Module engine 

The in-process Kokoro-82M engine.

Re-exports

use std::sync::Mutex; 🔒

use std::sync::mpsc; 🔒

use tokio::sync::mpsc as tokio_mpsc; 🔒

use crate::model::ModelFiles; 🔒

use crate::Chunk; 🔒

use crate::Engine; 🔒

use crate::Error; 🔒

use crate::SpeakRequest; 🔒

Structs

Job 🔒

Kokoro 🔒

The loaded engine, as the wire conversion sees it.

KokoroTts

A loaded Kokoro-82M pipeline. One render thread per engine, so a second utterance queues behind the first; the thread owns the ONNX session and the runtime the engine’s async API needs.

Constants

DEFAULT_VOICE

The voice an utterance renders in when the request names none.

SAMPLE_RATE

The wire’s sample rate: 24 kHz mono 16-bit. Kokoro renders at this rate natively; anything else is resampled once, here.

Traits

Renderer 🔒

What the engine’s inference looks like to the wire conversion, so a test can drive the conversion with no model load.

Functions

chunk_for 🔒

One render: the voice the request names, the model’s samples, and the wire’s own shape.

pcm16 🔒

The wire’s sample: clamp to [-1, 1], then the signed 16-bit range.

refused 🔒

A channel carrying one failure, for a render thread that is gone.

resample 🔒

Linear resampling to the wire rate; identity when the rates match.

select_voice

The installed voices a request is checked against.