Tract-precise

A reimplementation of the precise by Mycroft hotword listener (just the decoder, though) using Rust and TFLite.

Usage

Add this to your Cargo.toml

[dependencies]
precise-rs = {git = "https://github.com/sheosi/precise-rs"}

How it works

A little example:

fn load_samples() -> Vec<i16> {
    let mut reader = hound::WavReader::open("test.wav").unwrap();
    let samples: Vec<i16> = reader.samples().map(|e|e.unwrap()).collect();
    samples
}

fn main() {
    const WAKEWORD_THRESHOLD : f32 = 0.8;

    let mut precise = Precise::new("hey-mycroft.pb").unwrap();

    if precise.update(&load_samples()).unwrap() > 0.8 {
        println!("Wakeword recognized");
    }
}

This example loads the test.wav Wav file and runs that audio through the hey-mycroft.pb tflite model, note that also hey-mycroft.pb.args needs to be present, which is the configuration for that model. You can find more info here.

Obtaining the code

Before cloning you need to have git LFS (Large Files Storage).

First install git LFS, for Debian/Ubuntu it is:

sudo apt install git-lfs

For fedora:

sudo dnf install git-lfs

Then, no matter which OS you are under you need to initialize git LFS:

git lfs install

What works what not

Vectorizer

Only "mfcc" vectorizer is available and is one that's based very much like speechpy's, so anything using sonopy's mfcc vectorizer should work. This means that any model not requiring any vectorizer or setting it as "2" will work (that's speechpy's), any model using "1" should work (more testing is needed), and any model using "0" (mel_spec) won't work for now.

Delta

Delta is not implemented right now.

License

Licensed under Apache License, Version 2.0, (LICENSE-APACHE or http://www.apache.org/licenses/LICENSE-2.0).

Contribution

Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you, shall be dual licensed as using the Apache-2.0 license, without any additional terms or conditions.