FIXING in the Mix - I can't

  • Thread starter Thread starter rayc
  • Start date Start date
PM me with a link to the whole track and let me fuss with it. Perhaps a different approach may fix it?
 
Rokket, ta - you really helped a lot with the practical side. The track is usable now.
I'd really like some suggestion from thsoe in the know about how/where the noise comes from in the signal chain...anyone?
http://www.soundclick.com/util/getpl...d=6848840&q=hi
or
http://www.soundclick.com/bands/defa...&content=music
This is a link to a brief sample of the vocal track complete with problems.
I've tried noise reduction based on a sample of the track with no vox - ie" just noise".
I've tried filtering and I've tried EQing. In fact I've done all three in series to get improvement.
Have a listen, any suggestions?
The next post is the freq analysis of the sample unprocessed. You'll note a weird very narrow peak at around 11khz. The rad cut at about 17 would be the MP3 processsing that the file underwent before being sent to me.
Attached Images
 
Finally this is it...
 

Attachments

  • d.webp
    d.webp
    18.7 KB · Views: 101
Ray, I tried checking your files out, but Soundclick is telling me there's nothing there.

Is that freq graph strictly of the noise without the vocal or any other instrument? If so, what I find just as, if not more interesting as that 11k peak is the entire 150Hz-wide harmonic series starting at 150Hz. And no, I don't know where that's coming from, unless you're getting a bass peak mode in the room itself maybe? Just a wild guess, probably something else.

I had a synth composition I got from someone once that had some fairly bad soundcard noise on it. Standard noiseprinting via the Waves noise filter (and the Sonic Foundry one as well, which was worse) wound up artifacting the signal way too much and the synth wound up sounding like a sick whale. A similar FFT analysis to yours showed a harmonic series a lot like yours but at different frequencies. Here's what mine looked like:
attachment.php

Anyway, what I had to do there was use a harmonic filter (in my case, the filter on Roger Nichols Digital's Unequalizer) to knock down those harmonics first. A multi-band parametric EQ could do the job too, but that takes a little more manual setup. Then after removing those harmonics, I was able to take a new noiseprint with the Waves NR plug, and that worked MUCH better with almost no artifacting of the original signal.

G.
 
Last edited:
http://www.soundclick.com/bands/default.cfm?bandid=469622&content=music
Glen, ta, Soundclick seems to be throwing up probs all over the place at present. The above link ought to go to the main page & the file should be listed there as janglethan vocal sample.
The chart above is just the vocal track.
Below is the print of NOISE ONLY:
Yeah the harmonics stuff is pretty weird. I threw a low shelf on it when trying to work on the processing. The charts are unprocessed.
This is from my co-writer & singer's soundcard. It appears to be bog standard generic but of indetermiant age etc (the machine runs XP but that's not particularly new afterall). Terry's attempting to isolate the problem by recording with various stepes of his setup out of the chain. He has a very nice Sennheiser mic into a little Behri mixer into the soundcard.
You've given me some interesting info & a way forward. If I can isolate it to being the soundcard I might post the result in the thread about how generic s/cards aren't so bad.
On top of the obvious there's also a considerable DC offset when the mic is turned on as opposed to sections of lead in/oout where the mic is off.
Thanks for the insight so far.
Cheers
ray
 

Attachments

  • jangle noise only2.webp
    jangle noise only2.webp
    15.8 KB · Views: 81
Notice how the noiseprint alone has a definite harmonic series of it's own? Definitely noticeable at 500Hz, though there are possible odd subharmonics of 1/5th and 1/3rd at 100Hz and 166Hz. Then above 500Hz there's a regular harmonic series spaced at exactly 500Hz intervals after that.

I did also notice the DC offset reflected in the large rise at the left end of the graph on both charts, and, yeah, DC offset, while not necessarily common, is not exactly uncommon either in consumer-grade soundcards - especially laptops.

Also, while I'm not sure if this is science or not, it's possible maybe that the harmonics are a result of the MP3 encoder. It's a shame that all you have to work with is the MP3 file. If you can maybe get your man to noiseprint the original WAV, that would clear that up right off.

And, yeah, the 11k peak sticks out like a sore thumb now, along with a somewhat smaller spike around 15k. Those are probably unrelated to the harmonics, but they are real issues nonetheless.

G.
 
Last edited:
Glen,
THANKS. That's nice analysis. I didn't look at the intervals, I didn't even look for a pattern. I just obsessed over the 11khz peak & didn't notice the 15 - but it is quite clearly there.
I'll get Terry to send me a 10 second wave sample from his recording so I can do the chart, (I don't think he has the software for that). I'll let you know how the signal chain test and wave freq chart come out.
I had figured that a corrupted LAME thingo might cause some grief but you're right: it's a maybe in italics indeed.
Thank you very much sir!
 
rayc said:
I didn't look at the intervals, I didn't even look for a pattern.
Click to expand...
When there's a situation where what should work doesn't (like with my noise reduction situation), when even the the most dedicated and surgical EQ plan just doesn't seem to get it - like an itch that just won't scratch, or when wider EQ scooping takes out TOO much, look for such a pattern.

Look at your chart and think about it for a sec; unless you know the pattern is there and exactly where it is, attacking the problem with parametric EQ alone will be incomplete. Even if you managed to find the 500 and 1k bumps with a parametric sweep, the total energy distributed in all the remaining peaks exceeds what you've taken out with those two, meaning that you're still going to have an audible problem. And if you keep sweeping very carefully, the number of peaks you find will be daunting; you'll surely give up before you're done (unless you just happen to notice during the process that the bumps are evenly spaced and begin to notice the pattern that way.)

And if you try to remove that stuff with a graphic or a wide Q scoop, 90% of what you remove will be the good stuff you want to keep. So that doesn't work either.

That's when I turn to the FFT graph (and anyone who's been around here for a while knows that's not something I do very lightly) to see what's going on, and the easiest way to find the culprit is often to look for a non-random pattern in the graph. You gotta make sure it's a real pattern mathematically, but if it's there, it should pretty much jump out at you.

Now, there's no guarantee that this will completely solve your problem Ray (I still haven't had a chance to check out your file yet, I'll get to that later, hopefully), but I'll bet it'll help, at least.

G.
 
You must log in or register to reply here.
Back
Top