whisper.cpp

Supports Apple Silicon with Metal/Core ML, x86 with AVX, ARM NEON, and WebAssembly, enabling efficient deployment from servers to mobile devices, as highlighted in the feature list.

Integrates with NVIDIA CUDA, Vulkan, OpenVINO, and specialized NPUs for GPU inference, detailed in sections like NVIDIA GPU and Vulkan support.

Uses zero runtime memory allocations, mixed F16/F32 precision, and optional quantization to reduce model size and speed up processing, mentioned in the Efficient Inference and Quantization sections.

Implemented in plain C/C++ with no dependencies, allowing minimal footprint and offline use on edge devices, as stated in the project philosophy.