Blind De Reverberation

Started by jalopez14 3 years ago5 replieslatest reply 3 years ago107 views

I am currently doing some research into automatically removing reverb from audio files related to cleaning up speech. Is there any pointers you guys can share? I am looking into a couple papers from AES but was wondering if maybe there is a popular/standard way of doing this. Thanks

[ - ]
Reply by rbjJune 4, 2020

i remember seeing such papers in JAES.

had something to do with recognizing later replicas of earlier features.  i think from that you might be able to infer possible tap locations and tap coefficients for specific reflections and simply undo those taps by adding your own taps with the arithmetic negative coefficient.  but that's only the early reflections not the later reverberation.

best as i can tell, the problem is a female canine.

[ - ]
Reply by jalopez14June 5, 2020

thanks for the reply, currently looking at a method using IIR filters to remove the reverb component 

[ - ]
Reply by rbjJune 5, 2020

i dunno how that's gonna do it.  blind de-reverb removal requires some kind of automatic learning regarding your input-output relationship.  the effect of reverberation is likely broadbanded.

just removing some frequencies won't do it, but if you find that a room is sorta "tuned" to a frequency (we call those "room modes", it's like the room is ringing), you might be able to use an IIR notch filter (or a parametric equalizer in cut mode) to take out that annoying frequency.

[ - ]
Reply by jbrowerJune 4, 2020

Ja Lopez-

This music.stackexchange page gives some popular ideas -- as RBJ implies there probably isn't a standard method, and you may need bigger guns.


[ - ]
Reply by jalopez14June 5, 2020

I see, and yes that was my initial assumption, that there isn't a standard way of doing it.