Blind Room Parameter Estimation in Real Time from Single-Channel Audio Signals in Noisy Conditions
Mixed reality audio processing needs to adapt the acoustic rendering of virtual sound sources to the local listening environment. Such adaptation has been shown to dramatically improve the immersion and plausibility of spatial audio experiences. However, the parameters necessary for the signal adaptation are typically not available a priori for unknown environments. Given an estimate of the reverberation time and volume of an acoustic environment, the perception of virtual sound sources can be adapted through artificial reverberation. Here we focus on the attempt to blindly estimate the local room volume in real time from single-channel audio signals recorded in the environment. A dedicated dataset was prepared using impulse responses from a variety of environments and used to train a convolutional network for an approximate prediction of the volume.
Speaker Bios
PhD Candidate Andrea Genovese
Andrea Genovese is currently a PhD candidate and adjunct teacher at the Music and Audio Research Lab at New York University. His main areas of focus within the acoustics field are spatial audio analysis, psychoacoustics and sound-based interactions in multiplayer XR environments. Andrea holds an MEng degree from the University of York, UK, in Electronic Engineering with Music Technology Systems and has previously been an intern at THX and Fraunhofer IIS.
- Date:
- Haut-parleurs:
- Andrea Genovese
- Affiliation:
- New York University