kokoro-tiny 🎤

Minimal, blazing-fast TTS (Text-to-Speech) crate powered by the Kokoro model (82M params). Perfect for embedding in applications, system alerts, and smart tools!

🚀 0.5-2s time-to-first-audio | 📦 Single-file implementation | 🎯 Zero-config usage

Features

• ⚡ Extremely fast inference using ONNX Runtime
• 🎨 50+ built-in voices with style mixing support
• 🔊 Direct audio playback with volume control
• 📁 Multiple formats: WAV, MP3, OPUS, FLAC
• 💾 Smart caching - downloads model once to ~/.cache/kokoros
• 🛠️ CLI included - kokoro-speak for instant TTS

Quick Start

Add to your Cargo.toml:

[dependencies]
kokoro-tiny = "0.1"

Usage

As a Library

use kokoro_tiny::TtsEngine;

#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {
    // Initialize (downloads model on first run)
    let mut tts = TtsEngine::new().await?;

    // Generate speech
    let audio = tts.synthesize("Hello, world!", Some("af_sky"))?;

    // Save to file
    tts.save_wav("output.wav", &audio)?;

    // Or play directly (with 'playback' feature)
    #[cfg(feature = "playback")]
    tts.play(&audio, 0.8)?;  // 80% volume

    Ok(())
}

Voice Mixing

Create unique voices by blending:

// Mix voices with weights
let audio = tts.synthesize(
    "Creative voice mixing!",
    Some("af_sky.8+af_bella.2")  // 80% Sky + 20% Bella
)?;

CLI Tool

Install the CLI:

cargo install kokoro-tiny

Use it instantly:

# Speak text
kokoro-speak say "Hello from kokoro!"

# System alerts
kokoro-speak alert success "Build complete!"

# Pipe input
echo "Processing files..." | kokoro-speak pipe

# Context summaries (perfect for smart-tree!)
kokoro-speak context "Found 5 TypeScript files with 200 lines total"

# Save to file
kokoro-speak say "Save this speech" -o output.wav

# List voices
kokoro-speak --list-voices

Available Voices

50+ voices across accents and styles:

• American: af_sky, af_bella, am_adam, am_michael
• British: bf_emma, bm_george
• Special: af_heart (warm), am_echo (clear)

Use --list-voices to see all options!

Features

Enable additional capabilities:

[dependencies]
kokoro-tiny = { version = "0.1", features = ["all-formats"] }

• playback - Direct audio playback (default)
• mp3 - MP3 encoding support
• opus-format - OPUS for streaming/VoIP
• cuda - GPU acceleration
• all-formats - All audio formats

Examples

Check out the examples directory:

# Simple usage
cargo run --example simple

# Test all voices
cargo run --example test_voices

# Audio format comparison
cargo run --features all-formats --example audio_formats

Performance

• Time-to-first-audio: 0.5-2 seconds
• Model size: 82M parameters
• Audio quality: 24kHz sample rate
• Memory usage: ~200MB with model loaded

Use Cases

Perfect for:

• 🔔 System notifications - Build alerts, test results
• 📊 Smart tools - Audio context summaries
• 🎮 Game development - Dynamic NPC voices
• 📱 Accessibility - Screen reader functionality
• 🤖 Automation - Voice announcements for scripts

Model Details

kokoro-tiny uses the Kokoro-82M model:

• Lightweight transformer architecture
• Trained on extensive speech datasets
• Optimized for CPU inference

Contributing

Contributions welcome! This project is maintained by Hue & Aye at 8b.is.

For the full Kokoro implementation with advanced features, check out Kokoros.

License

Apache 2.0 - See LICENSE

Built with 🎉 by the 8b.is team | Powered by the amazing Kokoro model