Re: Removing transients in rejoined speech files (Jan Schnupp )


Subject: Re: Removing transients in rejoined speech files
From:    Jan Schnupp  <jan@xxxxxxxx>
Date:    Mon, 4 Dec 2006 18:56:58 +0000
List-Archive:<http://lists.mcgill.ca/scripts/wa.exe?LIST=AUDITORY>

Dear Allan (and Pawel), simply joining at zero crossings brings NO guarantee that you won't have click artifacts. If I take a sine wave and I cut it off at the zero crossing then I still have a 'sharp edge' at the zero crossing point where the signal snippet ends abruptly, and this will produce 'spectral splatter' and hence a potentially audible click artifact. You cannot completely avoid such transient artifacts at the joints of your cut and shuffled samples, but you can reduce them very considerably by cross-fading, and using something like a raised cosine ramp is likely to be better than a linear ramp. You are basically dealing with what signal engineers call a 'windowing' problem, and the literature on this is vast, so we could debate endlessly on what the 'ideal' approach would be, but a 3 ms cross-fade, as Pawel suggests, is likely to work well enough and is unlikely to raise too many eyebrows among potential reviewers. Jan On 04/12/06, Pawel Kusmierek <p.kusmierek@xxxxxxxx> wrote: > Allan J Goldstein wrote: > > I have tried looking for zero > > crossings and making the cuts at these points, but > > there is still the interference. > > Cutting at zero crossings should solve the problem... But, do you just > check for zero crossings, or for zero crossings in one direction? I mean, > if you do one cut at a zero crossing where the signal goes from negative > to positive, and the other one where the signal goes from positive to > negative, then you can get a strong transient in a rejoined file: the > signal will "bounce" from the zero line. > > If this does not help, about a short (1-3 ms) crossfade - unless it > interferes with your purpose. > > Pawel > > > > -- > Pawel Kusmierek PhD > Department of Physiology and Biophysics > Georgetown University Medical Center > The Research Building WP23 > 3970 Reservoir Road NW > Washington, DC 20007 > phone: +1 202 687-8851 > > -- Dr Jan Schnupp University of Oxford Dept. of Physiology, Anatomy and Genetics Sherrington Building - Parks Road Oxford OX1 3PT - UK +44-1865-272513 www.oxfordhearing.com


This message came from the mail archive
http://www.auditory.org/postings/2006/
maintained by:
DAn Ellis <dpwe@ee.columbia.edu>
Electrical Engineering Dept., Columbia University