Salvaging recording done with AGC?

[moment42]

Hi. A while back, I recorded a choir concert using my NH-1, and today I pulled it onto my computer to split into songs and burn a CD for sharing. However, when listening to the recording I noticed that during loud portions (especially volume surges), the volume is decreased suddenly. My guess is that I mistakenly left AGC on (why is AGC the default?). My question is: is there a good way to "salvage" the recording, i.e. "undo" the AGC? Of course it's possible to find all the places where there are sudden volume changes and manually adjust it so that there are no noticeable jumps, but that's still just "winging it". Is there a better way? (I imagine that if one knew the exact algorithm used to adjust the volume, it would theoretically be possible to undo it, but I guess only Sony know that.)

Cheers,

/ M

[greenmachine]

If a dynamic passage has been compressed to constant loudness, how would you tell the loud passages apart from the quiet for a reconstruction? All you can do is to guess. The result would be mediocre at best.

[dex Otaku]

It is theoretically possible to take an unprocessed copy of the recording and apply expansion to it to compensate for the AGC. It won't work 100%, and to really do it properly you have to know the threshold, attack, release, and ratio that the AGC works at to attempt to apply its reverse. These are not listed in any of the manuals. The only way to really find out what the values are is to repeatedly run test signals through the mic input of the recorder to try and figure them out, something which I honestly wouldn't want to invest the time in doing myself.

You could fudge it by putting expansion [which is the opposite of compression] on the recording and just fiddling with the settings until you get something that sounds less-bad than your original.

Good luck..

[moment42]

It is theoretically possible to take an unprocessed copy of the recording and apply expansion to it to compensate for the AGC. It won't work 100%, and to really do it properly you have to know the threshold, attack, release, and ratio that the AGC works at to attempt to apply its reverse. These are not listed in any of the manuals. The only way to really find out what the values are is to repeatedly run test signals through the mic input of the recorder to try and figure them out, something which I honestly wouldn't want to invest the time in doing myself.

You could fudge it by putting expansion [which is the opposite of compression] on the recording and just fiddling with the settings until you get something that sounds less-bad than your original.

Good luck..

*sigh* Yes, that's about what I expected. Oh, well, you learn something every day. Today I learned that the default recording settings on my NH-1 are useless for the situations I mainly use it for (and yes, I usually turn AGC off, I just forgot). At least the recording doesn't sound completely horrible - it's listenable, just very annoying. And the recording was for documentation, not publishing.

Re compression: I don't really know what compression is, technically, but I thought AGC just equated to holding the unit while recording and fiddling with the recording level (except that you can't do that, but anyway). What's the difference between just applying an "amplify" filter to portions of the recording and using expansion?

Thanks

[dex Otaku]

AGC = compression/limiting.

Compression means: when levels go above a certain point [the threshold], they will be lowered by a certain amount [compression ratio].

In its simplest form, straight 2:1 compression say, above -12dBfs [that's the middle mark on your recorder's level meter], means that at 1dB above -12 it's compressed 2:1 to give, well, 1dB above .. at 2dB above, it's compressed 2:1 to give 1dB above .. at 10dB above, it's compressed 2:1 to give 5dB above .. and so on.

Limiting is infinity:1 compression, meaning that anything above the threshold is reduced to the threshold itself.

AGC is usually a combination of the two, i.e. compression above a certain threshold and then limiting just below the maximum level so that hopefully clipping distortion never occurs. Another possibility for AGC is dynamic compression, where the higher the levels go above the threshold, the higher the compression ratio gets, until it reaches limiting.

Expansion is the exact opposite: when levels go above the threshold, they are -increased- according to an expansion ratio. Technically, there is no difference in the process involved between expansion and compression except that instead of being 2:1, it's then 1:2 coming out.

I am not adept at explaining these things, though. If you're interested in learning, a much better explanation can be found at wikipedia.

[ozpeter]

If you used an audio editing program which provides volume envelopes (a line representing how much the volume should be raised or lowered along the length of the file) you could in theory restore at least some of the damage, but it would be a labour of love... though as you went along you'd start to get a feel for the kind of slopes involved.

[Killroy]

Hello,

if you happen to have available one of those good old cassette tape decks with Dolby C noise reduction, it would be worth a try to record your gig to a cassette - with Dolby C off and the highest possible recording level. Played back with Dolby C on then, this recording will experience a considerable dynamic range enhancement that might undo at least some of AGC's compression.

Using one of this cassette adapters that feed the signal straight from your MD recorder to the tape unit's playback head (I think they are still in use for playing back music from iPods on car stereo systems with casette tape player only), you could even bypass the analog recording to tape. Just playback from yor MD unit and record to your computer via the soundcard's line in from the tape deck's line out.

Cheers - Killroy.

[dex Otaku]

Using one of this cassette adapters that feed the signal straight from your MD recorder to the tape unit's playback head (I think they are still in use for playing back music from iPods on car stereo systems with casette tape player only), you could even bypass the analog recording to tape. Just playback from yor MD unit and record to your computer via the soundcard's line in from the tape deck's line out.

That's an interesting method suggestion, killroy. It has one unfortunate side-effect, though: Dolby B, C, and S all use bandpass compansion, C far more obviously than either of the others [above above 10kHz], so the end results would be a completely flattened high-end at the very least.

[Killroy]

Dolby B, C, and S all use bandpass compansion, C far more obviously than either of the others [above above 10kHz], so the end results would be a completely flattened high-end at the very least.

You are of course totally right on this, and thus my advice to operate at a recording level as high as possible to avoid this side effect as good as possible. Adding high-frequency gain with the MD-unit's equalizer at recording time may also help a little to compensate Dolby's treble cut. In some cases, I achieved very good results with this methodology when I digitally recorded some of my vinyl albums - with the side effect of efficient noise reduction. However, as already stated, it's worth a try!

Btw: glad to see you back here in MDCF in your "old mood"!

Cheers - Killroy.

[dex Otaku]

Heh, "old mood"?

At high levels into the decoder stage, sure, C-type NR shouldn't have too much artifacting in the sense of de-emphasis, but then - at the highest levels, it also does little to no expansion, either. One of the premises of most noise reduction schema [dbx compansion being an obvious exception] is that (post-emphasis playback-processing) high-level signals approach an expansion ratio of 1:1 at their upper threshold. The end result of putting AGC'd material through this kind of playback processing should in theory be that in quiet parts the high end is quite audibly lost [up to 20dB with Dolby C], and in really loud parts, no expansion takes place in the range where you actually want it to [the peaks].

Mind you, that's theory based on my understanding of things [mostly based on things like this: the Dolby labs white paper on B, C, and S-type NR]. Both my theory and my understanding are potentially flawed.

The method itself is kind of cool, actually, and we all know that practise differs greatly from theory. I'd actually like to hear a before and after of the real-world results. [No, I don't currently have a cassette deck with better than B on it.]