Acoustic Scene Recognition

Description Acoustic scene recognition is the task of identifying an audio environment. Applications include devices that require environmental awareness, based on social or physical context, e.g. metro, office, park.
Task In this thesis, the student(s) will be extending a current RNN encoder-decoder model to develop a conditional recurrent sequence to sequence autoencoder for deep unsupervised representation learning suitable for use with acoustic data.
Utilises Conditional Variational Autoencoders, auDeep, Deep Spectrum System.
Requirements Preliminary knowledge in Machine Learning, Good programming skills (e.g. Python, C++).
Languages English and German.
Supervisor Shahin Amiriparian, M. Sc. (