Context & Problem definition

In the Meetup of November 21th 2018, we discussed the problem of noise pollution in the city. After a lot of complaints / reports about nuisance, the Ombudsman of Amsterdam decided to live in the center for a few months to gain some first-hand experience. During the meetup he played some sound samples of nightly street-noise he had collected in his bedroom. It became clear that the noise nuisance caused by visitors / tourists / horns (taxis etc) is disturbingly high. Moreover, the WHO last year issued a report in which sound nuisance is in the top 3 of health hazards.

Noise nuisance is hard to make tangible, it is difficult to get it prioritized on the city agenda and hard to enforce measures. Therefore, the City of Amsterdam, the Ombudsman and Waag.org looked at how the problem could be made more tangible and tackled more effectively.
They concluded that a new dimension could be added to policymaking, by providing more concrete data about what sounds, on what venues and at what times, caused most nuisance. So based on their experience in the
Europese Making Sense project, the Waag will facilitate citizens to analyze noise nuisance problems with the help of a network of sensors (smart microphones) in the top nuisance areas to provide more insight. They need to be smart microphones, not only measuring the sound pressure but also identifying the major sound sources.  “Decibels” in itself aren’t a enough for identifying the actual source of the nuisance.
For cost reasons however, the network will consist of many decibel meters combined with fewer microphones that can classify sound. For the latter, they have teamed up with Sensing Clues, who already are developing such smart microphones to help fight jungle poaching in Africa. Sensing Clues also benefits since this project also is an opportunity to improve their device for use in the jungle.

Meanwhile, Sensemakers and the Marineterrein were brainstorming about a new joint project, in addition to their water quality endeavor. Sensing sound seemed to be the most relevant (due to the nuisance at the Marineterrein) and interesting for the Sensemakers (new domain). Hence we teamed up with the Waag & Sensing Clues.

At the Waag, they have been working on a general kit that measures decibels. It would be a shame if we would repeat the development, so if you want to help create volume, and do a nice starter project to work/learn on, check this.

For the more experienced Sensemakers it’s interesting to work on the sound classification sensor. You will use digital signal processing, and deep learning for the classification of sound on the sensor.

We see AI moving towards the edge more and more. Another example being the SNIPs solution to do recognition of voice commands in house, without sending data to the cloud. This is important as we want the solution Privacy by Design.

Project Goal

Sensing Clues has a working prototype of a smart microphone, classifying sound in the device providing Privacy by Design. And just as important, allowing low bandwidth connections (LoRa, NB-IoT) in addition to WiFi, Ethernet, 4/5G and the like, as that would allow to use the free TTN-network.

However, it’s still a prototype with a lot of challenges:

  • Energy consumption is still way to high, draining batteries way too fast and heating up the device. We want to be as free as possible in device placement, indicating a battery powered solution. So, we radically need to find ways to get power consumption and heat production down.
  • Price.  We want many devices, both in the city and the jungle. At the moment a quite expensive microphone is required, making classification at a greater distance possible.
  • Robust design (casing, weather, tamper-proof).
  • Various communication options, it the wild we likely need LoRaWan, perhaps even LoRaWan to low orbit satellite. In the cities LoRa (free of charge by TTN, NB-IoT, LTE/M, M2M and perhaps even bursted WiFi are viable options.

Although a sound classification sensor for the jungle and for the city may differ somewhat in requirements, they can both be worked on. Actually, when the sensor can work in the jungle it certainly will be usable in the city. So as Sensemakers our goal is to work together with Waag and Sensing Clues to further develop a better and cheaper version of their prototype, deploy these in different pilots, share the learnings and publish (Privacy by Design) design solutions for others to copy, while allowing Sensing Clues to use the knowledge in their products.

Partners & roles

In this project we work together with:

  • City: funding, involved as stakeholder, co-responsible for measures to reduce nuisance
  • Waag, contractor, final responsibility for the official Amsterdam Sounds project; deciding on areas, involving residents & other stakeholders (like Sensemakers, those living in Amsterdam & experiencing  nuisance can also measure sound in their surroundings); collecting sound samples, hardware, dashboard.
  • Sensing Clues, working for the City/Waag as provider of the classification model (AI) for the smart microphones. Offered to share their present prototype, learnings and also the Classification Model with Sensemakers as to co-create a better/cheaper hardware solution, so that it’s more viable for citizens to buy their own kit to use in their local environment and share the result in the common dashboard.
    The Machine Learning to improve the model based on collected sound samples will be done by SenseingClues. Sensemaker’s focus will be on improving the hardware and to help citizens, rangers (in the real jungle against poachers) and others to collect relevant data.
  • Sensemakers: involved voluntarily to help create a DIY affordable, robuust and flexible sensing solution that impact actual problems, while learning about sensing, IoT etc.
  • Marineterrein: involved on voluntary basis as partner of Sensemakers, facilitates the testing of prototypes & other innovations on the Marineterrein.
  • Programma DI020 (collaboration of the city and several institutions): has developed a Privacy by Design framework, will test & improve it by applying it to this project.

Approach

Context & goal are set (a network of (smart) sound sensors to gain insight into what sounds, on what venues and at what times, causes most nuisance), as well as the involvement of Sensemakers (help further develop a better and cheaper version of a smart microphone that can classify sounds, share the learnings and publish design solutions for others to copy) with focus on the hardware.

On Feb. 6th we’ve met up with people interested in getting involved for a first kick-off and getting to know each other. Those that couldn’t make it, can still join March 6th when also the tech-people of Sensing Clues will join, and we start with the following:

  1. Defining and agreeing on requirements
  2. Collecting learnings from Sensing Clues and other sources
  3. Agreeing on the technical architecture (directions) of the sensor. Here we still look at all components (hardware, software, costs and complexity) in relation to each other. How the work together.
  4. For the most promising architecture(s) we then make a high level design for the prototype(s).

These can differ substantially from the original prototype Sensing Clues developed. E.g. we could decide to use a more, for deep learning, specialized alternative to the Raspberry Pi.

  • Finally, we agree tasks and assign them to the interested participants.

This is quite a lot of planning work, so it may take more than one session.  It may also be required that some more investigation or experimenting is required before we can choose a direction.