To develop an AI/ML-enabled adaptive noise cancellation (ANC) system that effectively suppresses stationary, non-stationary, and impulsive defence noises while maintaining high speech intelligibility and real-time performance on embedded hardware.

The receipts 🧾

Every red flag on this statement, in full. These are the four places it bites.

1. Exhibit A

   Impulsive noise is a fundamentally different problem from stationary noise — a gunshot is a broadband transient that masks speech completely for its duration, and a model that hits the targets on rotor noise will fall well short on gunfire unless you train and report for it separately

2. It gets worse

   The category is Hardware, so a notebook demonstration with excellent metrics does not answer the statement — the embedded real-time path with microphones and a headset is a requirement, not a stretch goal

3. Still reading?

   Training on synthetically mixed noisy-clean pairs generalises poorly to real recordings where the noise is acoustically coupled and reverberant, so include real noisy recordings in evaluation even though the statement only asks for synthetic generation

4. And the finisher

   Real-time means bounded end-to-end latency including buffering, and a model with excellent PESQ at 200 milliseconds of algorithmic delay is unusable in a duplex radio conversation

The damage report 📉

Every score this statement earned, and what each one actually costs you.

• Feasibility

  4/5

  Actually buildable, which on this slate is rarer than it sounds. Do not squander it on scope.

  This is one of the best-supported statements on the portal — clean speech corpora, noise banks containing gunshots, helicopters and engines, and speech enhancement benchmark sets are all freely available, the model architectures and the three named metrics are standard open implementations, and useful edge inference runs on boards far cheaper than the development kit the statement mentions.

• Innovation scope

  3/5

  Mildly interesting. The novelty will not carry the room; the build has to.

  The statement prescribes the pipeline in detail down to the transform, the complex-domain operation, the loss functions, the augmentation strategy and the optimisation path, so your real design freedom is the architecture and how you handle impulsive events, which behave nothing like the stationary noise the rest of the pipeline assumes.

• Clarity

  5/5

  The ask is unambiguous, which quietly removes your favourite excuse.

  Exemplary — it names the noise classes, the feature representations, the loss functions, the evaluation metrics with numeric targets of SNR above 15 decibels, STOI above 0.85 and PESQ above 2.5, the deployment hardware, the optimisation techniques and the microphone integration.

• Acceptance potential

  4/5

  Strong footing before you have written a line. Try not to waste it.

  Numeric targets mean success is defined rather than argued, the training data is entirely free, the demo works on a judge's own ears, and defence audio is a far thinner field than the general AI themes — the main caution is that the impulsive case is genuinely much harder than the stationary one and the targets are quoted as if they were uniform.

• Effort

  Heavy

  Heavy. Somebody on this team is not sleeping in week three. Pick who, on purpose.

  A dataset mixing pipeline, model training to stated targets, quantisation and runtime conversion, real-time streaming inference with bounded latency, and physical integration with microphones and a headset is five workstreams, and the embedded real-time path is where teams that trained a good model discover they cannot ship it.

• Demo-ability

  Easy

  Easy to demo — and so is everyone else's. Working is the floor here, not the achievement.

  This is an auditory demo and those are rare and visceral — a judge hears the difference in their own ears rather than reading a metric, and switching live between the raw and enhanced channels is more convincing than any chart of PESQ scores.

• Data

  None supplied

  No dataset comes with this one, so every accuracy figure you quote is a number about labels you invented.

  Nothing is provided with the statement. You are sourcing, cleaning and labelling it yourself, and that work is invisible in the demo but very visible in the questions.

The demo they will have already seen 🪞

Somewhere around 45–100 teams are heading here, and the description is doing the choosing for most of them. They will read the same brief, reach the same architecture, and build a version of the same demo you are planning. Being correct is the floor. If your five minutes could be swapped with the team before you and nobody in the room would notice, you have not picked badly — you have built predictably, which costs exactly the same and hurts more.

What survives 🛡️

The ground worth standing on when the questions start.

• The three performance targets are numeric, so you can prove you succeeded instead of claiming it — this is unusual and it makes the whole submission objectively assessable
• Clean speech corpora and environmental noise banks containing gunshots, helicopters and engine sounds are all free and abundant, so the dataset pipeline the statement asks for is genuinely straightforward to build
• An auditory demo lands harder than a visual one and almost nothing else at the event will have one — the judge experiences the result rather than evaluating a claim about it