C++ speech recognition inference engine using GGML — FunASR SenseVoice model, CPU/CUDA GPU, real-time microphone streaming, single GGUF model file, no Python dependency