A first-of-its-kind acoustic simulation platform for audio-visual embodied AI research. It supports training and evaluating multiple tasks and applications.
However, in the paper, I have some concern about the rendering process. In the paper about the detail of acoustic rendering process (section 3.1) It says,
The simulation operates in M logarithmically-spaced frequency bands (configurable), where it computes an energy-time histogram at the audio sampling rate.
My understanding is you first simulate the energy-time curve using the method mentioned in paper cited https://dl.acm.org/doi/pdf/10.1145/2980179.2982431 here. And then you would simulate the time domain signal by first assiging energy to different frequency band. And then I guess you are using some fft method to get back to time domain? But I am interested in the details about this process. Could you possibly justify my assumptions and provide more details?
Hi @ChanganVR , Really great work on SS 2.0!
However, in the paper, I have some concern about the rendering process. In the paper about the detail of acoustic rendering process (section 3.1) It says, The simulation operates in M logarithmically-spaced frequency bands (configurable), where it computes an energy-time histogram at the audio sampling rate.
My understanding is you first simulate the energy-time curve using the method mentioned in paper cited https://dl.acm.org/doi/pdf/10.1145/2980179.2982431 here. And then you would simulate the time domain signal by first assiging energy to different frequency band. And then I guess you are using some fft method to get back to time domain? But I am interested in the details about this process. Could you possibly justify my assumptions and provide more details?