VoiceStudio — audio.cpp Engine (Breeze-TTS-2)

audio.cpp is a pure-C++ ggml audio inference framework (prebuilt binaries for Windows/macOS/Linux, CUDA / HIP / Vulkan / Metal / CPU, no Python dependency). VoiceStudio drives its audiocpp_server over loopback HTTP — v1 serves the breeze_tts family: Breeze-TTS-2 (BreezeBlue, 3B params, English + Chinese, voice clone + voice design + voice direction, 24 kHz).

Opt-in, and never a default. Select audiocpp explicitly in Model Catalogue → Engines (or OMNIVOICE_TTS_BACKEND=audiocpp).

License — read before enabling

audio.cpp code: Apache-2.0.
Breeze-TTS-2 weights (upstream BreezeBlue/Breeze-TTS-2 and the audio-cpp/audio.cpp-gguf GGUF repack): research and non-commercial use only under the BreezeBlue Research and Non-Commercial License. Self-hosted outputs inherit the restriction; a BreezeBlue paid subscription covers hosted-platform outputs only, not this engine.

Platform support

Host	Binary	Compute
Windows x64	prebuilt (Vulkan default; CUDA 12.4/13.3 opt-in)	GPU · CPU
Linux x64	prebuilt (Vulkan default)	GPU · CPU
macOS arm64 / x64	prebuilt (Metal)	GPU · CPU
Linux aarch64	none upstream	unavailable in v1

Notes:

Upstream ships no Linux-CUDA prebuilt. The Vulkan prebuilt runs on NVIDIA/AMD/Intel GPUs; for maximum CUDA speed, self-build audio.cpp with ENGINE_ENABLE_CUDA=ON and point OMNIVOICE_AUDIOCPP_BIN at it.
The Windows CUDA zips additionally need the matching cudart archive extracted next to the binaries (upstream packaging, not VoiceStudio).
VRAM: Q8_0 GGUF is ≈ 4.73 GiB plus graph/session workspace; 6 GB+ GPU memory is required.

Install

Download the v0.7.2 prebuilt for your platform from audio.cpp releases (CPU/Vulkan ≈ 20–70 MB; Windows CUDA ≈ 270 MB + cudart) and extract it.
Set OMNIVOICE_AUDIOCPP_BIN to the audiocpp_server binary (audiocpp_server.exe on Windows):
```
# macOS / Linux
echo 'export OMNIVOICE_AUDIOCPP_BIN=$HOME/apps/audio.cpp/audiocpp_server' >> ~/.zshrc
source ~/.zshrc
```
Alternatively set OMNIVOICE_AUDIOCPP_DIR to the directory containing it.
Restart VoiceStudio. The ~4.73 GiB breeze-tts-2-q8_0.gguf downloads from audio-cpp/audio.cpp-gguf (not gated) into the shared HF cache on first generate — resumable, hash-verified.
Pick audiocpp in Model Catalogue → Engines. The server starts lazily on first generate (server.json + server.log live under the app data audiocpp/ directory).

Voice modes

All three go through the one speech endpoint — reference presence selects:

Clone: ref_audio + ref_text (exact transcript, as upstream).
Direction: ref_audio + ref_text + instruct (e.g. "Speak slowly with a restrained, serious tone").
Design: description (or instruct) with no ref_audio (e.g. "A warm, thoughtful young woman…"). Upstream strengthens instruction-following with guidance_scale ≈ 4.

Optional env knobs

Variable	Default	Purpose
`OMNIVOICE_AUDIOCPP_BIN`	—	Absolute path to `audiocpp_server`.
`OMNIVOICE_AUDIOCPP_DIR`	—	Directory containing `audiocpp_server`.
`OMNIVOICE_AUDIOCPP_MODEL`	pinned auto-download	GGUF file or directory override.
`OMNIVOICE_AUDIOCPP_PACKAGE`	`breeze-tts-2-q8_0.gguf`	Package filename (`…-bf16.gguf` for full precision).
`OMNIVOICE_AUDIOCPP_BACKEND`	`metal` (macOS), `vulkan` (Win/Linux), `cpu`	Server compute backend.
`OMNIVOICE_AUDIOCPP_PORT`	`17860`	Loopback port.

Common errors

`audiocpp_server not found ...`

The binary isn't installed. Follow Install — the message carries the exact release URL and SHA for your platform.

`audiocpp_server exited during startup ...`

Usually a backend the binary wasn't built with, or a taken port. Check the server.log next to server.json in the app data audiocpp/ directory; set OMNIVOICE_AUDIOCPP_BACKEND=cpu as a fallback.

`Breeze-TTS-2 package ... missing after download`

The upstream audio.cpp-gguf layout changed. File an issue with the package listing — the allow-list in bootstrap.py needs updating.

audio.cpp runs as a managed native server (no Python venv, no transformers conflict). Only the downloaded GGUF counts toward sidecar disk usage.

4.6 KiB Raw Blame History Unescape Escape