whisper.cpp
whisper.cpp is a dependency-free C and C++ port of OpenAI's Whisper speech-recognition model, built to run efficiently from phones to servers with no external libraries.
It is optimized for Apple Silicon through ARM NEON, the Accelerate framework, Metal and Core ML, adds AVX support for x86 and further backends for POWER, Vulkan, NVIDIA and AMD ROCm GPUs, AMD Ryzen AI NPUs, OpenVINO, Ascend NPUs and Moore Threads GPUs, and supports mixed F16/F32 precision, integer quantization, zero runtime memory allocation, and built-in voice-activity detection.
It runs on macOS, iOS, Android, Linux, FreeBSD, WebAssembly, Windows and Raspberry Pi, and ships a C-style API plus a whisper-cli example that transcribes a downloaded ggml-format model against a 16-bit WAV audio file, with common quantized model sizes documented alongside their approximate memory use.
MIT licensed.