audio.cpp

Why we included this project

audio.cpp runs a broad mix of audio models over a single compiled binary, so you get text-to-speech, speech-to-text, voice cloning, VAD, and music generation in one place without standing up a Python stack. Because everything is built on ggml, the same model files move between NVIDIA, AMD, Apple Silicon, and CPU-only machines, which helps on laptops and edge hardware where a heavyweight Python setup is impractical. The Arena UI is handy when you want to compare local models side by side, and the HTTP server mode makes it straightforward to feed outputs into your own application. The speedups over the Python reference paths and the GGUF quantized loading options mean you can usually simplify deployment without giving up much quality.