opus-head-sys

Rust FFI bindings to the Opus audio codec with vendored source code, including support for AI/DNN features (DRED, OSCE).

Overview

This crate provides low-level bindings to libopus, built from vendored source code. The goal is to track Opus master as closely as possible with frequent updates.

Features

Vendored source - Opus is compiled from source, no system dependencies required
AI/DNN support - DRED (Deep Redundancy) and OSCE (Opus Speech Coding Enhancement) via runtime weight loading
Full API coverage - Core, Multistream (surround), and Projection (ambisonics) APIs
Static linking - Single binary with no runtime dependencies on libopus
Small crate size - Under 5MB (DNN weights loaded at runtime, not embedded)

What this crate is NOT

This crate focuses on simplicity and staying close to upstream Opus. The following are non-goals:

Dynamic linking to system libopus
pkg-config or system library detection

If you need these features, consider other opus-sys crates like audiopus_sys or opus-sys.

Using AI Features (DRED/OSCE)

The AI features require loading DNN weights at runtime. This keeps the crate small while still supporting the full Opus AI capabilities.

1. Enable the features

[dependencies]

opus-head-sys = { version = "0.1", features = ["dred", "osce"] }

2. Download the weights file

Download opus_data-<hash>.bin (~14MB) from the releases page.

Place it in your project or a known location to load at runtime.

3. Load weights at runtime

use opus_head_sys::*;

// Read the weights file
let weights = std::fs::read("opus_data-<hash>.bin").expect("Failed to read weights");

unsafe {
    // Create encoder
    let mut error = 0;
    let encoder = opus_encoder_create(48000, 1, OPUS_APPLICATION_VOIP as i32, &mut error);
    
    // Load DNN weights into encoder
    opus_encoder_ctl(
        encoder,
        OPUS_SET_DNN_BLOB_REQUEST as i32,
        weights.as_ptr() as *const std::ffi::c_void,
        weights.len() as i32,
    );
    
    // Now DRED encoding is available
    // Enable DRED:
    opus_encoder_ctl(encoder, OPUS_SET_DRED_DURATION_REQUEST as i32, 100); // 100ms of redundancy
    
    // ... encode audio ...
    
    opus_encoder_destroy(encoder);
}

The same applies to the decoder for OSCE (speech enhancement) features.

Why runtime loading?

The DNN weights are ~14MB, which would exceed crates.io's 10MB limit if embedded. Runtime loading also allows:

Updating weights without recompiling
Sharing weights across multiple encoder/decoder instances
Optional AI features (don't load weights if you don't need them)

Vendored Version

The vendored Opus source is tracked in vendored/OPUS_VERSION. Run the update script to sync with upstream:

python vendor_opus.py

License

The Rust bindings in this crate are licensed under MIT OR Apache-2.0.

The vendored Opus codec is licensed under the BSD 3-Clause License. See NOTICE for full copyright information and patent licenses.

opus-head-sys 0.1.0