On 3 Dec 2006, at 04:04, William wrote:
> You were talking about the relaxation of the stylus after a click.
> Is the
> length of the click determined at all by the amplitude of it?
> Bigger click,
> longer relaxation.
Absolutely. The monster clicks that get clipped, can take up to a
hundred sample points (two thousandths of a second) to recover. The
small clicks (which may be as small as a single sample point, but
audible none the less) are often of less amplitude than the
surrounding music. Both present problems.
> Is there any way to determine from the click the
> direction of the needle vibration? Seems some directions would
> make worse
> clicks than others.
On my system, clicks are always a jump upwards (increase in value).
As well, usually both tracks are affected, but seldom equally. And
there seems to be bias towards track 1. I am sure that there is a
> Just realized something, was about to delete this when I wondered,
> measuring the resonant frequency of things present help determine the
> smaller spikes?
I am sure that the stylus acts like a damped spring and that one
could model its action. I may do that. The smaller clicks tend to
follow a more regular pattern. On the other hand, the big clicks seem
to be like snow flakes - no two alike. I think that the problem there
is with the scratch causing the click. Some clicks are extremely
violent, with two or three oscillations before recovery.
> I'm not sure how octaves and such work together, but from a
> different understanding that I have of things, wouldn't a pop front
> end be
> like a square wave, with too many too strong harmonics in it from
> the main
> signal? If it picks up on that, then analyse to see if you've got
The ear seems to operate with frequencies (All those little hairs
resonating to different frequencies) and I am sure that we detect
clicks as a sudden burst of amplitude at all frequencies. I would
love to reproduce that on a computer. I could detect a lot more than
just clicks that way. But the computer operates linearly on overall
amplitude and calculating and analysing a moving spectrum is just not
very fast. The problem is hardware not software. It would be
interesting (and possible) to build a sort of microphone that works
like the ear, sending a signal that is the result of an array of
My program goes a little crazy with artificially produced square
waves. It looks at them as really big clicks and tries to remove
them. The front end of a click usually happens from one sample point
to the next. That is why looking at the second difference is
effective. The acceleration is enormous.
> I come up with things from weird angles sometimes. Hope I helped
> and didn't
> just waste time.
Not at all. My program is based on my experience looking at economic
time series, which is also a weird angle.
> ----- Original Message -----
> From: "John Butterill" <butterillj@...>
> To: <audacity-nyquist@...>
> Sent: Saturday, December 02, 2006 8:36 PM
> Subject: [Audacity-nyquist] speed and clicks
>> I joined this board with the specific intention of asking about a
>> problem I was having with a nyquist program. But in formulating the
>> question, I found the answer, and so I am now posting this message to
>> thank the board, for its mere existence I suppose, and, since it may
>> be of interest, to briefly describe the problem, the solution, and
>> the program.
>> I periodically use Audacity to rip vinyl to digital. The vinyl has
>> clicks (pops) and crackle which is annoying. The click removal
>> routine in Audacity (version 1.2.4) doesn't seem to do anything. I
>> discovered that I could write my own plug-in, in nyquist. So, based
>> on my thirty years experience with time series analysis, I wrote my
>> own click removal program.
>> When a stylus hits a scratch, it is violently jerked aside, much more
>> violently than would happen from a natural sound wave. My program
>> scans the sound sample array for second differences (acceleration).
>> Anything beyond ten standard deviations, is invariably a click. The
>> program then determines the bounds of the click and replaces it with
>> a nice fitted cubic curve. Since most clicks are less than a milli-
>> second in duration, the patches are, to my ears, inaudible and often
>> even invisible when graphed. Anything outside the patch is untouched.
>> The method is about 90% effective. What is modified are only obvious
>> clicks, but some clicks are too soft to be detected in this way. I
>> would rather leave some clicks than accidently modify the music. As
>> well, the relaxation of the stylus after a click can be quite complex
>> so that it is often difficult to detect the extent of a click. There
>> is still some work to be done.
>> I would be quite happy to discuss the methodology in more detail with
>> anyone who is interested and to share the program. I am not familiar
>> with how other click removal programs work, but I believe that some
>> apply filters, presumably intending to smooth the clicks. This would
>> also smooth out the legitimate sounds (or noise) which, to me, is not
>> The problem which prompted this note, was that, for a ten second
>> sound sample, the program required about three minutes to run the
>> first time and twenty seconds subsequently. This was not good. And
>> the problem persisted even when I did no calculations beyond read and
>> write. I also noticed a lot of garbage collection, so I began to
>> suspect a memory problem. A few minutes with Google found a reference
>> that noted that (for a Fourier transform) speed considerations
>> required the "expand" command. I discovered that "(expand 5000)" made
>> the program usable.
>> However, even with the fix, the program runs at double real time -
>> that is ten seconds of music requires twenty seconds of processing.
>> This is ok for experimenting, but not for a forty-five minute album.
>> Obviously it needed to be compiled, and since nyquist doesn't seem to
>> be compilable, I transcribed the program into Java, my usual
>> language. The Java version reads the WAV file directly. After some
>> tuning, this version is running at about nine percent of real time
>> (about four minutes to process fifty minutes of music), which is
>> John Butterill
>> Take Surveys. Earn Cash. Influence the Future of IT
>> Join SourceForge.net's Techsay panel and you'll get the chance to
>> opinions on IT & business topics through brief surveys - and earn
>> Audacity-nyquist mailing list
> Take Surveys. Earn Cash. Influence the Future of IT
> Join SourceForge.net's Techsay panel and you'll get the chance to
> share your
> opinions on IT & business topics through brief surveys - and earn cash
> Audacity-nyquist mailing list