phonon-31m English for llama.cpp
phonon-31m-en converted for
llama.cpp's llama-tts and llama-server.
It is the same model; only the file format differs.
These files need a llama.cpp build with Phonon support, which is not merged yet.
For the Phonon runtime (command line, Python, browser, iOS, Android), use gradium/phonon-31m-en. For audio.cpp, use gradium/phonon-31m-en-audiocpp.
Files
| File | What it holds |
|---|---|
phonon-31m-en-Q8_0.gguf |
The flow-matching transformer, 8-bit, with the tokenizer and the default voice. |
mmproj-phonon-31m-en-F32.gguf |
The flow head and the Mimi decoder, at full precision. |
llama.cpp needs both: the first is the model (-m), the second its projector (-mm).
Quickstart
llama-tts -hf gradium/phonon-31m-en-llamacpp -p "Hello from Phonon." -o speech.wav
With the files downloaded:
llama-tts -m phonon-31m-en-Q8_0.gguf -mm mmproj-phonon-31m-en-F32.gguf \
-p "Hello from Phonon." -o speech.wav
As a server:
llama-server -hf gradium/phonon-31m-en-llamacpp --port 8080
curl -X POST http://127.0.0.1:8080/tts -H "Content-Type: application/json" \
-d '{"input": "Hello from Phonon."}' -o speech.wav
Good to know
- Voice: the default voice is built in, so no speaker file is needed. The model has no voice cloning from audio. The other voices of gradium/phonon-voices do not work in llama.cpp yet.
- Text: llama.cpp sends the text to the tokenizer as written. Write numbers, symbols and
abbreviations out in words. The Phonon runtime does this for you with
--lang en. - Characters: letters the tokenizer has no piece for, and emoji, reach the model as an unknown token. Prefer plain text.
- Sampling: temperature 0.3, the Phonon runtime's default.
License
Apache 2.0, like phonon-31m-en. Phonon builds on Pocket TTS and the Mimi codec, developed by Kyutai.