whisper.cpp
Efficient, offline Whisper speech recognition on almost any device
Overview
whisper.cpp is a C/C++ port of OpenAI's Whisper speech-to-text built on ggml, running locally on Apple, Android, Linux, Windows, Raspberry Pi and WebAssembly with Metal, CUDA, Vulkan and NPU acceleration, plus streaming, VAD, timestamps and Parakeet support. Free and MIT-licensed.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 1 Oct 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
What you can produce with whisper.cpp
- Local Whisper speech-to-text in C/C++
- Metal, Core ML, CUDA, Vulkan and NPU acceleration
- Real-time streaming transcription
- Word-level timestamps and karaoke video output
- Voice-activity detection and speaker-turn detection
- NVIDIA Parakeet model support
whisper.cpp is the go-to way to run Whisper transcription locally and privately on ordinary hardware, from iPhones and Macs to Raspberry Pi and the browser, with quantised models small enough for phones and acceleration on most chips. It is free, MIT-licensed and actively maintained, now under Hugging Face. For high-volume batch transcription on NVIDIA servers, batched GPU alternatives can be faster. An excellent default for offline and on-device speech-to-text.
What people say
whisper.cpp is a high-performance C/C++ implementation of OpenAI's Whisper speech-recognition models, built on the ggml machine-learning library. It runs Whisper locally on a wide range of hardware — Macs, iPhones, Android phones, Linux and Windows PCs, Raspberry Pi and in the browser through WebAssembly — on CPUs or on GPUs and accelerators through Metal, Core ML, CUDA, Vulkan, AMD's ROCm and Ryzen AI NPUs, Intel, Ascend and Moore Threads, and an option called ANEForge runs the encoder directly on Apple's Neural Engine. Quantised models shrink large-v3 from 2.9 GB to about 1.1 GB, and the turbo version to around 550 MB. Tools include a command-line transcriber with word-level timestamps and karaoke-style video output, real-time streaming transcription, voice-activity detection with Silero, speaker-turn detection with a tinydiarize model, and a simple HTTP inference server, along with official Docker images and bindings for many languages. Since June 2026 it also supports NVIDIA's Parakeet speech models. Releases come every few weeks; the latest, 1.9.4, was published on 11 September 2026.
The project began in September 2022 and is maintained under ggml-org by Georgi Gerganov's company ggml.ai, which Nat Friedman and Daniel Gross backed with pre-seed funding. Hugging Face took on ggml.ai in February 2026, and NVIDIA announced an agreement to acquire Hugging Face in September 2026.
Everything is free under the MIT licence, and OpenAI's Whisper weights are also MIT-licensed. There is no hosted service or paid tier.
On NVIDIA GPU servers, batched alternatives such as faster-whisper can deliver higher throughput, so whisper.cpp's real strength is efficient, private transcription on everyday devices. It suits developers who need offline speech-to-text in apps and on edge hardware.
Summary of public user & expert reviews, compiled by RECATOOLS.
Frequently asked questions
About this listing
This entry was compiled from publicly available data including whisper.cpp's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with whisper.cpp unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to whisper.cpp directly →
Spotted something out of date? Suggest an update →
More in Video & Audio
