whisper.cpp
High-performance Whisper speech recognition in plain C/C++ — no dependencies.
About
whisper.cpp is a high-performance port of OpenAI's Whisper automatic speech recognition model in plain C/C++, with zero runtime dependencies. It's the engine behind a huge share of local transcription tooling.
It's aggressively optimized: first-class Apple Silicon support (ARM NEON, Accelerate, Metal, Core ML), AVX/VSX intrinsics for x86/POWER, mixed F16/F32 precision, integer quantization, Vulkan and NVIDIA/AMD GPU support, and zero memory allocations at runtime. MIT licensed, very actively maintained.
What it can do
- Whisper ASR with no dependencies
- Quantized models for low RAM use
- GPU acceleration (CUDA, Metal, Vulkan, ROCm)
- Streaming and server examples
- Bindings for Python, JS, Go, Rust and more
Setup
- Clone the repo and build:
cmake -B build && cmake --build build --config Release. - Download a model, e.g.
./models/download-ggml-model.sh base.en. - Transcribe:
./build/bin/whisper-cli -m models/ggml-base.en.bin -f your-audio.wav. - For apps, link the library or use one of the many bindings.
Prerequisites
A C++ toolchain and CMake to build; prebuilt binaries and bindings exist for most platforms.
Use cases
Works with
Trust score — 9/10
Scored 9/10 (verified tier, metrics-v1). Evidence: 54.3k stars — the foundation of local speech-to-text; pushed 2026-10-12 — very active. Manual review: zero-dependency C/C++ with broad hardware coverage (Metal, CUDA, Vulkan, ROCm, Core ML); quantization and zero-alloc runtime are real engineering. License: MIT.
★ 54,332 GitHub stars
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 9.0/10
- Maintenance · recency, releases, issue load
- 10.0/10
- Openness · license + docs
- 9.0/10
- Manual review · sanity read + red-flag check
- 9.0/10
- · 54.3k stars — the foundation of local speech-to-text
- · pushed 2026-10-12 — very active
- · manual review: zero-dependency C/C++ with broad hardware coverage (Metal, CUDA, Vulkan, ROCm, Core ML); quantization and zero-alloc runtime are real engineering
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.