I can work on your project.

Find me! Call DAP at 214.350.7678 or email rene@dallasaudiopost.com. Also check out echocollectivefx.com for custom sfx, and tonebenders.net for my podcast.

Tuesday, December 28

Distant train extraction

So while I was out recording footsteps on rubber tracks and revolving metal turnstile doors this past weekend I managed to catch a pretty solid distant train as I was walking from one place to the next. I had heard the train a few minutes before, so I kicked on the prerecord buffer in my D50 and started walking towards the track that was a few hundred yards away.

Sure enough that train got closer and started blowing its horn. Here's the raw audio I caught - notice that you can hear me stop walking, turn and click the record button since the prerecord buffer was on. Also note the amount of wind I had going on there.

Distant train by Rcoronado

when i got back to the studio I was pretty disappointed with the amount of noise in the file so after noodling a bit I chalked it up to a swing and a miss. Regular Izotope eq and noise reduction was just leaving too many artifacts. Right before I deleted the file I decided to try one more thing on a whim: sample the train and use the tonal noise reduction algorithm to listen to the noise only.

At this point I wasn't really trying to clean up the file anymore, I was just playing around with the tools. What I got was pretty remarkable in terms of clean audio though, so I'd like to detail what I ended up doing:

The beautiful thing about the way that Izotope RX works is that it is very flexible and able to be tailored to any specific type of processing. Most other desnoising programs don't have separate controls for tonal noise vs broadband noise, and this feature really sets RX apart. I highlighted a clean section of the train whistle and used that as my noise print, then I clicked the "output noise only" box at the top right. At first I experimented a bit with bending the noise reduction curve to accentuate the train tonality and ignore the rest, but the result was a lot less natural than just the flat line there, so I left it at that.

I kicked the tonal reduction up to the max and the broadband reduction down to zero, but again that felt a little artificial so I dialed a touch of the broadband noise back in to smooth the edges out.

I actually went through this process twice, which yielded very good results.

The denoiser didn't do much for my wind and general city traffic tone though, so I ran a linear phase highpass just under the tonic note of the train whistle, and I ran it pretty steeply. That lopped off the bottom end and left me very clean.

What was left was peaking out around -30 or so, and I added about 12db of gain back into the file because I can. This is why we always record at 24 bits. :)

The end result looked and sounded like this:

Distant train-extracted by Rcoronado

I went from having something completely unusable from an impromptu recording to having something that I could layer in with some traffic (to taste this time) and have it sell. The artifacts aren't the ugly glassy things you normally get with noise reduction software either. They sound much more like an expander working, which is something you can get away with much more easily.
also, all of the reverb pretty much stayed intact, which would never happen if I hammered that with broadband reduction.

This kind of thing is pretty specialized though, as it probably won't ever work on things that aren't both harmonic and static. For things such as horns, alarms and sirens heard in noisy cities though, its pretty darn interesting.

One strange side effect came at the end, when my voice kind of vocoded against the reverberating distant tones of the train. I really didn't expect that, but I'm curious now about how to exploit it. I guess recording someone's voice while playing a static synth tone in the same room would do it, but it may also cost me friends. :)

1 comment:

Леша said...

In RX2, there's a DeConstruct module that can do a similar thing: separate your tonal and noisy signals and re-mix them.