Tempest

Poor man's linux voice automation.

Goals

• Wake phrase "Tempest rise" and sleep phrase "Tempest rest"
• Shortcut customization using config files
• Recording built into the binary
• Sending fuzzy questions to an LLM through Ollama's API
• Built-in understanding of fuzzy terms (similar sentences are treated as equal)
• Static builds

Getting started

Prerequisites:

• Rust toolchain, either via your package manager or rustup
• C++ standard library
• clang
• cmake
• pkg-config
• ONNX Runtime
• Audio library for your OS (linux distros have package names like alsa-dev, alsa-lib or alsa)

Luckily, if you use NixOS with flakes, you can run nix develop in the project directory to get a dev shell with all the dependencies installed.

Once you have installed the necessary tools, clone this repo.

git clone https://github.com/lavafroth/tempest
cd tempest

Change the bindings in the config file to suit your needs.

Run the following to build the daemon and the client:

cargo build --workspace --release

Daemon

The daemon is optional and is only needed if you want phrases in your bindings to perform keyboard shortcuts. Since performing keystrokes is a privileged action, you must run the daemon as root.

sudo ./target/release/tempest-daemon

This will give a token to authenticate with the daemon.

Client

If you have the daemon running in the background, in a different terminal tab, run

./target/release/tempest-client \
the_token_from_the_daemon

where the_token_from_the_daemon is the token provided by the daemon.

You can alternatively run the client as standalone. However, config bindings with keyboard shortcuts will not work.

./target/release/tempest-client

On first run, the client will prompt you to download the models.

Acknowledgements

A huge thank you to these folks in helping me build this tool:

• Josh Habdas for creating the original C FFI using bindgen.
• abb128 for the C implementation of April ASR.