whisper.cpp — A high-performance C/C++ implementation for inference of OpenAI's Whisper automatic speech recognition (ASR) model, designed for lightweight and cross-platform deployment.

A high-performance C/C++ implementation for inference of OpenAI's Whisper automatic speech recognition (ASR) model, designed for lightweight and cross-platform deployment.

About this project

whisper.cpp is a plain C/C++ port of OpenAI's Whisper model that enables high-performance speech-to-text inference without heavy dependencies. It is designed for efficiency, featuring zero memory allocations at runtime and support for mixed F16/F32 precision and integer quantization. Key capabilities include: - Extensive Hardware Acceleration: Optimized for Apple Silicon (ARM NEON, Metal, Core ML), x86 (AVX), NVIDIA GPUs (CUDA), AMD GPUs (ROCm), Vulkan, OpenVINO, and various NPUs (AMD Ryzen AI, Ascend). - Broad Platform Support: Compatible with macOS, iOS, Android, Linux, FreeBSD, Windows, WebAssembly, and Raspberry Pi. - Advanced Features: Includes Voice Activity Detection (VAD), real-time audio input streaming, word-level timestamps, and confidence color-coding for transcriptions. - Integration: Provides a C-style API and Docker images for easy deployment across different environments.