Page 1 of 1

Linearly averaged noise profile

PostPosted: Mon Jan 17, 2005 9:46 pm
by schiller
I mainly use WC on shellac, and there, due to all the parameters of the format and the scanning process, a noise profile sampled in the lead-in will typically differ much from a noise profile as found in the lead-out.

When de-noising with one of these, it is typically found to work quite well close to where the profile was sampled and degrade in efficiency the further away one gets.

HENCE:

Why not allow to specify TWO noise samples N1, N2 say at times t1, t2 (beginning and end of track) and employ a time-dependent noise profile that is a linear mix of the two when de-noising the signal in between:

N(t) = N1x((t2-t)/(t2-t1)) + N2x((t-t1)/(t2-t1))

PostPosted: Tue Jan 18, 2005 10:08 am
by Derek
Ah, that's interesting. The noise must be varying according to the stylus velocity. I have never considered that before.

Your suggestion would be quite easy to implement. My only hesitation would be that I wouldn't have expected the hiss filter to be very effective anyway against this type of 'shellac' noise. I'd have thought the 'spikiness' of the noise would cause a lot of unwanted artefacts. Have you not found this to be the case?

PostPosted: Sat Jan 22, 2005 1:41 am
by schiller
Yes, I think the absolute stylus velocity is the most prominent cause. After all: with the big 25cm records, outer to inner is about 3:1.

With disks that are either being scanned with steel needles (and recorded through a mic) or those that have been scanned that way before, there is perhaps a second source of time dependent noise spectrum: Their outer radius, when starting the scan, will meet a fresh and standard-conforming stylus, which then steadily wears off and snug-fits itself to the groove profile at hand.

You are right in assuming that de-hissing shellac is sort of sensitive. Using default settings does indeed somewhat "overdo" and may give rise to artificial sound effects. It sort of depends on the granularity of the meterial. But with a noise profile generally damped down I found it working. The more important it is to keep the noise model close to what really is present momentarily; hence my idea.

Rgds
Harald

PostPosted: Sun Jan 23, 2005 3:35 pm
by Derek
I'll see what I can do in the next release, although I'll not promise because I think this is rather a specialised requirement.

As I said before, it should be quite easy to implement. However, it'll need to be done in a way that doesn't confuse the novice users.

PostPosted: Mon May 23, 2005 6:30 am
by mannie.gross
The way I have tended to use this feature is to separate the music into individual tracks (not always possible) and then apply the hiss removal to each track separately. If it were possible to interpolate the noise to remove between tracks in software it would be great.

Thanks and regards,
Mannie

PostPosted: Wed May 25, 2005 8:06 am
by Derek
I'm not sure I fully understand your request. However, it is not really practical to run the hiss filter on track files after an album has been split. This is because the track files will not have sufficient lead-in/lead-out period that is essential for the hiss filter to work effectively.

Sorry.

PostPosted: Wed May 25, 2005 8:25 pm
by mannie.gross
How much time is required for a noise profile?

Regards,
Mannie

PostPosted: Wed May 25, 2005 8:51 pm
by Derek
The auto-profile is about 200ms. If you manually select a profile it has to be in the range 100ms to 1 sec.

PostPosted: Thu May 26, 2005 5:29 am
by mannie.gross
I apologize for causing confusion. What I was trying to suggest was an extension to Schiller's original post. If there are say 5 tracks on the side of an LP then automatically extend the formula to t6 and N6. However so long as the end user can pick the points to sample for noise I would be very happy with this feature as originally suggested.

Thanks and regards,
Mannie

PostPosted: Wed Nov 02, 2005 8:05 pm
by mannie.gross
Hello Derek,

Any chance you might implement this suggestion in the next release? It seems to have fallen off the radar.

Thanks and regards,
Mannie

PostPosted: Thu Nov 03, 2005 10:06 am
by Derek
Yes sorry, I did not get around to it in the current release. It is still in the To Do list so yes it should appear in the next release.

PostPosted: Mon Jul 31, 2006 8:41 pm
by mannie.gross
I don't think I can see this in the just released beta. Pity, I was looking forward to this feature.

Regards,
Mannie

PostPosted: Tue Aug 01, 2006 7:12 am
by Derek
Yes, sorry about that. I am still looking at it and it could still be included in this release cycle. It was given a low priority because I'm not sure how big a requirement there is for this.

Do you (or anyone else) have a sample file that exhibits this phenomenon, ie where the noise varies linearly between the start and end of the file? I'd like to be able to test the effectiveness of any change I make. We have an ftp site that you could upload to. You can pm me for details.

PostPosted: Tue Aug 01, 2006 9:13 pm
by mannie.gross
Sorry, I don't have one that I know of. How would I tell? I suppose what I am hoping for is something that will give me better noise removal. I could test this by listening to the results.

Regards,
Mannie

PostPosted: Mon Aug 07, 2006 9:44 am
by Derek
The latest build of the beta now implements this functionality. See http://www.wavecor.co.uk/wavebeta.html