Top News - Artikel
forum | IKS: Vortrag von Prof. Dr.-Ing. Nilesh Madhu
Machine Learning for Audio
Freitag, 05. Februar 2021
virtueller Konferenzraum - Einwahldaten auf Anfrage
Im Rahmen der IKS‐Veranstaltungsreihe referierte Herr Prof. Dr.-Ing. Nilesh Madhu, Universität Gent über das Thema:
"Machine Learning for Audio"
Incorporating a priori knowledge increases the performance of audio processing algorithms. Classical, knowledge-based approaches depended on hand-engineered features and painstaking statistical models for this. In recent years these have been increasingly superseded by data-based approaches using (deep) neural networks (DNNs). In this talk Prof. Madhu will contrast knowledge-based and data-based approaches for different applications and also discuss some pertinent questions in relation to the use of DNNs such as: How can we allow DNNs to generalize better, and what are some pitfalls? These shall be illustrated on specific applications.
The overall aim of the talk is to present a general (qualitative) appreciation of the two classes of approaches. Perhaps this will stimulate work on how to obtain the "best of both worlds."
Den Vortrag verfolgten knapp 70 Teilnehmer.