HMGCC Co-Creation Challenge: Cutting eavesdropping risks using AI

About the challenge

Government offices are often found in multi-occupancy buildings and open plan offices. All offices are designed to National Protective Security Agency specifications, to ensure a standard in physical and cyber protection. But there is more to learn. Understanding risk in this type of working environment is an important function. If there is an opportunity for eavesdropping, either accidentally or by a nefarious party, we would like to understand how challenging it would be to cancel out the irrelevant ambient noise to focus in on the conversation of significance.

The latest challenge launched by HMGCC Co-Creation sets out to understand the threat of third parties using artificial intelligence (AI) / machine learning (ML) to cancel out randomised and unwanted noise.

Within office environments, there is a general noise from heating, ventilation, air conditioning systems (HVAC), desk fans, doors closing and background conversations. All of this constitutes random noise generation.

What is already known about how to cancel this noise out? Digital signal processing with adaptive filtering is well known. We want to know more about the threat of cutting- edge methods to increase signal to noise ratios, used to focus on specific conversations.

There has been a rapid rise in recent years of AI and ML adoption in most sectors. There has also been interest and advanced research into using deep learning and neural networks to provide real-time noise cancellation. HMGCC Co-Creation is now seeking to better understand the threat through testing advanced noise cancellation capabilities.

Example use case

Government employee Sam is having a private call in an office booth. The booth is open but designed to dampen leaking sound to the external office. For other office users standing close by there is limited sound leakage, so Sam’s conversation can stay private.

The outer area of the office is open plan and there are various online calls happening, as well as general office noise.

Shauna is a few metres away from Sam, using her phone. The phone is picking up all the audio in the room but with the general office noise, Sam’s conversation is unlikely to be recorded to an intelligible level, whether the audio was picked up accidentally (over a phone call) or nefariously (by taking an active recording).

If audio was downloaded from Shauna’s phone, there are existing software packages that could be used to remove background noise. However, the following points should be considered:

• Software packages are typically focused on the commercial market, such as podcasting, music recording, and online calls, where there is a controlled and predictable environment of microphone placement and high signal to noise ratio.
• Software packages are unlikely to filter between other less sensitive conversations in the office and focus on Sam’s private conversation.
• Shauna’s microphone is dynamic, she may decide to sit in a different area or move about, all while still picking up audio. Shauna’s phone contains a single microphone, not an array of microphones.

In this scenario, it is highly unlikely that Sam’s conversation could be intelligibly intercepted from Shauna’s phone even with modern post processing techniques. But could advances in AI / ML audio processing pose more risks?

Eligibility

This challenge is open to sole innovators, industry, academic and research organisations of all types and sizes. There is no requirement for security clearances.

Solution providers or direct collaboration from countries listed by the UK government under trade sanctions and/or arms embargoes, are not eligible for HMGCC Co-Creation challenges.