I have a dataset of single-channel audios containing speech, recorded in a single room. I need to extract the room IR's. The goal is to simulate the room acoustics and adjust new speech audios recorded with close microphone such that the reverberation is similar to the room in question. What are my best bets? I read a few papers about the topic, but they mostly assume that the distance between the speaker and the microphone is known and that the speaker produces white noise. I am looking for more literature regarding this problem. The papers I read:

  • 1
    $\begingroup$ Hi Jav- Most approaches would utilize a known undistorted souce and then the copy of this source as received by a microphone- using those two signals we can determine the channel impulse response for the specific channel between the speaker and microphone, at the frequencies for which the source has spectral occupancy (thus the desire to have either a white noise source or a frequency chirp which would both optimally cover all frequencies equally). $\endgroup$ – Dan Boschen Sep 9 '20 at 13:26
  • 1
    $\begingroup$ DanBoschen is absolutely right. The best bet would be to have a "reference" signal and then an "output" signal and by "comparing" them extract the effect the channel (room + mic + position) has on the reference signal, which for a linear and time-invariant system (you assume your room is such a system) would completely describe it. Now, the first paper you propose uses a Bayesian approach to estimate the room impulse response, the second seems to need a microphone array (just managed to read the abstract) and the last one is most probably the most promising one, since it uses deep learning. $\endgroup$ – ZaellixA Sep 22 '20 at 11:54
  • $\begingroup$ thanks. the problem is i do not have the reference signal. $\endgroup$ – JAV Sep 22 '20 at 13:58

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.