On the Estimation of Early Reflections from Multichannel Impulse Responses using Deep Learning Techniques
* Präsentierender Autor
Zusammenfassung:The acoustic reflection coefficients of surfaces are important parameters for geometric modelling of rooms. Various methods have been proposed to estimate these from multichannel impulse response measurements using one or more sources and microphones distributed in the environment of interest. Based on measurements from rigid spherical microphone arrays, various beamforming techniques have traditionally been applied to address this problem. Deep machine learning (ML) approaches have proven to be efficient and robust in many tasks. We elaborate on the design, training and application of ML to estimate the incidence angles and times of early reflections. The influence of practical limitations, like the number of microphones and their self-noise, is investigated to assess the robustness of the approach.