whisper.cpp Download: Fast Whisper Speech Recognition in C/C++
whisper.cpp 1.9.5 · for Windows, macOS and Linux
Developer: ggml (opens in a new tab)
- Windows
- macOS
- Linux
The download link points to the program’s official source; the file is not hosted on this site. Link address: github.com
Link not working? Let us know
whisper.cpp is a high-performance C/C++ port of OpenAI’s Whisper speech recognition model with no dependencies, developed by the ggml team. The latest version is 1.9.5, released on October 6, 2026; it runs on Windows, macOS, Linux and many other platforms and is MIT-licensed.
whisper.cpp brings OpenAI’s Whisper to plain C/C++, without Python or PyTorch. That makes it fast, small and easy to embed, from desktop apps to phones and a Raspberry Pi. It comes from the same ggml team as llama.cpp.
What does whisper.cpp do?
- Local transcription: Turns speech into text using Whisper models in the GGML format.
- Efficient: Plain C/C++ with no dependencies, integer quantization and zero memory allocations at runtime.
- Many back ends: Apple Metal and Core ML, NVIDIA CUDA, Vulkan, AMD ROCm, Intel OpenVINO and more, plus CPU-only use.
- Ready-made tools: whisper-cli for files, whisper-server with an HTTP API, and whisper-stream for live microphone input.
- Extras: Voice activity detection, plus experimental word-level timestamps and speaker segmentation.
When to choose whisper.cpp
It suits developers who want to build speech recognition into their own software, and anyone who wants fast offline transcription on modest hardware.
Building whisper.cpp
Clone the repository, download a model and build with CMake:
git clone https://github.com/ggml-org/whisper.cpp.git
cd whisper.cpp
sh ./models/download-ggml-model.sh base.en
cmake -B build
cmake --build build -j --config Release
./build/bin/whisper-cli -f samples/jfk.wav
whisper-cli currently reads only 16-bit WAV files, so convert other formats first:
ffmpeg -i input.mp3 -ar 16000 -ac 1 -c:a pcm_s16le output.wav
Docker images are also published at ghcr.io/ggml-org/whisper.cpp, including CUDA and Vulkan variants. To build with NVIDIA support yourself, add -DGGML_CUDA=1 to the first cmake command.
Model sizes and memory
| Model | Disk | Memory |
|---|---|---|
| tiny | 75 MiB | ~273 MB |
| base | 142 MiB | ~388 MB |
| small | 466 MiB | ~852 MB |
| medium | 1.5 GiB | ~2.1 GB |
| large | 2.9 GiB | ~3.9 GB |
It runs on macOS (Intel and Apple Silicon), Windows, Linux, FreeBSD, iOS, Android, WebAssembly and Raspberry Pi. A GPU is optional.
whisper.cpp is free and open source under the MIT license.
Pros
- Fast and light, no Python required
- Runs on CPUs, many GPUs and small devices
- Includes a server and live microphone tool
Cons
- Usually has to be compiled from source
- The CLI only reads 16-bit WAV files
Frequently Asked Questions
What is the difference between Whisper and whisper.cpp?
Whisper is OpenAI’s original Python implementation. whisper.cpp runs the same models in plain C/C++, which makes it lighter, faster on ordinary hardware and easier to embed in other software.
How much memory does whisper.cpp need?
From about 273 MB for the tiny model to about 3.9 GB for the large model, according to the project’s memory table.
whisper.cppOpen source
Download (opens in a new tab)
