Promote Sign in Sign up

Audio model

Phonon 1

Phonon 1 is an open speech recognition model for English. It downloads in 415 MB and runs on a laptop or a datacenter GPU, transcribing an hour of audio in about two and a half minutes. It was trained at 2.4 bits per weight from the start, making it a low bit, on device capable transcription model built on a quantized Qwen3 ASR base.

Fermion Research Audio Closed apache-2.0
Visit model website ↗ Hugging Face
Type
Audio
Weights
Closed
Downloads
262
Likes
9
Pipeline
automatic speech recognition
License
apache-2.0

Model hub

FermionResearch/Phonon-1

Repo
FermionResearch/Phonon-1
Library
mlx
Access
Public
Technical tags and languages
en mlx apple-silicon speech-to-text asr stt low-bit ternary quantization-aware-training on-device streaming automatic-speech-recognition base_model:Qwen/Qwen3-ASR-0.6B
Open on Hugging Face

More from Fermion Research

3

About Fermion Research

1

Fermion Research builds compute-efficient foundation models, including the Neutrino language models and the Phonon-1 open speech recognition model, along with ternary quantization methods and inference runtimes for CUDA, Apple silicon, and x86.