Almost done with my documentary audio-scrub. I just hit the worst piece of audio in the whole film.

It's a person talking with a VERY LOUD bird-call ringing out, two 4-5 second bursts. Here's the file:

http://www.teachingdrum.org/adjul/elby2.wav

(1:37-1:41 and 1:47-1:52)

Don't worry about the sea of noise that the dialogue is swimming in; that's standard for these recordings and I can deal with that part.

I'm working in RX4 and tried copying and pasting "clean noise" over the bird call frequencies (there's one in the 1.5-3k and another grouping in the 7-15k but then it just sound like repetitive noise patterns which is just as distracting.

I plan to spend the morning 'painting' the spectrogram patterns and reducing the gain, and/or trying out the spectral repair in RX4.

any suggestions or attempts very welcome!

Only 20 minutes left out of 2 hours of scrubbing the audio!

The extreme bird call sounds pretty much like a pair of alternating frequencies. But there is also a significant amount of "glissando" where the pitch glides from higher to lower. So my first concept of simply filtering out the two frequencies didn't work. I cut out "chunks" of frequency/time and it improved a bit, but we don't know what is "acceptable" to you?

I have to wonder why the overall signal to noise ratio is so poor? And why recording continued with such a very loud external noise. Were you monitoring the audio while recording?

I didn't record it, the guy appeared to be an amateur. And he was recording a group of 30 people sitting in a very wide circle with one on-camera microphone, and no budget.

Alternating frequency works okay. The calls are by no means on just two frequencies, although their locations ACROSS two frequency ranges roughly match.

Oops, gotta reread your post i see now

Richard, the sample you posted is a start (still very distracting with the alternating blank audio freqs)....I brought it back in rx4 and tried 0-harmonic-level spectral repair, and pasting "regular noise" at a matching frequency over it. Still, the signal is so weak it almost doesn't even matter. Not sure what more I can do, maybe I'll just cut the audio way down at those segments and subtitle.

OK, yes the Noise to Signal ratio is unfortunate. So to start with I added 26x 1db Gain steps to gauge that I had enough signal to play with. Do turn your monitors way-down. I then became aware of HUM. I then applied 3x Hum Reduction steps. It was only then that I started with DeNoiser.

If you haven't prepped your WAV this way, then do try it.

Grazie

If the visual elements allow this noise to be in context, I'd leave it well alone. It sounds like the cameraman kept laughing - and making that less annoying would be my quest. All the treatment on the bird removes important detail from the voice. If the image is of people, sitting around a table in a hut with foliage and blue sky visible through open windows, I'd leave it alone. I on the other hand, it's a nondescript hotel room, then you're stuck I suspect. My attempts at repair and disguise made the voice sound quite different, and most listeners would notice the change.

+1 for Mr. Bernard and Mr. Johnson.

The most annoying thing to my ear is the cameraman's goofy guffaws. Reduce them manually. Then bring the gain up, so the NR has enough bits to work with. Notch out the hum, then run a broadband adaptive NR. I'd also suggest HF rolloff starting at 10kHz, so you are down to nothing by 14kHz ... this gets rid of the nasty HF spike around 15kHz (probably 15,750Hz from NTSC horizontal sync) and there's no useful audio up there anyway.

You can not remove the bird, it's right in the middle of the most important voice frequencies.

I agree with Paul.

OK, as a last treatment I applied a LOUDNESS EBU control over the whole to make an evened-out or at least a non-"starling" response by the listener.

I think if you put some ambient behind it you could be good to go.

I've tried to Upload sample, but it isn't getting here?

Grazie

Okay...so I had given up on that piece and either deleted the birdcall and repalced with ambient, or dulled down the birdcall with manual gain selections...and was going to subtitle it.

And I'm on the home stretch of my cut-by-cut audio scrub.

Then I tried this 26x +1 db gain, adaptive noise reduction, and ECU loudness control....and it's actually cleaner then my dialog NR scrub...cleaner and more evenly loud, of course.

So...I REALLY don't want to go through this whole thing again. I'm thinking to just render my master wav file UNCLEANED, then running the whole thing through 26x +1db/adaptive NR/EDU loudness...and then either before or after that jumping in and taking care of the big hum sections. Would that work? Will adaptive noise reduction adapt to the different noise profiles of different clips WITHIN a wav?

Actually, Bernard, your 26x +1db works even better with a dialog denoiser and then your EBU advice. Looks like I can just pretty much line up all of the audios in order, clean them, and then do the fine-tuning (clicks and hums).

NOW I need to consolidate all of my 4-5 versions of the audio into the originals and get them on a single track, yeck!

And Graham, why 26x of +1db gain increase? How did you come up with that number? And why not just do one big +26db jump?

Actually, Bernard, your 26x +1db works even better with a dialog denoiser and then your EBU advice. You're very welcome.

Grazie

And Graham, why 26x of +1db gain increase? How did you come up with that number? And why not just do one big +26db jump? Oh simple, really, I wanted to "hear" and "see" what was happening. Maybe x15 or x28 or whatever would have done it. No special math reason for x26 - I wish!

Glad you liked and could use my remedies.

Grazie

And yesterday I further experimented with pushing-back that tweetie-pie, using a selection of Brushes and -db gains. Are you interested?

Grazie

I'm definitely interested!

Really enjoying being able to "chunk" larger groups of audio and then go in for fine-tuning later. My earlier method looks pretty neurotic in retrospect. :)

I'm definitely interested! Yeah, well, . . . I'm not happy with tweetie-pie - has become a rusty bike chain . . . . I also removed some more White Noise

http://www.dvinfo.net/forum/attachment.php?attachmentid=35316&stc=1&d=1442409094

Of course, as I started cleaning-up I came across other artifacts.

I also added some SFx - iZo Ozone5 "Dramatic Width and Open Up Mid-Side" to a Brush-ed Selection - neat! Gonna use that again.

Grazie

I've just listened to the original audio file.

Honestly, just tell them that the audio is terrible and to go back and do a re-shoot. If they can't then it's their problem - otherwise we're effectively subsidising their cheap sense of reality.

Andrew

Andrew...lol, if i had my time machine, I would go back in time and ream out the camera man (it was filmed 13 years ago). But also, as someone that has done a 1-man-shoot-on-a-budget for near-impossible documentary-filming situations, I can empathize. His poor work inspired my low-budget multi-microphone vector approach to mic'ing large groups.

The good news is, I learned a lot. About pop and crackles, clicks, how to remove lots of coughs and sniffs (the camera man not only laughed a lot, but sniffed even more) and those birds...

Graham: I found that the birds are best treated with RX4s attenuation from .9 -> 2.0. The solo ones can be totally removed, where the long stretches you worked on can be nicely blended into the voice...better than your bike chain at least ! :P

I ended up using a simple technique:

x26 gain on every clip in the film
apply EBU loudness over the whole clip
apply decrackle over the whole clip
Low and high cuts, 90db and a 10-14k db fade-away

From there I went in and manually removed any further clicks, coughs, sniffs, and birds. With RX4s attenuation, partials-spectral-repair-tool, declicker and occasionally just de-gaining. For hums I don't know how to identify and nail exact frequencies as well with RX4 (it tends to want to take out multi-bands) so I used Audacity to seek out hums and removed those to a greater or lesser extent.

Here's my final cut of the horrendous clip in question:

http://www.teachingdrum.org/adjul/showntell.wav

thanks for all the help...about a week away from finishing this documentary I've been working on for two years :)

The solo ones can be totally removed, where the long stretches you worked on can be nicely blended into the voice... You're welcome.

As to the HUM remover, I've found that the multiple harmonics can and do work very well; I can also de-group the harmonics; I also use the Spectrum Analyzer to nail a particular Harmo:http://www.dvinfo.net/forum/attachment.php?attachmentid=35339&stc=1&d=1442818399

I like your results and still think that RX is just plain VooDoo! - Oh, that's IT Engineer Terminology.

Grazie

Izotope RX is powered by magical electronic smoke. :-D

Andrew

Graham - thanks for pointing out the analyzer and the harmonics...simplified my workflow

While you're at it...Graham, could you tell me if there's a key-command in RX4 that stops the playback at the last place the live-play ended at? For me, spacebar returns to the last place I clicked the cursor, and enter starts at the start of the sample.

While you're at it...Graham, could you tell me if there's a key-command in RX4 that stops the playback at the last place the live-play ended at? For me, spacebar returns to the last place I clicked the cursor, and enter starts at the start of the sample.

The RX solution/way is not intuitive. At the moment you/we have SPACEBAR for PLAY, that plays and then RETURNS to the same place. In RX we need to use a "TOGGLE" command to stop this happening, and that is Ctrl+R.

The SPACE Bar plays and returns, and CTRL+R stops this Return happening. Now, when we WANT to revert to the PLAY from start position to play again, we need to, again, press Ctrl+R. I think this is clunky and unnecessary User control.

Grazie