Your first listener

Listener is the batteries-included API. It opens a microphone, converts its audio to the format the model expects, owns a Detector, and suppresses immediate repeats.

1. Add the code

use micro_wakeword::Listener;

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

    println!(
        "Listening for {}",
        listener.detector().config().wake_word
    );

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

    Ok(())
}

2. Run in release mode

cargo run --release

Speak the model’s wake phrase. A detection contains:

• wake_word: the label from your configuration;
• probability: the smoothed model score from 0.0 to 1.0.

What each call does

1. Listener::from_config(...) parses and validates JSON, resolves its model path, loads TensorFlow Lite, opens the current default input, and starts the audio stream.
2. next_detection() waits for microphone audio and processes it until a detection occurs or the stream ends.
3. The listener applies its default one-second cooldown after a detection.
4. The while let loop continues until next_detection() returns Ok(None). Microphone failures are returned as errors rather than silently hidden.

“Default” is selected once

The operating system’s default microphone is selected when the listener is created. An existing stream does not migrate when the system default later changes. See Disconnect & reconnect.

Configure it with a builder

use std::time::Duration;
use micro_wakeword::Listener;

# fn run() -> micro_wakeword::Result<()> {
let mut listener = Listener::config_builder("models/wake-word.json")?
    .device("USB Mic")
    .cooldown(Duration::from_secs_f64(0.5))
    .build()?;
# Ok(())

The device can be its numeric index, its exact name, or an unambiguous case-insensitive part of its name.