Acoustic Property
Acoustic property research focuses on characterizing and modeling sound propagation in various environments, aiming to accurately estimate parameters like reverberation time and room dimensions from audio recordings. Current research emphasizes developing robust and efficient algorithms, often employing neural networks (including attention-based models) and self-supervised learning techniques, to estimate these parameters even from noisy, single-channel signals. This work is crucial for advancing applications in areas such as audio processing, speech recognition, and virtual/augmented reality, where accurate acoustic modeling is essential for realistic sound reproduction and human-computer interaction. Furthermore, understanding and quantifying acoustic properties is vital for improving the design of spaces for optimal sound quality and human experience.
Papers
Machine-learning applied to classify flow-induced sound parameters from simulated human voice
Florian Kraxberger, Andreas Wurzinger, Stefan Schoder
Realistic sources, receivers and walls improve the generalisability of virtually-supervised blind acoustic parameter estimators
Prerak Srivastava, Antoine Deleforge, Emmanuel Vincent