Extrapolation of Audiosignals with Non-Linear Methods

Practical project, Bachelor's thesis & Master's thesis

Topic Description
The prediction of future signal values based on past measurement values is known as signal extrapolation. Such methods are familiar from trend analysis in the financial world or weather forecasting, for example. Extrapolation is also used in the audio sector, e.g., to reduce algorithmic latency or in packet loss concealment (PLC), where voice signal sections lost during transmission are reconstructed. In contrast to interpolation, which uses information from the future for reconstruction, this work focuses exclusively on extrapolation, i.e., prediction from past values.

Tasks
Depending on prior skills, this work will explore nonlinear model-based or data-driven methods, e.g., deep neural networks, for the extrapolation of audio signals. The tasks include:

• literature research on relevant methods
• implementation in Python
• evaluation of the implemented methods
• comparison of the different approaches

Required Background

• good understanding of signals, systems, and algorithms
• good programming skills in Python