Re: making a sound unrecognizable (Brian Gygi )

Subject: Re: making a sound unrecognizable
From:    Brian Gygi  <bgygi@xxxxxxxx>
Date:    Tue, 12 Sep 2006 11:31:08 -0700

Hi Ursula,

As Valeriy mentioned, a few years ago I created some "chimerical" 
environmental sounds by combining the envelope of one with the spectrum of 
another (different from the "chimerical sounds" in Lewicki, 2002).  While 
they were unrecognizable to untrained listeners in open set format, 
listeners who were trained on the original sounds were remarkably good at 
selecting at least one of the sounds that went into the composite (most 
often they could name the sound that had contributed the spectrum).  So in 
order to make a sound totally "unrecognizable" you must greatly obscure 
both the spectral and temporal structure, at which point it is a valid 
question as to whether it is that "sound" anymore.

Fastl also did some manipulations on train sounds to make them 
acoustically similar to train sounds but not recognizable as train 
sounds.

If you like I can send you the Matlab code

Brian