A Better Way To Find Dropouts?

I often work with DAT transcriptions that contain dropouts like the ones shown below. No tool in Wavelab seems to find them all. I’m wondering if anyone has a bright idea to share?

Many people suggest using Global Analysis > Errors (or the Error Correction tab which works the same way). This tool searches for spikes or other abrupt changes in the waveform, and it picks up things like this…

image

… but it won’t pick up dropouts like this one unless the sensitivity is so high that it flags way too many false positives.

image

Another option is to Analyze > Auto Split > Split at Silences, and then drop a marker at each point, but this has two problems. The first is, the minimum duration it catches is 100 ms, and it misses examples like this:

The other drawback is that the tool looks for a minimum RMS level, which works OK when the dropout is at -0 dB but that’s not always the case. The default setting of -45 db is too low to catch this (the scale is exaggerated):

Setting the RMS threshold higher results in too many false positives, especially when the minimum duration is short enough to pick up the 30ms dropouts.

What I think would work better would be a tool that looks for the absence of change in the waveform over a fixed number of samples. For example, 10 ms of “digital silence” of the sort I’m looking for is, at 44.1 kHz, 441 identical samples. In my world, 441 identical samples seems very, very wrong, because even the “silence” at the end of the tape tends to run -75 db or so and looks like this:

image

So… is there a way to search for these glitches based on the absence of variation in the waveform, rather than by the abruptness of variation (Analysis) or the average RMS quietness (Auto Split)? If not, could there be? Or is there a plugin that could do this? Thanks.

The Error Correction tab is a totally different analysis tool than Global Analysis, and much better for the concern of error detection.
Since the Error Correction tab has error sensitivity settings, I strongly recommend to experiment with this.

You could also benefit from the new Visual Analysis tool window and some of the analysis curves there, to visually pin points the gaps.

You could also use the Spectrogram: at gap edges, you generally have high frequency contents, and that is easily visible.

You could event trim the new Raibow waveform color to pin points high frequency contents.

Thanks, PG. I’m still using WL11.2 so I will have to upgrade and try the new spectral features.

I want to respond to your suggestion about the Error Correction tab. I was hopeful when it came out (WL 11, I believe?) and I have experimented with it. It does great things but it doesn’t do the job I want. For example, it flags this (false postive):

image

but misses all of these (orange markers are mine) …

… except two (here’s one it caught):

No combination of settings for “Crackle” “Clicks” and “Pops” picks up more than a couple of these errors and I still think it’s b/c of an abrupt change of amplitude. Even to get that many hits the tool picks up 1000+ non-errors. (This is with 2 hour audio files, and selecting a smaller area means doing it twice or three times or more.)

One great thing about the Error Correction tool is that I can set the sensitivity to 30, which is as high as it will go, and then Wavelab will report this:

image

I can then lower the sensitivity, step by step until that message goes away, and watch as the error markers disappear from the overview window. It’s easy to notice the visual changes, which confirm that the tool isn’t picking up the errors I’m looking for. Likewise, I can also lower the Minimum Audio Level and watch the errors drop out and return. That’s a terrific feature, and I’m mentioning it for anyone else reading this… but doesn’t solve my problem.

Also, the “Clipping” setting doesn’t work because, I assume, you can’t clip until you get to 0dB and the Minimum Audio Level settings are being ignored? Whatever the reason, I got not results.

Anyway, this is a plea to consider adding a feature to WL to detect repeated samples like these. I can’t be the only person restoring damaged audio files around here.

P.S. Digital errors like this, and similar things, stick out like a sore thumb to human eyes and ears. When AI comes to Wavelab some day, can I help train it? I’ve got lots of damaged audio to play back!

It would be nice to send me a couple of audio files where this type of audio error exists. I don’t say I will have a solution next week, but that would be a first step.

Happy to, thanks. Please let me know how? These files run 800 MB or so. I will chop them down a bit, but still much to large to e-mail.