This challenge awaits you:

With advances in AI and machine learning, powerful methods for automatic processing and analysis of audio data are being used. Machines can extract information from audio signals, similar to humans. Applications include the identification of noises in vehicles, glass break detection, and animal noise classification. Extensive training data is required for reliable detection of acoustic events. Challenges arise with rare events that are difficult to label. Solutions include the use of growing amounts of data or optimized learning methods that require less training data. The following topics can be selected in this context: 

Semi-supervised learning: It can be assumed that, in the case of acoustic events that are particularly difficult to label, there is a large amount of data that contains the sound but is either unlabeled or incorrectly labeled. Semi-supervised learning methods can create reliable models for recognition and classification from a limited amount of labeled data and a large amount of unlabeled data.

Incremental learning: With incremental learning, a system can continuously learn and adapt to new acoustic events without requiring complete retraining. The amount of initial training data can also be kept manageable. The increments with additional training data can be obtained through continuous inference of the models on a large, unlabeled dataset, allowing the system to acquire new training data on its own.

Few-shot learning: The use of few-shot learning techniques enables the training of models that can deliver accurate results with only a few training examples. In this thesis, a method will be developed to detect and classify vehicle-related acoustic events using few-shot learning with a minimal dataset.

The aim of this thesis is to investigate one of the aforementioned methods for acoustic event detection and to enable the use of larger training data sets without additional labeling effort, or to achieve increased accuracy when using a small available (training) data set.

Your Tasks:

• You conduct literature research and familiarize yourself with the topic
• You develop and implement methods for detecting acoustic events or anomalies
• You analyze and optimize existing models for detecting and classifying acoustic events
• You conduct experiments and tests to validate the developed models
• You document and present your work results

Necessary Skills:

• Ongoing studies in computer science, mechatronics, electrical engineering, or a comparable technical or scientific course of study
• Programming experience in at least one high-level programming language
• Excellent English skills, both written and spoken
• Good German skills (at least B2 level)

Desired Skills:

• Experience or interest in the following programming languages / frameworks / tools: Python, Tensorflow, Keras, PyTorch, Kubeflow
• Knowledge in machine learning, neural networks, or pattern recognition
• Independent and structured way of working
• Willingness to learn, strong self-initiative, as well as communication and teamwork skills

Remuneration is based on our collective wage and salary agreement. The current monthly salary for this position is EUR 979.00.