We have two papers selected for presentation at this year’s ICBEN Congress scheduled to take place in Copenhagen. This is the 15th Congress on Noise as a Public Health Problem that brings together researchers and policymakers to share the latest discoveries and develop public health strategies addressing noise-related challenges. We are looking forward to the multi-disciplinary discussions ahead.
At the Congress, representatives from our Aviation and Building Acoustics Sector teams will be presenting their research. The abstract of their papers are below:-
Paper 1
Embedding a longitudinal Quality of Life study within an airport noise insulation scheme: delivery challenges and exposure methods – Robin Monaghan and Nicole Porter
Abstract:
Large airport noise insulation programmes can reduce indoor aircraft noise, but demonstrating wider benefits and maintaining community trust requires more than installation counts. We present a programme-level view of a longitudinal Quality of Life (QoL) study embedded within an ongoing insulation scheme, delivered through a partnership between an airport, an academic research team, and a service-integrator consultancy responsible for field delivery, resident engagement and quality assurance.
The study is designed to be light-touch for households while supporting repeated measurement. It combines complementary exposure approaches: (1) standardised facade sound insulation measurements using a controlled loudspeaker source to characterise repeatable building performance, and (2) aircraft noise measurements to capture real-world operational variability. These data are linked to repeated resident questionnaires and scheme milestones, allowing change to be interpreted in the context of both building performance and the noise environment.
We focus on practical challenges that materially affect data quality and representativeness at scale: recruiting and retaining households over multiple years; appointment logistics in occupied homes; consistency of field teams, equipment and evidence standards; governance across multiple delivery partners; version control and traceability across large property cohorts; and transparent, fair communications as households move through phased roll-out. We outline controls used to manage these risks, including standard operating procedures, equipment calibration, metadata capture, audit trails, and a governance cadence for exceptions and resident complaints.
In our experience, longitudinal QoL evaluation embedded within routine scheme delivery is still uncommon in consultancy practice. The lessons presented are transferable to other large-scale community noise mitigation programmes.
Paper 2
Interpretable urban sound source classification using projection pursuit regression and third octave spectra – Octavio Lora Aranda
Abstract:
Accurate identification of dominant urban noise sources is increasingly important for interpreting community noise exposure and targeting mitigation. Many modern machine learning approaches can achieve high classification accuracy, but are often difficult to interpret and may require intensive computation. This work explores Projection Pursuit Regression (PPR) as an interpretable alternative for sound source classification, with a focus on features aligned to environmental acoustics practice.
PPR models the response as a sum of smooth functions applied to linear projections of the predictors, enabling non linear relationships and interactions to be represented through a small number of informative projections.
We apply PPR to classify urban sound events using publicly available machine learning libraries, using third octave band spectral features rather than raw audio waveforms. Input features are derived across 20 Hz to 16 kHz; spectra are min max normalised to emphasise relative spectral patterns, and time varying information is represented via third octave band spectrograms computed with a 50 ms Hann window and 50 percent overlap.
Model performance is demonstrated on a subset of environmental sound event classes (birds, dogs, cars, sirens, trains), using progressively larger training sample sizes per class (10, 30, 50). Test performance increases with training data, highlighting both the promise of PPR for this application and the risk of overfitting when training sets are small.
We discuss practical next steps for field deployment, including expanding training data, handling unknown events using confidence thresholds and out of distribution detection, and evaluating the use of measured Leq,T spectra as a lower cost and more sustainable alternative to processing full audio recordings.
