AES SGS Research Colloquium #3: From Algorithmic to Neural Beamforming – Two Approaches to Multichannel Speech Separation in Pro Audio

Attendees: Elena Shabalina, Rafael Kassier, Jonathan Ziegler, Beat Hohmann, Jamie Angus, Ryan Corey, Martin Rieger, Thomas Lechner, Rozenn Nicol, Oliver Sampson, Jim Cox, Reinhard Gretzki, Chris Hold, Thomas Gmeiner, Tobias Goldmann, Lukas Benedicic, Boris Rehders, Jeroen De Rycke, Alexander Goldmann, Till Gerneth, Greg White, Zane Rusk, Karim Sune, Javier Frutos-Bonilla and Andrea Rodriguez among others

Jonathan’s presentation focused on microphone arrays that use beamforming to enhance their output signal by manipulating and combining the individual sensor signals. He analyzed an algorithmic and machine learning approach for different microphone array configurations. Under laboratory conditions and in practical applications, he examined the tracking performance, which he also made audible to the audience through several audio examples. For more complex applications, he developed a neural network beamforming approach. Jonathan explained the basic network architecture and highlighted advantages of this method. In addition, various aspects of the usability of the approaches in different applications were emphasized.

We thank Jonathan for a very impressive talk with a professional and well understandable presentation of this complex topic. We wish him all the best for his PhD defense.

A recording of the presentation is available on YouTube.

More information about Jonathan’s work