Master Thesis: Low-Delay Audio Compression Using Deep Neural Networks

Join our Team

About this opportunity:

Audio compression enables the efficient transmission and storage of digital audio. It is used in mobile and satellite communications, emergency services, broadcasting, and many other applications. Speech and audio compression reduces the number of bits needed to represent a signal while maintaining sound quality appropriate to the application.

Recent work has explored using powerful AI-based autoencoders in both the analysis and synthesis stages of audio compression and prediction. These approaches aim to deliver good listening quality at low bit rates of 1–2 kbps and moderate delays.

In this thesis, you will investigate low-delay audio compression methods and explore how deep neural networks (DNNs) can improve their performance. The work will consider both short- and medium-frame approaches, as well as legacy signal-processing-based codecs designed for very low latency.

What you will do:

• Review existing mono, low-rate audio compression methods using short and medium frame lengths (10–20 ms).

• Review legacy signal-processing-based low-latency codecs using frame lengths of 1–2.5 ms.

• Select a promising core low-delay compression method.

• Enhance the selected method by incorporating DNN-based analysis and synthesis.

• Optimize the resulting algorithm for deployment on hardware with limited computational resources.

The skills you bring:

• You are pursuing a master’s degree, preferably in Computer Science, Electrical Engineering, Engineering Physics, or a related field.

• Coursework in signal processing and machine learning is essential.

• You have good programming skills in MATLAB, C, or Python.

• You are proficient in English.