Installation
RFWhisper runs on Linux, macOS, Windows, and Raspberry Pi 5. Pick your platform below.
| Platform | Minimum | Recommended | Notes |
|---|---|---|---|
| Linux | Any x86_64/aarch64 from 2015+, ALSA or PipeWire | Ubuntu 22.04+ / Fedora 38+ | Best DX for hacking |
| macOS | Intel or Apple Silicon, macOS 12+ | Apple Silicon M1+ | CoreML provider gives you a free perf boost |
| Windows | Windows 10/11 x64 | Windows 11 with DirectML | VB-Cable required for virtual routing |
| Raspberry Pi | Pi 5 (4 GB) + active cooling | Pi 5 8 GB + NVMe HAT | RNNoise fallback for Pi 4 / Zero 2 W |
Shared prerequisites​
- Rust (stable toolchain via rustup)
- Git with
git-lfsfor samples/models - A working audio stack on your OS
- (For v0.2+) GNU Radio 3.10.x with
gr-soapy+gr-dnn
One-liner install (from source, alpha)​
git clone https://github.com/jakenherman/rfwhisper.git
cd rfwhisper
cargo build --release
./target/release/rfwhisper models fetch
./target/release/rfwhisper --version
The build produces a single self-contained binary at target/release/rfwhisper — copy it anywhere on your PATH (no interpreter or virtualenv needed).
Then jump to Quick Start to denoise your first file.
Verifying the install
rfwhisper doctor
Runs a diagnostic checklist: build info, ONNX Runtime providers, audio backend, model SHA-256s, and platform-specific quirks. Paste the output into any bug report.