whisper.cpp

An open-source speech-to-text engine that runs Whisper locally on desktop, mobile and in browsers, with CPU-only inference and optional GPU acceleration.

whisper.cpp runs OpenAI's Whisper speech recognition models on your own hardware. It's for developers adding transcription to applications and people who need offline speech-to-text.

Local transcription: Process speech on-device without sending audio to a cloud transcription service, including fully offline use on iPhone.
Application integration: A C-style API lets developers embed speech recognition in desktop, mobile and browser applications.
Speech detection: Voice activity detection identifies the parts of an audio input that contain speech; an example application also demonstrates an offline voice assistant.
Hardware acceleration: Run inference on a CPU or use NVIDIA CUDA, AMD ROCm or Vulkan acceleration; Apple Silicon supports Metal and Apple Neural Engine encoder processing through Core ML or ANEForge.
Smaller models: Integer quantization reduces model memory requirements, which can help when running on constrained hardware.

The project is open source under the MIT license. Supported platforms include macOS, Windows, Linux, FreeBSD, iOS, Android and Raspberry Pi, with WebAssembly for browser use and Docker support for server deployments. A GPU isn't required. OpenVINO can accelerate encoder processing on compatible Intel CPUs and GPUs, while supported AMD Ryzen AI processors can use their NPU.