Blind Estimation and Dynamic Binaural Rendering of Multichannel Room Impulse Responses (en)
* Presenting author
Abstract:
For the perceptually convincing rendering of a virtual sound source among real sound sources in a given real-world environment, as required for auditory augmented reality, the acoustic properties of the virtual environment must match the real-world acoustics. However, the acoustic properties of a user’s environment are typically unknown and must be estimated. We recently proposed a method that facilitates the blind estimation of multichannel room impulse responses from a few seconds of recorded speech. The method was evaluated for various types of microphone arrays, such as spherical, equatorial, and head-worn arrays in a glasses form factor. The estimates were shown to reproduce acoustic properties like the reverberation time and the direct-to-reverberant energy ratio with low error and to be perceptually comparable to a measurement from the real-world environment. In this contribution, we summarize the recent advances and describe a flexible way to achieve dynamic binaural rendering of the estimates for arbitrary microphone arrays by exploiting measured anechoic array transfer functions and an end-to-end magnitude least squares binaural rendering approach.