whisper.cpp

A dependency-free C/C++ port of OpenAI's Whisper that transcribes speech on CPUs, Apple Silicon, GPUs and phones.

  • Run locally
  • MIT · Open source
  • macOS, Windows, Linux
  • Verified

whisper.cpp is a plain C/C++ implementation of OpenAI's Whisper automatic speech recognition model. It has no dependencies and treats Apple Silicon as a first-class target through ARM NEON, the Accelerate framework, Metal and Core ML. It also supports AVX on x86, Vulkan, NVIDIA GPUs, AMD ROCm, OpenVINO and several NPUs, integer quantization, and CPU-only inference.

According to its README it runs on macOS, iOS, Android, Linux and FreeBSD, Windows, WebAssembly and Raspberry Pi, and it comes with examples such as an iPhone app and a command-line voice assistant, plus voice activity detection. It belongs to the ggml family of projects, like llama.cpp. You download Whisper models converted to the ggml format and run them from the command line or through a C-style API.

It runs Whisper models, and the license is MIT. Newer speech-recognition models, such as those on our speech recognition list, ship with their own runtimes, so check each model card before assuming whisper.cpp supports it.

Good for

  • Offline transcription on a laptop, phone or Raspberry Pi
  • Adding speech-to-text to a C or C++ application
  • Fast transcription on Apple Silicon with Metal or Core ML

Watch out for

  • It runs Whisper models only, so check a model's card for supported runtimes before choosing another speech model
  • Whisper can write text that was never spoken, so check transcripts that matter
  • Models have to be downloaded in the ggml format that the project provides

Links