Model reference · open weights
phonon-en-audiocpp is an open-weight audio or speech model from gradium. phonon-31m-en-audiocpp (BF16) weighs 85 MB; the smallest configuration that runs it is RTX 3060 12 GB.
For self-hosting, compare this build's requirements below with the GPU you plan to rent.
Summary of the gradium/phonon-31m-en-audiocpp model card, 2026-11-02
What it is
| Released by | gradium |
|---|---|
| Released | 2026-10-23 |
| VRAM | 85 MB for the weights |
What it runs on
| Card | One stream | Memory |
|---|---|---|
| RTX 3060 12 GB | fits | 11.6 GB |
| RTX 4060 Ti 16 GB | fits | 15.4 GB |
| RTX 3090 24 GB | fits | 23.4 GB |
| RTX 4090 24 GB | fits | 23.4 GB |
| RTX 5090 32 GB | fits | 31.0 GB |
| L40S 48 GB | fits | 44.0 GB |
| A100 80 GB | fits | 78.2 GB |
| H100 80 GB | fits | 78.1 GB |
| RTX PRO 6000 Blackwell 96 GB | fits | 93.8 GB |
| DGX Spark (GB10) 128 GB unified | fits | 107 GB |
| H200 141 GB | fits | 138 GB |
| B200 180 GB | fits | 176 GB |
From the model card
phonon-31m-en converted for
audio.cpp's gradium_phonon family.
It is the same model; only the file format differs.
These files need an audio.cpp build with the
gradium_phononfamily, which is not merged yet. They do not load in llama.cpp: for llama.cpp, use gradium/phonon-31m-en-llamacpp.
For the Phonon runtime (command line, Python, browser, iOS, Android), use gradium/phonon-31m-en.
| File | What it holds |
|---|---|
phonon-31m-en-q8_0.gguf | The model, with its config and tokenizer. The transformer's linear layers are 8-bit; the rest, the Mimi decoder included, keeps the original precision. |
default-voice.safetensors | The default voice, the same file as in phonon-31m-en. |
From an audio.cpp checkout:
python3 tools/model_manager_v2.py install gradium_phonon_31m_en_q8_0 --models-root models
audiocpp_cli --task tts --family gradium_phonon --model models/phonon-31m-en \
--backend cpu --text "Hello from Phonon." --out speech.wav
--backend selects another audio.cpp backend, such as metal, and --mode streaming
streams the audio.
--voice-id, the default voice is used. More voices are in
gradium/phonon-voices: copy one into the
model folder as embeddings/.safetensors and pass --voice-id . The model has
no voice cloning from audio.--lang en.Apache 2.0, like phonon-31m-en. Phonon builds on Pocket TTS and the Mimi codec, developed by Kyutai.
Quoted from the model card on Hugging Face. The full card is behind the Hugging Face link above.
How it works