Smartcity/ sound recognition

Municipality of The Hague

Algorithm that detects which noise can be seen in the spectrogram based on spectrograms. The algorithm aims to get an overall picture of where, when what types of noise are present so that further research can be done on them.

Last change on 23rd of August 2024, at 15:53 (CET) | Publication Standard 1.0

• Publication category: Impactful algorithms
• Impact assessment: DPIA
• Status: Out of use

General information

Theme

Public Order and Safety

Begin date

Field not filled in.

Contact information

[e-mail]

Link to publication website

De Cyrb-sensoren zijn toegevoegd aan onderstaande kaart. https://ddh.maps.arcgis.com/apps/webappviewer/index.html?id=8531785e9a8c4450be8839385003f1bc Bewoners zijn geïnformeerd over de diverse testen die worden uitgevoerd in het kader van het Living Lab Scheveningen programma. https://www.denhaag.nl/nl/in-de-stad/wonen-en-bouwen/ontwikkelingen-in-de-stad/ontwikkelingen-scheveningen-kust/nieuwe-uitvindingen-in-living-lab-scheveningen.htm

Responsible use

Goal and impact

The algorithm was developed to recognise sounds. In the current application, for sounds above a certain decibel level, it identifies what the nature of the loud noise was. This gives an insight into the kind of noises that cause nuisance. The algorithm aims to get an overall picture of where, when what types of noise nuisance are present so that further research can be done on them. The results of the algorithm are not intended as grounds for immediate decisions or actions without verification of the results.

By better understanding the cause of noise pollution experienced by citizens, the municipality can better perform its task in maintaining a pleasant comfortable outdoor space. Besides being able to identify the type of noise (noise recognition), the noise level (decibels) was also measured in a test setup. At present, no citizens are interacting with the algorithm. Two residents of the area where this algorithm was tested were involved for the duration of the test to investigate whether the intended concept application works. They gained insight into the decibel levels measured in the outdoor area through a dashboard (nb: the algorithm itself does not measure decibels). At these two residents' homes, a sensor was hung on their facade to verify what the noise levels are at their homes. Besides these two intensively involved residents, a larger event was organised during the summer of 2020 to inform visitors and residents of and on the promenade, during the Smart @Sea festival.

Considerations

There is no alternative to the noise recognition algorithm. The algorithm is an addition to noise level measurement. The algorithm is not used for decision-making.

Human intervention

The algorithm can be stopped by shutting down the processes on the sensor. This can be done remotely by Cyrb or by the administrator by switching off the power to the sensor. No processes flow directly from the algorithm's outcomes.

Risk management

The algorithm is trained on public datasets of similar sounds. Because these do not always sound the same and have the same characteristics (and nature) as the sounds captured on location, the algorithm could make errors in sound recognition. Because the algorithm only serves imaging purposes, it does not pose any direct risks to the citizens and entrepreneurs involved. The risk of data leakage (recorded sounds) has been overcome by not storing the sound on the device but directly executing the sound recognition on the device.

As long as the system is not used, there is also no need to update the 2019 DPIA.

Impact assessment

Data Protection Impact Assessment (DPIA)

Operations

Data

The algorithm processes only spectrograms that were in turn generated from self-collected audio files.

Technical design

In evaluating the model, it uses the confusion matrix and the categorical-crossentropy as a metric.

The algorithm converts 4-second sound clips into spectrograms, visual representations of the sound. The spectrograms are images of 128x128 pixels with greyscale values. These spectrograms are then classified by an image recognition algorithm into a number of categories, such as "stationary traffic", "music", "honking", etc. This image recognition model is a relatively small convolutional neural network of five layers. (Three convolutional layers followed by max-pooling and two fully connected layers). The model runs in python, using the Tensorflow library.