[LAU] Dereverberation

Fons Adriaensen fons at linuxaudio.org
Mon Sep 12 12:07:10 UTC 2016


On Sun, Sep 11, 2016 at 04:14:33PM -0400, Ivica Ico Bukvic wrote:

> Does anyone have access to an easy to implement (as in code)
> dereverberation algorithm assuming one has access to the room
> impulse response? I am looking for a way to clean-up recorded audio
> signal in order to improve signal clarity (e.g. speech). Any info on
> this topic is most appreciated.

This is not an easy matter. There is no single 'room impulse response',
there is a different one for every combination of source and receiver
position. And even these can't be assumed to be constant except maybe
at low and low mid frequencies.

So in practive any given room IR is just a hint, and algorithms
that work need to be adaptive. They usually work in the frequency
domain, and build up statistics of dynamics in each frequency band.
Using these it's possible do attenuate some of the reverb. How 
much depends strongly on the source material. This is still a very
active research topic. Most research is related to speech pickup,
not music. 

Ciao,

-- 
FA

A world of exhaustive, reliable metadata would be an utopia.
It's also a pipe-dream, founded on self-delusion, nerd hubris
and hysterically inflated market opportunities. (Cory Doctorow)



More information about the Linux-audio-user mailing list