FdafAec: Frequency Domain Adaptive Filter for Acoustic Echo Cancellation

fdaf-aec is a Rust library that implements an Acoustic Echo Canceller (AEC) using the Frequency Domain Adaptive Filter (FDAF) algorithm with the Overlap-Save method. It is designed to be simple to use while providing a robust solution for real-time echo cancellation in voice communication applications.

What is Acoustic Echo Cancellation?

In a typical hands-free communication setup (like a video call), the sound from the far-end user (played through your speakers) can be picked up by your microphone. This causes the far-end user to hear an echo of their own voice, which is disruptive. An AEC is a signal processing component that identifies and removes this unwanted echo from the microphone signal, resulting in a clean audio stream.

graph TD;
    subgraph "Far-End User"
        A["Voice from Call"];
    end

    subgraph "Your Room"
        B["Loudspeaker"];
        C["Your Voice"];
        D["Microphone"];
        E["Acoustic Echo Canceller (AEC)"];
    end

    subgraph "To Far-End User"
        F["Cleaned Voice"];
    end

    A -- "Reference Signal" --> E;
    A -- "Played out loud" --> B;
    B -- "Acoustic Echo" --> D;
    C -- " " --> D;
    D -- "Mic Signal<br>(Your Voice + Echo)" --> E;
    E -- "Removes Echo" --> F;

How It Works

This library uses a Frequency Domain Adaptive Filter (FDAF). Here's a simplified overview of the process:

Buffering: It takes a frame of audio from the far-end (speaker) signal and the microphone signal.
FFT: It transforms these audio signals into the frequency domain using the Fast Fourier Transform (FFT).
Echo Estimation: In the frequency domain, an adaptive filter (represented by a set of complex weights) models the echo path. It uses the far-end signal to predict what the echo should sound like.
Subtraction: The predicted echo is subtracted from the microphone signal, leaving (ideally) only the near-end user's voice.
Adaptation: The filter constantly adjusts its weights using the Normalized Least Mean Squares (NLMS) algorithm to adapt to changing room acoustics and echo paths.
IFFT: The cleaned signal is transformed back into the time domain (audio samples) and returned.

The Overlap-Save method is used to efficiently process the audio in blocks, making it suitable for real-time applications.

Features

Real-time capable FDAF implementation.
Adjustable learning rate (step size) to balance convergence speed and stability.
Simple and straightforward API.
Minimal dependencies for the core library.

Getting Started

To use fdaf-aec in your project, add it to your Cargo.toml:

[dependencies]
fdaf-aec = "0.1.0" # Replace with the latest version

Then, initialize the FdafAec and process your audio frames:

use fdaf_aec::FdafAec;

// Configuration
const FFT_SIZE: usize = 1024; // Must be a power of two. Determines filter length.
const FRAME_SIZE: usize = FFT_SIZE / 2; // Should be half of FFT_SIZE.
const STEP_SIZE: f32 = 0.02; // Learning rate. A small value between 0 and 1.

// Create a new AEC instance
let mut aec = FdafAec::new(FFT_SIZE, STEP_SIZE);

// Your audio processing loop
// loop {
    // let far_end_frame: &[f32] = get_far_end_audio_frame(); // Should have FRAME_SIZE samples
    // let mic_frame: &[f32] = get_mic_audio_frame();     // Should have FRAME_SIZE samples

    // Process the frames to get the echo-cancelled signal
    // let output_frame = aec.process(far_end_frame, mic_frame);

    // Use the `output_frame` for your application...
// }

Examples

The project includes several examples in the examples/ directory to demonstrate its functionality.

1. Basic Simulation

This example runs a simple simulation with sine waves to show the core principle of echo reduction. It prints the RMS energy before and after processing.

cargo run --example basic_simulation

2. Generated Signal AEC

This example provides a more realistic test by generating white noise (far-end) and a simulated voice signal (near-end). It models a simple room echo and saves the audio before and after AEC to WAV files in the output_generated/ directory for listening.

# The --release flag is recommended for faster execution
cargo run --example generated_signal_aec --release

3. File-Based AEC (CLI Tool)

This is a command-line utility for processing your own WAV files. It's the most practical way to test the AEC's performance on real-world recordings.

Usage:

# Place your WAV files in the project root directory
# For best results, use mono, 16kHz WAV files.
cargo run --example file_based_aec --release -- \
  --farend your_farend_file.wav \
  --mic your_mic_file.wav \
  --output processed_output.wav

License

This project is licensed under the MIT License.

fdaf-aec 0.1.0