Currently, homes and buildings around us become smart and more than ever connected to us and our devices. A plethora of sensors enables us to detect events in our environment. There are sensors to detect temperature, smoke or general movement. Especially the latter – the detection of movements – is in most cases related to sound of some sorts: opening of windows and doors, breaking of glass or many more. Each one of these events produces a more or less common pattern of sounds. The abstraction of these events by their sound patterns is the core of this project.
One of our missions with this project is to make the sensing of events cheaper and therefore more broadly available. We have high ambitions to develop hardware enabled connected home services that ease our daily personal and business lives. During this thesis a prototype for processing audio signals and pattern matching to detect certain activities should be developed and integrated into the Smart Energy Living Lab at fortiss.
Since the recognition of a lot of different sound patterns is a very complex target, at the first stage generic activities need to be recorded and classified. Hereinafter the interesting patterns are subjected to a more detailed analysis, to draw conclusions about the specific activities (computer work, house work, eating, etc.) or context information (number of persons, ongoing presentation, conference call, etc.). The goal is to utilize the background noise (non-speech) for assumptions about certain activities and develop a concept and prototype of an audio based activity detection component. This component will be integrated into the SMG2.0 middleware for a self-balancing smart grid node. Based on the requirement analysis with an partner from industry, 3-5 use cases will be implemented and evaluated.
– Analyse and evaluate related work in the domains of audio processing and acoustic activity detection
– Development of different scenarios and requirements elicitation
– Development of a generic concept for audio (only) based activity detection
– Priorisation of use cases
– Implementation of the proposed concept as a SMG2.0 component (a part of the existing Smart Energy Living Lab demonstrator)
– Statistical/Practical Evaluation (e.g. false positive and false negative rates) of the proposed solution
Compulsory qualifications:
– Work independently
– Enjoy working in teams
– Experience in Java and the Eclipse platform
– Beneficial: Experiences in
o Audio processing, feature extraction
o Machine learning (e.g. Tensorflow), anomaly detection and classification
General Conditions:
Date: from now
Duration: 4 – 6 Month
Supervisor: PD Dr. Bernhard Schätz
Advisors: Markus Duchon, Patrick Löchelt (thiinc GmbH), Till Klocke (thiinc GmbH)

If we have aroused your interest, send your application to career (at) Please note that applications without this code cannot be considered!
Application code: FB1-NOP-BAM-02-2016
Contact Person: Markus Duchon