audio.cpp: C++/ggml runtime for 12 audio models, up to 5x faster TTS

View on Reddit

Supports 12 audio model families including Qwen3-TTS and VeVo2. Claims up to 5x speedup over Python on CUDA in a single-file ggml-based runtime.