Audacity Forum

<<<Erm....yes I was getting a bit embarrassed at the length of this ... no idea why it should have attracted so much interest, except that many, many people have problemsssssss with their ....sssssibilants.>>>

Not at all. Trust me, if we weren't enjoying this enormously, the thread would have stopped dead by itself days ago.

<<<I am hoping that you will allow me to use a recording of your voice for testing purposes.>>>

We are hoping you'll let us do that.

What he said.

Straight recording as best you can do with no processing. Therapeutic cats and all.

You don't have to host anything. I'm pleased to post the performance on my web site--kozco.com. We just need to get the WAV file to California. Can it be burned to a data CD? I'll post my mailing address (and send you a blank CD and mailer if you want).

...And redouble my efforts to get my hosting company FTP (file transfer) service to work--by far the easiest way out of this. I'm paying for it and it's never worked right.

Koz

<<<Koz will think he's the only one who got the link>>>

Koz got a non-working link......

See messages.

Koz

<<<I intend to have a go at making a de-esser plug-in>>>

There must be a reason this is so hard to do. Yes, I'm listening carefully to what happens during the damaged portions and what makes them key in the performance.

IF you have a sudden increase in high frequency energy, AND suppressed low frequency energy (note that sibilance and low frequency voice energy are never there at the same time), AND the overall level is above a certain point, AND the average noise level is low, THEN.....what? What steps do you take to help the performance? Apply the Drum3 equalization curve?

Don't forget look-ahead, so you don't get leading edge ticking.

It is an adventure.

Koz

Progress report.

I've done some initial experimentation using nyquist for de-essing. So far I have been running multiple passes with different scripts which has just been to test my basic algorithm. I've still got the hard work of programming it all into one script and designing a usable, versatile (works on many different voice types), but simple (user friendly) interface.

Here are a couple of samples to demonstrate the method. The first is the original audio sample, normalized and converted to mono. Normalizing the sample first makes it far more simple to calculate levels. Converting to mono cuts out a few lines of code, speeds up processing time (my computer is very slow - just 500 MHz processor), and halves the size of the uploads/downloads. Once I have the plug-in, it will be simple enough to make it capable of processing stereo.

The second sample is a direct copy of the first (Ctrl+D) but has been processed with my de-esser algorithm. For my personal tastes I have over processed it (applied too much de-essing), but that is deliberate for demonstration purposes. the final interface will include a slider for selecting how much of the effect should be applied, hopefully with a default value that will be about right for most cases.

The samples can be downloaded from here: http://easyspacepro.com/audacity/deesser/

(comments welcome)

Just to throw mud in the game, did you do that from the original sound clip? I think I recognize an early stab at noise reduction in there where I didn't get rid of all the artifacts, so the clip has competing patches.

Here's the original sibilant, noisy WAV sound file.

http://www.kozco.com/audacity/Drumbeat% ... 20PART.wav

...and if you're not up to typing all that in, I stuck a link at the bottom of my home page.

http://www.kozco.com/

Koz

I will probably post that clip, too. I'm messing about with my web pages (bring them up to W3C Validation) and doing a project for work and trying to keep up with this thread.

<<<I'm glad I'm here and not there.>>>

Manchester? It's raining, right?

Koz