AES-Hackathon

Contest Guidelines

Hackathon Goal

Participants will develop a MATLAB audio plugin that helps podcasters and social-media creators achieve clear, professional-sounding speech recordings in noisy, budget environments. By combining Voice Activity Detection (VAD) with AI-based enhancement and classic DSP, the plugin will automatically suppress background noise and polish vocals with a simple, user-friendly interface.

Video Intro

Watch Dr. Bennett describe the contest and give some tips for successful submissions.

Slide Deck

Download the PowerPoint presentation with useful tips and functions

Starter Template

A barebones audioPlugin with VAD implemented

Example Processing Chain

These blocks, and there ordering, are totally optional - this is just a starting off point for a signal chain for this type of plugin effect.

High Pass Filter

Removes low-frequency rumble and proximity effects

VAD Soft Gate

Uses 'voiceActivityDetector' to compute speech-presence probability.

Applies gating to attenuate non-speech.

AI-Based Enhancement

Calls 'enhanceSpeech' to denoise and reduce residual artifacts

De-Esser

Reduces harsh sibilants with a narrow-band compressor around 5–10 kHz

Parametric EQ

Adds presence with a gentle boost at 3–5 kHz for clarity

Compressor

Smooths level variations (e.g. ratio 3:1, attack 10 ms, release 100 ms)

Limiter

Prevents clipping and controls peaks

User Gain

Final output gain controlled by the user (–12 dB to +12 dB)

Intelligibility

Measure the STOI of the speech signal

Judging Criteria

Here are some tips for scoring highly with the judges!

Fulfillment of Coding-Challenge Criteria

VAD Integration: Properly incorporate the provided VAD output into your signal chain (e.g. gating or level control).
Built-In System Objects: Try to use MATLAB Audio/DSP System Objects (e.g. noise gate, compressor, EQ).

Technical Merit

Noise Reduction Performance: Objective STOI improvement on the provided noise/speech files.
Speech Clarity & Naturalness: Minimal gating “chatter,” pumping, distortion, or tonal artifacts.
Efficiency: Frame-by-frame processing runs in real time on a standard laptop without dropouts.

Innovation & Originality

Creative Signal Flow: Novel ordering or combination of DSP
Adaptive Behaviors: e.g. dynamic attack/release, parameter switching driven by VAD statistics.
Preset Mode Design: Intelligent presets (e.g. auto-tuned for “Subtle” vs. “Broadcast” scenarios).

Creative Value

User Experience: Intuitive labels, meaningful defaults, and clear “one-click” improvement.
UI Simplicity: Avoid overwhelming the lay user; fewer, more impactful controls.
Polish: Visual metering, tooltips, or simple visualization that enhance usability.

Execution & Code Quality

Code Structure & Readability: Well-commented, modular design
Documentation: comments clearly describes signal chain, parameters, and usage instructions.
Robustness: Graceful handling of edge cases (e.g. different sample rates, stereo vs. mono mismatches).
VST: supplying a compiled VST

Welcome