Feat: Parler TTS Support

[DogeDark]

Adds basic Parler TTS support using the Candle integration. This supports both Mini and Large Parler models.

This implementation is very basic with no streaming and limited output methods. I did not (knowingly) implement any of the optimization tips.

This implementation is slow. The optimization tips link mentioned the capability of 500ms to-first-speech with streaming implemented. On my machine, it takes a good 10 seconds on the mini model w/RTX 3080 cuda.

Wav file output is behind the wav feature flag.

Demo

yodude1.mp4

#[tokio::main]
async fn main() {
    // Get parler
    let parler = ParlerBuilder::default()
        .with_source(ParlerSource::MiniV1)
        .build()
        .await
        .unwrap();

    // Get a task handle
    let task = parler.generate(
        "Yo dude! This Parler text-to-speech integration for Kalosm is sick!",
        "Will's voice is enthusiastic and slightly fast in delivery, with a very close recording that almost has no background noise.",
    );
    
    // Set settings
    let task = task.with_settings(
        GenerationSettings::new()
            .with_temperature(Some(1.0))
            .with_top_p(Some(0.2)),
    );

    // Get the decoder and output to wav file.
    let decoder = task.await.unwrap();
    decoder
        .to_wav(PathBuf::from("./parler-output.wav"))
        .unwrap();
}

To-Do:

• Maybe try optimizing it.
• Kalosm-sound re-export?