repair dropouts

WinXP SP3 Audacity 2.1.2

I’ve got an MP3 recording that has numerous dropouts, as seen below.

Dead silence for about 0.021 secs.

This is too long to use the “repair” effect.
Is there any way to either smooth them over or just delete them?

There are longer stretches of intentional silence. But I think just targeting silences of this length or less might work.

If I delete the silence and then select the join and do “repair”, I get a tolerable result.
But there are too many to do them all manually.
dropout.png

You could use the “Truncate Silence” effect (http://manual.audacityteam.org/o/man/truncate_silence.html), but (a) you would need to take care to avoid truncating silences that are supposed to be present, and (b) you may still need to “repair” some of the joins manually.

And you’ll need to manually inspect the whole thing.

Do you know why that happened? This is dangerously close to being an unrecoverable show.

Koz

That won’t work. Even if I select the region containing a dropout, it’s too short to trigger it. Anyway, this is a spoken word track and the space between words is silent (dead silent, presumably by design) so there are thousands of “correct” silences.

There are several silence detecting tools, but like this one, all I’ve looked at select silences longer than a threshold, not shorter.

No idea, I didn’t make it. I pointed out the problem to the uploader, but with no response so far.

Anyway, it seems that either I delete it or use a mental filter and try not to notice. Manually editing it would take 10 times longer.

You said that the length of silences are “about 0.021 secs”.
In Audacity 2.1.2 the “Detected Silence > Duration” setting can go down to 0.001 seconds. http://manual.audacityteam.org/o/man/truncate_silence.html#detection

In spoken word recordings the gaps are rarely completely silent, but OK if they are in this recording, then that puts an end to any hope of automating repair.

Sorry, got that mixed up somehow. But though “Truncate” can detect them, when set to “truncate to 0.0 seconds” it only reduces a 20 ms dropout to a 6 ms dropout. Still noticeable.
And since I’d have to manually select the silences to act on, even if it worked, it’s not useful in this case.


I specified “dead silence” because I know in speech normally there is a background. Maybe a noise gate was used. I do that sometimes, but the filter I use doesn’t create glitches like this.

In the current 2.1.2 version of Audacity, the detected “silence” (below the detection threshold) may be reduced to zero.
However, this is academic if the silence between words is “absolute silence” because there is no way that Audacity can know what is “intentional” silence and what is a flaw.

There is simple criterion: silence >5ms and < 25ms. That should not catch any of the inter-word pauses. But none of the silence detectors I’ve found have such a setting.

But none of the silence detectors I’ve found have such a setting.

Too specific.

If you were to develop such a tool, Audacity stands ready to consider your submission.

You didn’t hear back from the content developers? Maybe they know and they’re just getting tired of hearing about it.

ACX/AudioBook has two levels of testing to guard against problems like this. There is the standards robot which blindly tests for noise, loudness and peak distortion and then there’s Human Quality Control. This would never have made it past HQC.

Koz

Of course I don’t mean those numbers should be hard wired. But, as I mentioned, the existing effects all let you select for silence longer than a given threshold, but not shorter, or in a specific range.

If you were to develop such a tool, Audacity stands ready to consider your submission.

I’m not going to spend a few months learning how to code Audacity plugins. And I don’t expect anyone to do it for me.
It does seem like it would be a simple extension to existing effects though.

It does seem like it would be a simple extension to existing effects though.

I doubt it’s an extension. I’m guessing you’re talking about crumpling up and throwing away about a third of the plugin.

It might be handy, but you’re the first poster I know to ask for it. Also, there’s the requirement to change the user interface and user instructions. I’ve done instructions. They’re not fun.

We’re considering changes one of the Generate tools because it was so universally comprehensive and broadly versatile nobody could figure out how to use it.

Koz

Of course, Silence Finder and Sound Finder should have this ability. There is an experimental Sound Finder in circulation that does what you want, but not as far as I know a Silence Finder.

Gale