sets, Whisper |
| English text normalizer, greedy decoding. |
|
|
| ## Run it |
|
|
| ```bash |
| pip install fermion-research |
| |
| fermion transcribe recording.wav --model FermionResearch/Phonon-1-Big |
| ``` |
|
|
| Or serve an OpenAI-compatible endpoint: |
|
|
| ```bash |
| fermion serve --model FermionResearch/Phonon-1-Big |
| |
| curl -s http://127.0.0.1:8000/v1/audio/transcriptions \ |
| -F "file=@recording.wav" \ |
| -F "model=FermionResearch/Phonon-1-Big" |
| ``` |
|
|
| 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](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. |
|
|
