Many of my older recordings have significant background noise that I find distracting. I have noticed for the most part that ADD recordings have noise, while DDD have little-to-no noise.
Has anyone else fooled around with noise reduction software to improve the sound of their recordings? I have tried audacity, but can't manage to get a clean output. Either the transients are degraded (as verified with the residual option) or the complex passages near the noise floor are followed by "musical noise". I'm being very picky when I say this. I think 90% of the population would be happy with the noise reduction in audacity, but this forum is probably almost exclusively people in the other 10 percent.
Has anyone had success with commercial products?
I ended up writing my own routine in matlab/octave. Eventually I found a sweet spot for the parameters that resulted in no audible degradation (also confirmed with a residual that consists of the noise that was filtered out). The down side is it takes a while to run, but it's worth it to me.
From what I can tell, there is next to no interest in this, as I can't find any discussions on it from music listeners. The only people who seem concerned are those with home studios.
So I thought I'd try stoking a discussion on my own
I'm interested to know if anyone else is bothered by that noise. For me it's like a constant reminder that I'm listening to a recording and not a live performance. It ruins the illusion. Does that happen to anyone else?
Has anyone else fooled around with noise reduction software to improve the sound of their recordings? I have tried audacity, but can't manage to get a clean output. Either the transients are degraded (as verified with the residual option) or the complex passages near the noise floor are followed by "musical noise". I'm being very picky when I say this. I think 90% of the population would be happy with the noise reduction in audacity, but this forum is probably almost exclusively people in the other 10 percent.
Has anyone had success with commercial products?
I ended up writing my own routine in matlab/octave. Eventually I found a sweet spot for the parameters that resulted in no audible degradation (also confirmed with a residual that consists of the noise that was filtered out). The down side is it takes a while to run, but it's worth it to me.
From what I can tell, there is next to no interest in this, as I can't find any discussions on it from music listeners. The only people who seem concerned are those with home studios.
So I thought I'd try stoking a discussion on my own
I'm interested to know if anyone else is bothered by that noise. For me it's like a constant reminder that I'm listening to a recording and not a live performance. It ruins the illusion. Does that happen to anyone else?
It is one of my retirement projects....ha ha.
I have several single-ended diy projects that I have intended to build. The most interesting is from the old copy of "Audio". (Maxwell Strange? may be) With matched FETs, a pre-filter and 12db/ove variable low pass filters. Intended for 78's but of course adaptable to 331/3 rpm and other formats.
Another has a more sophisticated noise analyzer, again from an American mag', where they argue that their 9db/ove slope is the most pleasant audibly.
I think I have around half a dozen designs.
Have you looked at multi-band expandors? i.e. like a more refined 3dbx. Again one of my future projects. I have several NE572's. In a four band system and a modest expansion ratio that should lessen the subjective noise factor. I KNOW this is all sooooooooooo 70's but if you are prepared to chase up an article from the JAES from that time there is a wonderful piece that goes into different attack and decay times for each band etc.. all designed to minimize the subjective effects. (Look out for an errata a month or so later that corrected some errors in a diagram or chart axis...)
They suggest that a multi-band expandor with a variable low-pass filter is optimum....
But none of the above is software of course...........just empathizing with your frustration.....
For what its worth a friend of mine who is a good amateur muso has a very lo-fi system and tells me it is all psychological and that I should just "tune it out" and focus on the inherent beauty of the music....
Cheers,
Jonathan
I have several single-ended diy projects that I have intended to build. The most interesting is from the old copy of "Audio". (Maxwell Strange? may be) With matched FETs, a pre-filter and 12db/ove variable low pass filters. Intended for 78's but of course adaptable to 331/3 rpm and other formats.
Another has a more sophisticated noise analyzer, again from an American mag', where they argue that their 9db/ove slope is the most pleasant audibly.
I think I have around half a dozen designs.
Have you looked at multi-band expandors? i.e. like a more refined 3dbx. Again one of my future projects. I have several NE572's. In a four band system and a modest expansion ratio that should lessen the subjective noise factor. I KNOW this is all sooooooooooo 70's but if you are prepared to chase up an article from the JAES from that time there is a wonderful piece that goes into different attack and decay times for each band etc.. all designed to minimize the subjective effects. (Look out for an errata a month or so later that corrected some errors in a diagram or chart axis...)
They suggest that a multi-band expandor with a variable low-pass filter is optimum....
But none of the above is software of course...........just empathizing with your frustration.....
For what its worth a friend of mine who is a good amateur muso has a very lo-fi system and tells me it is all psychological and that I should just "tune it out" and focus on the inherent beauty of the music....
Cheers,
Jonathan
Last edited:
Sorry, memory going. There are a few important corrections in that last edit in #2 above.
I think it is right now.....got attack and decay times mixed up with expansion ratios initially.
Not just from the 1970's but I am getting too close to the 70's myself now. Doh.
Cheers Jonathan
I think it is right now.....got attack and decay times mixed up with expansion ratios initially.
Not just from the 1970's but I am getting too close to the 70's myself now. Doh.
Cheers Jonathan
You might look at "Goldwave". trial download. I find that the hiss removal works pretty well there. It is a general purpose audio editor, similar to Audacity, but easier to use and with more filters.
How disturbing the noise is to me is dependent of the actual type of noise. Phono surface noise ticks and pops are annoying. Hum and air handler is bad. Hiss varies with spectrum.
I use Adobe Audition 3. Effects/Restoration/Noise Reduction(process) works well if you can find a spot about 1/10 second of the noise to use as a profile. The 'trick'is to not over do it. I typically run it at 9 dB reduction and the 'noise reduction' slider at 85. I've used this on old analog rock and classical recordings with good to excellent results. Don't be surprised if the noise reduction removes a little reverb but that can be replaced. If you keep a light touch on it no one will know it's been altered. It will just be pleasing.
G²
I use Adobe Audition 3. Effects/Restoration/Noise Reduction(process) works well if you can find a spot about 1/10 second of the noise to use as a profile. The 'trick'is to not over do it. I typically run it at 9 dB reduction and the 'noise reduction' slider at 85. I've used this on old analog rock and classical recordings with good to excellent results. Don't be surprised if the noise reduction removes a little reverb but that can be replaced. If you keep a light touch on it no one will know it's been altered. It will just be pleasing.
G²
I did some more testing, with a couple more classical music pieces. So far I'm comparing my own algorithm to audacity and Izotope RX 5. I'll see about adding other software like those mentioned above.
The results really seem to depend on the source material. There's one piece with a lot of pizzicato on a solo violin. Audacity lost a bit of the transients with any substantial amount of noise reduction. My algorithm was able to maintain the transients with substantial noise reduction, but Izotope was able to keep the transients and reduce the noise another dB or two. Since that was the first piece I tried, I was initially quite impressed with Izotope.
Then I tried another piece with bowed cellos. Izotope can't seem to isolate the overtones. When the strings are bowed, much of the noise comes back until they stop. The noise is reduced during the rests, so it's modulating on and off. Audacity didn't have the same problem, but it has a small amount of "musical noise" during a prolonged diminuendo. On that piece, my own algorithm was superior to either of the commercial products.
The results really seem to depend on the source material. There's one piece with a lot of pizzicato on a solo violin. Audacity lost a bit of the transients with any substantial amount of noise reduction. My algorithm was able to maintain the transients with substantial noise reduction, but Izotope was able to keep the transients and reduce the noise another dB or two. Since that was the first piece I tried, I was initially quite impressed with Izotope.
Then I tried another piece with bowed cellos. Izotope can't seem to isolate the overtones. When the strings are bowed, much of the noise comes back until they stop. The noise is reduced during the rests, so it's modulating on and off. Audacity didn't have the same problem, but it has a small amount of "musical noise" during a prolonged diminuendo. On that piece, my own algorithm was superior to either of the commercial products.
I also use (and like) GoldWave. Its noise suppression routine works, but definitely needs experimenting with the parameters to ensure that its artefacts aren't worse than the noise you are suppressing.
By the way, GoldWave, or at least the version I'm using, has no dithering. You can easily fix this by adding dither yourself with the equation editor before storing files in a format with limited resolution. For example
wave👎+rand(1/y)+rand(1/y)-1/y
with y = 32768 before storing it in a 16-bit format.
By the way, GoldWave, or at least the version I'm using, has no dithering. You can easily fix this by adding dither yourself with the equation editor before storing files in a format with limited resolution. For example
wave👎+rand(1/y)+rand(1/y)-1/y
with y = 32768 before storing it in a 16-bit format.
Last edited:
I have downloaded goldwave. So far I have only tried it on the pizzicato solo violin. It took a long time to adjust it so it wouldn't reduce the transients. I had to put the scale all the way down to 30%. At that point, I would say it was doing better than audacity. My own algorithm was able to reduce the noise an extra couple of dB relative to goldwave (without impacting the transients). Izotope still did the best on that piece. I'll try some other pieces.
Goldwave performed very well on the piece with low SNR bowed cellos. No problems reducing noise during the notes like Izotope, and no tinkle bells during the extended diminuendo. In fact, I haven't heard it produce a tinkle bell (aka musical noise) artifact yet.
I recently did some live recording (voice only), in a live kitchen, with refrigerators and freezers running. Goldwave was able to sample the noise spectrum, and did a very good job of removing the background. Not so good at removing her "esses", though. Enough eq to do that muffled the voice too much. Moved the mic to the side and that helped.
Thanks, jplesset, that brings up a point I wanted to ask about as I'm evaluating these different software suites:
What instruments / sounds are the most revealing of issues with noise reduction software?
What instruments / sounds are the most revealing of issues with noise reduction software?
I find that the simpler the sound is, the more sensitive I am to hearing background noise. Cello, flute, solo voice, piano make hearing background pretty easy. Full orchestra, not so much.
Hi bbutterfield
There was a burst of interest for noise removal and declicking I noticed in this thread http://www.diyaudio.com/forums/analogue-source/298896-digitizing-vinyl.html (from post#58 to post #197)
George
- Status
- Not open for further replies.