Search papers, labs, and topics across Lattice.
This paper extends a diffusion-based inpainting framework to interpolate Room Impulse Responses (RIRs) for microphone array processing. The key finding is that this interpolation method enhances the performance of multi-microphone array processing tasks, even when using real-world RIR data. This demonstrates the potential of diffusion models for handling missing or incomplete spatial audio data.
Diffusion models can now reliably fill in the gaps in real-world spatial audio data, boosting the performance of microphone arrays.
Room Impulse Responses estimation is a fundamental problem in spatial audio processing and speech enhancement. In this paper, we build upon our previously introduced diffusion-based inpainting framework for Room Impulse Response interpolation and demonstrate its applicability to enhancing the performance of practical multi-microphone array processing tasks. Furthermore, we validate the robustness of this method in interpolating real-world Room Impulse Responses.
