Crate rlx_kittentts 

KittenTTS — lightweight ONNX text-to-speech for RLX.

Backends

Feature	RLX runtime	ONNX Runtime EP
onnx	—	CPU (default)
native	RLX runtime	Decomposed kitten_tts_mini_rlx graph (no ORT)
rlx	RLX runtime	ORT inference + RLX path deps for future parity
metal	Metal	CoreML (macOS / iOS)
mlx	MLX	CoreML (Apple GPU)
cuda	CUDA	CUDA
rocm	ROCm	ROCm
gpu	wgpu	DirectML / CUDA / CoreML
full	all above	all ORT EPs + RLX path deps

Re-exports

pub use assets::DEFAULT_LOCAL_DIR;

pub use assets::ModelLayout;

pub use assets::default_model_dir;

pub use assets::default_native_weights_dir;

pub use assets::find_native_weights;

pub use assets::find_rlx_bundle;

pub use backend::OrtSession;

pub use backend::build_onnx_session;

pub use backend::execution_providers_for;

pub use backend::validate_device;

pub use config::DEFAULT_HF_REPO;

pub use config::ModelConfig;

pub use download::fetch_default;

pub use download::fetch_to_local_dir;

pub use features::cuda_feature_enabled;

pub use features::enabled_backend_labels;

pub use features::espeak_feature_enabled;

pub use features::gpu_feature_enabled;

pub use features::metal_feature_enabled;

pub use features::mlx_feature_enabled;

pub use features::native_feature_enabled;

pub use features::onnx_feature_enabled;

pub use features::rlx_feature_enabled;

pub use features::rocm_feature_enabled;

pub use infer_opts::SAMPLES_PER_DURATION_UNIT;

pub use infer_opts::recommended_native_compile_opts;

pub use model::KittenTTS;

pub use model::MIN_AUDIBLE_PEAK;

pub use model::SAMPLE_RATE;

pub use model::peak_amplitude;

pub use npz::NpyArray;

pub use npz::load_npz;

pub use npz::parse_npy;

pub use phonemize::DEFAULT_LANG;

pub use phonemize::is_espeak_available;

pub use phonemize::phonemize;

pub use phonemize::phonemize_lang;

pub use phonemize::set_data_path;

pub use tokenize::ipa_content_len;

pub use tokenize::ipa_style_index;

pub use tokenize::ipa_text_style_index;

pub use tokenize::ipa_to_ids;

pub use tokenize::warn_unknown_ipa_chars;

Modules

assets

Model directory discovery and path layout.

backend

Map rlx_runtime::Device to ONNX Runtime execution providers.

backend_kind

Inference backend selection (ONNX Runtime vs native RLX).

cli

CLI for IPA → WAV synthesis (ONNX Runtime or native RLX graph).

config

config.json schema from KittenTTS Hugging Face repos.

download

Hugging Face Hub download for KittenTTS checkpoints.

features

Compile-time feature probes and backend labels.

infer_opts

Native compile sizing and ONNX duration → waveform length mapping.

model

ONNX model runner — mirrors Python’s KittenTTS_1_Onnx.

npz

Minimal NPZ / NPY loader.

phonemize

Text → IPA via the pure-Rust espeak-ng crate.

tokenize

Character-level tokeniser — mirrors Python’s TextCleaner.

Enums

Device

Target device for graph execution.

Functions

fastest_device

Highest-priority backend that is compiled in and live on this host.

is_available

parse_device

Lower-case Cargo feature names and common aliases → Device.

parse_device_list

Parse comma/semicolon/whitespace-separated device lists (RLX_DEVICES=cpu,metal).