Breeze TTS 2 is available in audio.cpp release 0.7.2. 6GB VRAM 4x faster than realtime. No Python. Demo included
Breeze TTS 2 is available in audio.cpp release 0.7.2.
RTF = 0.21 to 0.24 on RTX 5090 (4x realtime). Longform (6000+ char) RTF = 0.23. Peak VRAM is about 6GB.
If you’re tired of wrestling with Python environments and dependencies just to try an audio model, give audio.cpp a try!
Star or watch the repo so you don’t miss important updates!
Wow, this is really awesome ! Thanks for posting this !
Does audio.cpp support using cloning with direction on how say it like emotions using natural language because the base model supports that
Does audio.cpp support using cloning with direction on how say it like emotions using natural language because the base model supports that
Yes. Check https://github.com/0xShug0/audio.cpp/blob/dev/docs/models/breeze_tts.md. On the UI you need to click the model parameter section and input your instructions there
Does audio.cpp support using cloning with direction on how say it like emotions using natural language because the base model supports that
Yes. Check https://github.com/0xShug0/audio.cpp/blob/dev/docs/models/breeze_tts.md. On the UI you need to click the model parameter section and input your instructions there
{
"host": "127.0.0.1",
"port": 8787,
"backend": "cuda",
"device": 0,
"threads": 1,
"lazy_load": true,
"models": [
{
"id": "breeze_tts",
"family": "breeze_tts",
"path": "models/Breeze-TTS-2-GGUF/breeze-tts-2-q8_0.gguf",
"task": "tts",
"mode": "offline",
"load_options": {
"language": "english"
},
"session_options": {
"language": "english"
}
},
{
"id": "qwen3-asr",
"family": "qwen3_asr",
"path": "models/Qwen3-ASR-0.6B",
"task": "asr",
"mode": "offline"
}
]
}
Why isn't it working?
"path": "models/Breeze-TTS-2-GGUF",
"task": "clon",
or:
"path": "models/",
"task": "clon",
audiocpp_server --config server.json
v0.7.1 , Even so, the model file in the following path cannot be scanned: models\Breeze-TTS-2-GGUF\breeze-tts-2-q8_0.gguf
Could you share your server.json file? Why can't I scan the gguf model in the models?
@sunnyboxs It's in the dev branch...release prebuilts are based on main.
Update: now you can try prebuilts v0.7.2
This is a great post, but the license of the model keeps it from really being used. There is no value in porting it if you don't own the outputs nor can you use it for commercial or production use.