Hi, I’m new here. Am a 50’s some audio amateur, need some help.
I recorded a lovely chorus+piano concert this weekend on my Oly DM-520. Having previously used mp3 and auto, this time I wanted better quality - so used PCM settings, manual audio, internal mike, did a sound check, etc. .
But after listening to first half concert I determined some loud passages were distorted (apparently over-high gain on the signal). Switched to auto, and seemed better (but one can hear the AGC, alas).
Back home in Audacity I used “find clip” of 1.3x beta - Audacity shows no clipping - but the waveform is clearly flattened at the top/bottom.
I used A’s plug-in Clipfix and it seemed to make no difference.
I’ve read messages here, and elswhere - and this type of distortion is tought to deal with, I gather.
Is there any way to
a) recover some of the natural waveform (clipped volume) … and
b) remove the staticky sound of the distortion?
An expander will make peaks more peaky : it increases the gain above a threshold, [It won’t make a peak out of a plateau though].
But I wouldn’t get your hopes up of being able to cure this type of distortion.
An expander performs the opposite function [to a compressor], increasing the dynamic range of the audio signal.
If it is only very slight, then the clipfix.ny plug-in can be very effective (should only be used on short sections to “recover” the occasional clipped peak). By the time the distortion is really noticeable it becomes a salvage job rather than restoration.
in general not really.
if you clipped then you lost the information needed to change anything.
if you didnt clip for more than a couple of milliseconds maybe you can fix it good enough.
don’t know about oly but zoom H series have input gain settings high low and mid , plus a knob to tweak volume. Most audio recorders opposed to voice oriented ones will also have such features.
always try to record -18 to -24 dB lower than your expected max.
you can always amplify it later to max the volume.
if you have a really wild dynamic range then you may have to also use some compression option which is a smaller evil than clipping.
The “ideal” amount of headroom depends on several factors. You only ever need >0dB headroom - the important thing is that clipping distortion is avoided.
The more headroom that is given, the less dynamic range is available.
Allowing for lsb inaccuracy and overhead from filters, 16 bit provides < 90dB dynamic range. If you intend to allow 24dB headroom, but estimate the level a few dB on the conservative side, you could easily be restricting the dynamic range to less than 60dB, which is lower than can be achieved with a good cassette recorder.
Greater degrees of headroom can be allowed when using equipment at greater bit depths without significant detrimental effects, but at 16 bit there is still the need to aim for a “substantial” recording level. In part the “ideal” amount of headroom to allow will depend on your confidence in how accurately you can estimate what the maximum peak level will be. In this sense “headroom” is essentially allowing room for error.
An expander performs the opposite function [to a compressor], increasing the dynamic range of the audio signal.
You can attach afew seconds of mp3 of a distorted bit to your next post, [file size less than 1Mb].
Thanks to all for your advice - especially to keep some headroom. (If 24db is too much to retain full dynamic range, maybe 10-15? my device doesn’t have numbers/marks to indicate, so I imagine the moving bars s/be about 10-15% below the peak)
Attached is brief sample from two voices … listen carefully to hear the crackly top.
If an expander merely expands (say) the clipped parts by, somehow, sensing what would have been higher amplitude, that won’t help. What I most want to do is remove the distortion sound effect - the crackly or staticky sound that comes from overmodulating the signal. Is there a noise remover that works on say 10-15 sec sections to pull out only that sound?
Woodhands you’ve attached an Audacity (.aup) file which doesn’t contain any audio.
To attach audio to a post it needs to be in mp3 or wav (or another audio format put in a zip file), not bigger than 1Mb, (typically that’s only about 10 seconds).
If the peaks are clipped flat, (rather than just a bit squashed), then yes an expander is no use " it can’t convert a plateau into a peak".
If they are clipped absolutely flat then clipfix is your only hope: it estimates what the clipped peak should have looked like by interpolation.
NB: you may have to reduce the amplitude of the waveform before applying clipfix to give headroom for the clipped peaks to be drawn in by clipfix.
If the peaks are clipped flat, (rather than just a bit squashed), then yes an expander is no use " it can’t convert a plateau into a peak".
If they are clipped absolutely flat then clipfix is your only hope: it estimates what the clipped peak should have looked like by interpolation.
Understood.
NB: you may have to reduce the amplitude of the waveform before applying clipfix to give headroom for the clipped peaks to be drawn in by clipfix.
Yes, I did that. And clipfix in 1.3x beta didn’t seem to do anything.
Hmmm … I thought the Audacity file included the data in a new format; is it just a file with data for manipulation but with no sound in it? How does that work?
Sorry … here is the mp3 clip of two different voices, recorded sections.
Not a lot that can be done. I’ve butchered the spectrum, (have a look at the before and after frequency analyses to see what frequencies I’ve amputated),
and applied “Floorfish” (free) expander at different frequencies …
I slso tried “centre pan isolation” using Kn0ck0ut, it has less overload distortion but sounds computery, (digital artifacts), basically swapping one type of distortion for another …
do that and you will clip a lot unless you record a flat line signal
headroom ensures the peaks don’t hit 0dbFS
that is why it is called headroom
The more headroom that is given, the less dynamic range is available.
Allowing for lsb inaccuracy and overhead from filters, 16 bit provides < 90dB dynamic range. If you intend to allow 24dB headroom, but estimate the level a few dB on the conservative side, you could easily be restricting the dynamic range to less than 60dB, which is lower than can be achieved with a good cassette recorder.
so record with 24 or 32 bits and amplify later before converting to 16
… the “ideal” amount of headroom to allow will depend on your confidence in how accurately you can estimate what the maximum peak level will be. In this sense “headroom” is essentially allowing room for error.
you need headroom that will be enough for the signals to come that you have not seen and do not know how much louder they will be.
there are recommended standards and most are in the -18 to -24 range for headroom.
all recordings have way too much dynamic range anyway. 30db is about the most you can use in normal environments. if you have an anechoic chamber and use headphones you might want more. most of us prefer our music so the highs dont wake the baby and the lows are not lost under the traffic noise outside. and that might be too much range for use in a car stereo.
The singer spends most of his time in severe peak distortion and that’s usually fatal. Clip-Fix and most tools like that are good for occasional tick here and there, not when the performance is a quarter distortion.
You might try the clip fix, though. One of the oddities of the tool is that it sometimes works better when you apply it multiple times.
And that’s precisely the reason Chris developed his famous compressor. To listen to opera in the car without turning the performance into garbage.
Listening room aside, it’s still necessary to capture live performances in the volume they’re being produced. I generally use -15dB as a design center and that’s only because I can reach over and smack the voice talent or singer if they go over. In uncontrolled conditions, I can understand wanting more.
In the case of a rock band, you might also want a microphone with the thermo-chemical or nuclear detonation option.
Television “normal” level is -20 in the US and -18 in Europe.
It is very impressive that you took the time to do this for me … thanks. I’m learning a lot in this thread. I almost like the Floorfish result, but it changes the character enough I’d be tempted to apply it to the entire vocal parts in order to get consistent timbre. I assume you meant to see Plot Spectrum in Aud … yes, you clobbered a lot above 7k. The “fish” tools look interesting. Did it take you a while to learn how to do them? Are you an “audio engineer”?
i fail to see why i have to record them at their SPL.
I can and will use normalise/amplify and compress to set the final level and loudness on the cd. as long as there is no noise on the low end you can record them waaaay below their level and increase the headroom even more to remove all risk of clipping.
being cautious , i leave plenty of headroom as the lesser evil.
-24dBFS seems to work for me with most sound sources as long as I can get some preliminary testing to see their typical level.
and I compress the last several db in RT as further insurance.
typically record 24 bits , process - normalise eq etc. - depending on the music, i may also compress – , dithertrunc to 16, and burn cds.
for rock bands i just back up a few hundred yards.
(we were 80 feet from phish with hearos and earmuffs and the sound was still too loud! )
The floorfish results are about the best you’re likely to get.
I’ve had a go using a different method, and restricting the effects to those included in Audacity.
The method here was to generate a bit of pink noise, and deliberately clip it severely. This could be done by amplifying it with the “allow clipping” box selected, then amplifying it negatively to drop the level down again - I actually used the Nyquist prompt to do the clipping, but its the same effect.
I then used that noise to create a noise profile in the Noise Remover effect, then used Noise Remover on the track.
Finally I used a bit of Eq (Equalizer effect) to drop down frequencies in the 4 kHz range.
The result is certainly no better than Trebor’s “floorfish” example, but still a reasonable salvage job. What flavour of bad sound do you want
As you said at the very beginning woodhands, “this type of distortion is tought to deal with”.
Hmmmm … why use pink noise for the noise sample? It seems so different than the type of sound heard from overmodulating as here.
I tried Noise Remover on a particularly staticky section, but no result.
I imagine one would want a simple signal (pure 440Hz tone?), amplify it so it distorts, then extract that pure tone signal - leaving the noise alone. Can that be done in A?
It started as pink noise, but then was severely truncated (clipped) so that it had a waveform like this:
It’s a valid point that you raise - I picked truncated pink noise because it has visual similarity to the “noise” that we’re trying to remove. Experimenting with other noise profile could produce better results, but I doubt that the improvement would be enough to justify the time.
To get any noticeable effect on your recording requires that the amplitude of the noise profile is very large - I used a noise sample with roughly the amplitude shown in the above screenshot.
how can it flattop before 1.0? bug in the linear display ??
dB view would be more useful IMHO - magnify the top part to really see what is happening wrt clipping.
