Smart acoustic sensor systems for your products and processes

Smart hardware that is easy to integrate 

The Oldenburg Branch for Hearing, Speech and Audio Technology HSA of Fraunhofer IDMT develops hardware and software for recording and processing acoustic data in both cabled and wireless networks for all sectors.

The systems are assembled so that they meet the requirements of the respective market and of customer-specific applications. This begins with the selection of suitable sensor elements, computing platforms, housing, power supply and communication capacities and continues in the area of software. Corresponding processes are deployed depending on the application, e. g. signal preprocessing, noise cancellation, localisation of sound sources or data reduction. In addition, modern machine learning and artificial intelligence methods as well as classical algorithms are used as software modules for modern service architectures.

The sensor solutions are capable of processing large amounts of data on-premises. This relieves the load on communication channels and at the same time satisfies data protection and privacy requirements.

User group-specific interfaces and gateways for end users and experts round off the sensor systems and make them operable.

Acoustic sensor networks

For complex applications, acoustic sensor networks can be built. Meta-information forwarded from individual smart sensors enable the spatial, temporal and statistical modelling of acoustic events as well as conflation with multimodal information of other types on a central server.

The performance of smart acoustic sensor systems will improve even more in the future through the integration of AI-capable microelectronics. Last but not least, we are constantly improving our systems via research and development projects.

Multimodality is the technology of the future: To better map more complex situations, we are already linking acoustic information with other sensor modalities. The combination of different sensors makes it possible to obtain more comprehensive information.

Smart sensors can be an independent component of manufacturing machinery, identifying faults in advance and optimising processes, such as the detection of pneumatic actuator leakage. Higher operating costs, lower machine efficiency or isolated machine failures can only be linked to a loss of compressed air at a late stage. Consistent monitoring makes machine downtime plannable and less cost-intensive.

Researchers at Fraunhofer IDMT in Oldenburg have developed the SensOhr® platform to study smart sensor systems and their use in humans. The main areas of application are scenarios that require monitoring – be it in health applications or in communication. Sounds such as snoring, heartbeat, swallowing or speech can be combined with recordings of (other) vital data such as EEG or oxygen saturation. In this way, SensOhr® can form the basis for recording various biosignals. It is a platform for the sensor-based recording of biosignals close to the ear. The main area of application is in monitoring, for example in healthcare. This includes, for instance, the monitoring of speech or sleep disorders. But it could also be used in nursing care or hospitals, for example to monitor coughing or breathing.

When using smart sensors in the security sector, we focus above all on noise reduction with the help of several microphones, known as microphone arrays, to separate relevant and less relevant sounds. In addition, such setups make it possible to locate sources acoustically and combine them directly with acoustic monitoring. In so doing, we are opening up the possibility to detect and localise drones that infiltrate security areas. By detecting broken glass or spray cans, we can contribute to combating burglaries and vandalism. Directional filtering, also called beamforming, makes it possible to direct the microphone to where it is needed. For example, the beamformer amplifies those audio signals which the user wants and simply blocks out those noise sources that are unwanted, such as other people’s voices.

The car is learning to hear: In our project “The Hearing Car”, smart sensors help the car to hear. Machine learning methods teach it to recognise sirens, for example. For this purpose, we are applying our cross-group expertise to develop AI algorithms for acoustic event detection or optimal signal capture through sensor positioning, signal preprocessing and enhancement, and noise cancellation.

Fraunhofer IDMT offers robust signal processing methods and a broad portfolio of detector models that can be adapted to different application scenarios.