rust-tts-wrapper 1.5.1

//! Pocket-TTS with real word timings: our own streaming inference over the
//! Kyutai ONNX graphs (lm_main + attention tap / lm_flow / mimi encoder +
//! decoder / text conditioner), so every generation step's attention over
//! the text tokens is captured — the model's own alignment, no aligner.
//!
//! Layout + recipe: ~/models/pocket-onnx/SPRINTS.md (validated in Python by
//! run_tapped.py; whisper cross-validation within ~0.17s on matched words).

pub mod model;
pub mod timings;
pub mod tokenizer;

pub use model::{PocketConfig, PocketTtsModel};
pub mod engine;
pub use engine::PocketTtsEngine;
pub use timings::{word_boundaries, word_boundaries_grouped};
pub use tokenizer::{AnyTokenizer, PocketTokenizer, WordLevelTokenizer};