Teach Rust to listen.

Run small wake-word models locally, from a microphone or any 16 kHz PCM stream. No cloud. No audio uploads.

Get started See how it works

listener.rs

listening on Default Microphone

detectedhey foxy92.4%

local by defaultstreaming detectiontwo APIs high + low level

Start with what you have

A microphone

Listener handles capture, resampling, detection, and cooldowns for you. Build your first listener

An audio stream

Detector takes 10 ms PCM blocks from files, calls, sockets, or your own audio engine. Bring your own audio

A wake-word model

Load microWakeWord JSON directly, or describe a standalone .tflite model in Rust. Set up your model

One pipeline, two ways in

Microphone  ->  Listener  --+
                           +-->  Detector  -->  Detection
Your audio  ->  PCM blocks --+

The detector converts every 10 ms audio block into microWakeWord frontend features, evaluates the TensorFlow Lite model, smooths recent probabilities, and reports a match when the configured cutoff is reached.

Listen in a few lines

cargo add micro-wakeword

use micro_wakeword::Listener;

fn main() -> Result<(), Box<dyn std::error::Error>> {
    let mut listener = Listener::from_config("wake-word.json")?;

    while let Some(hit) = listener.next_detection()? {
        println!("Heard {} ({:.1}%)", hit.wake_word, hit.probability * 100.0);
    }
    Ok(())
}

Keep the model JSON and .tflite file together. That is enough to start listening.