The only way to get rid of unwanted ambient sound to get the wanted sound recorded at a higher level, i.e. get the microphone closer to the wanted sound.
Certainly directional mics will help and if the unwanted sound is not broad spectrum some work in post can help get rid of it however getting the original recording as clean as possible should always be the first line of attack.

In this case headset style mics would be the go. There's plenty of these on the market and you pretty much get what you pay for. The best are almost invisible to the camera.

Bob.

Use really close miking on the talent so their voices are as high a level as possible above the noise. Directional mikes pointed away from the noise source may help though if it's a small room it might not help much.

Then use sony's noise reduction plugin or even better Izotope RX.

Get plenty of recordings of just the freezer noise. Have long gaps between question and answer in the interviews so you can sample the freezer noise from the gaps in between. You can edit the gaps out after you've processed the audio.

It is pretty impressive what you can do with the noise reduction tools if you can get good samples of just the noise. After processing you may be able to get a little more with EQ by rolling off high frequency noise and if there are any big thumps going on a high pass filter on the low end can reduce those. Whines can be hit with narrow band reject filters. RX has some really powerful ones that can be extremely narrow.

Close miking to begin with is the key - better to reduce the noise on the original recording first.

rocky

I'll second Bob's suggestion of a headset / earset mic.

I bought one of these (the "poor man's Countryman") and it's surprisingly clean for the price, slightly warmer than the e6i. Has worked well for us live, and a local church bought two on my rec. Only drawback is it won't handle extremely high SPLs (belters) like the 135dB e6i. Biggest advantage is it is extremely rugged and bendable, doesn't break like the e6 models.

http://www.microphonemadness.com/products/mm-psm-directional.htm

Jack, this is one time that you may want to seriously consider hiring a professional sound person.
He/she will have the gear that you'll never be able to afford, has probably run into similar situations before and will know the best way to get the cleanest possible sound for you.
There's a user on here by the name of "tomaras" (Charles Tomaras) who lives in Seattle and is a long-time pro sound guy.
Drop him a line through here or give him a call (phone number is on his site).

Mike

In principle it is possible to employ noise cancelling techniques for such a problem, but I don't know off-hand any suitable commercial software.

What you would do is place a microphone close to the noise source so that it picks up noise only (no signal!) and another mic close to the wanted signal, which picks up signal plus noise. The mics and noise source should be stationary. During a noise-only time, the relationship (transfer function) between the noise picked up by each mic is determined, and then later when the signal is present, the noise picked by the signal mic can be computed and subtracted. 20 dB noise reduction should be possible without generating artefacts you get with single channel processing.

Sorry if this not helpful, but maybe someone knows someone with suitable equipment.

All the above will help. A highly directional mike to get the sound level as high as possible over the noise. Then use Diamond Cut 7 or 8 to eliminate the noise.

@PeterDuke,
Your theory is only plausible in an acoustically neutral space, with no walls or reflective surfaces, no movement of subjects or their microphones, no introduction of the vocal sound into the space, and no phase dynamics at the source. In other words, impossible.

You see, the "transfer function" (we call it an "impulse") to which you refer is not a static constant, but at the recording point is a dynamic, swirling sandstorm of direct reflections, enharmonics, phase differentials, combination tones, and flanging; and at today's technology levels, is neither predictable nor quantifiable.

The scheme you suggest could easily turn the sandstorm into a maelstrom given the alternating phase suppression and reinforcement cycles of mid-high frequencies over short time periods.

As someone who has designed and recorded sound from synchronized multiple points in acoustically large spaces, the "acoustic swirl" is well known to me, although in well-designed spaces, it sometimes enhances the spatial effects in 5.1 recordings. At worst case, it creates pulsating or flanged feedback (not pleasant).

The only adaptation of that technique that is sometimes used in industrial settings is realtime negative feedback of <40Hz rumble frequencies using large subwoofers, where the wavelengths are long enough to preclude any possibility of phase reinforcement.

For that matter, having the mics in different positions will make the process implausible too. The mic closer to the noise source is going to record different material than the mic farther away so they won't cancel each other.

I give you the noise cancelling microphone. Unfortunately these are primarily targetted at speech intelligibility and can be quite large and ugly, not a practical solution in the current context but it does show that it can be done.

On a more practical note it may be possible to reduce the problem by building a mini studio in situ. Hospitals have a good supply of mattresses and blankets. Prop mattresses up against the walls and hang the blankets over them.

Bob.

Musicvid

If you haven't tried it, don't knock it. I have! (I used to work in a telecommunications research laboratory.)

I have said that the noise source and microphones should be stationary. The speaker could move provided that he/she was reasonably close to the signal+noise mic. Also, very little (preferably none) of the speaker's voice should be picked up by the noise-only mic.

The mics and noise should be stationary so that the transfer function (frequency domain term) or impulse response (time domain term) is fixed. Otherwise the filter modelling the impulse response needs to be adaptive, which is also feasable (and I have implemented), where the filter adapts during non-speech intervals and is frozen during speaking. This is more practicable in a telephone situation where people speak in turn but less feasible in the proposed interview scenario above.

I came upon this product during a web search, but it is not clear to me whether it is processing along the lines I have discussed.

http://www.grittec.com/multi-microphone.html

I should add that the method works best in a low reberberant environment. The filter is implemented as a Finite Impulse Response (FIR) type and its length has to be longer than the reverberation time of the room.

i would run a 'test' before committing to buying / hiring anything.

try a few takes and then run the results past the client for 'approval'.

if he reckons that there isn't a problem, then you're good to go - and any 'clean-up' you carry out afterwards will certainly earn you credit from the client.

if he realises the problem and that the results are too 'muddy'. maybe he'll have second thoughts.

There have been lengthy discussions of using impulse technology for acoustic cancellation (as opposed to modeling) going several back years on this forum. Even knowledgeable pros have tricked themselves into thinking our very own Acoustic Mirror could be used for this purpose. There are simply too many realtime variables to make this feasible at this point in history. It has never been demonstrated to my satisfaction that an "adaptive impulse" application is anywhere near on the horizon. One guy even posted a demonstration. Turns out he cheated by using the output as the impulse (like in the 1940's) . . .

One promising emerging technology is described here:
http://www.izotope.com/products/audio/anrb/

What you imagine is conceptually possible, given you could nail down enough variables to keep from making the noise worse - I just haven't seen it in application at any practical level. It is simply not predictable at any point in time whether a given complex reflection (or any part of it) is going to come back at you in phase or out of phase; that relationship is dynamic, and doesn't fit a discernible matrix. Not going to belabor this any further, yet I eagerly await your solution.

In fairness to Peter I have used some quite impressive noise cancelling microphones for telephony / video conferencing. These things are rather expensive, take a few seconds to tune themselves etc.
The problem is they're fine for telephony however what most of us here would call "quality speech" and telephony are two quite different things.

Bob.

Noise canceling mics are proven technology and I agree with Bob they are a viable option for communications and sports announcers. They use an open-back technology that applies noise as direct negative feedback, not unlike pressure zone mics. Our crews couldn't communicate on clearcom during live shows without them.

And two-mic setups could be effective as a noise-reduction array. But the second mic would have to be in the same ambient space as the talent, not at the source of the noise, which is of absolutely no interest in this situation. It's use would be to record a feedback track in realtime, not an unreferenced impulse. And positioning is absolutely critical, because it must pick up more ambiance and very little talent, meaning it is also likely to pick up camera and crew noise which defeats the whole purpose. I can't imagine an effective video location design that would accommodate this visually and acoustically, but maybe someone has had success with it (location video isn't my strong suit anyway).

JackW,

This whole discussion reminded me of another technique, adapted to digital over a decade ago, that may reduce the effect of noise in post; realistically, maybe -3dB.

See my post on echo reduction here and it may be of some help.
http://www.sonycreativesoftware.com/forums/ShowMessage.asp?ForumID=3&MessageID=545526

However, despite all the useful and not-so-useful advice offered here, I like ushere's the best!

Noise cancelling microphones, at least the ones I am familiar with, use a different principal to what I was talking about. They rely on the fact that being close to the lips, the speech is unidirectional (high pressure gradient), but being far from the noise sources, the noise is relatively omnidirectional (low pressure gradient). The diaphram is open on both sides, so noise presses more or less equally on both sides at once, getting nowhere, but speech only presses on one side.

I don't know how noise cancelling headphones work but I have assumed that they work something like what I was talking about (monitor the noise outside the headphone and inject a filtered version of opposite phase to cancel the noise inside the headphone). If anyone knows otherwise then please enlighten me.

PS
If the noise cancelling mics that Bob talks about need to adapt then they may be different from what I said above in this post and probably more like what I was originally talking about, although they may still depend on the different pressure gradients I mentioned.

Former $dayjob produces a cheap line of incredible noise cancelling mics for telephone applications. They are completely passive, requiring no tuning or setup at all. The usual test site was the loading dock with three industrial air compressors running. Ambient noise levels could reach 130dB, yet someone whispering into the phone near them could be understood clearly.

The frequency range was a mere 250 to 3KHz, which is impressive for telephone applications and certainly usable for speech applications. The things were pretty cheap too, in the $12 to $15 range. Of course, they were all built into drop-in housings that fit regular telephone handsets so they wouldn't be that easy to use in a video interview situation, but the same technology could easily be fit into more socially acceptable microphone shapes.

I have a similar passive noise cancelling mike on the headset I use for Skype. Unit was made by Plantronics.

The other unit that I was talking about, the active one, was made by Polycom. That was a conference phone / mic. I think at the time we paid over $2,000 for it. Put it in the middle of a big table and it would pickup speech with perfect clarity, full duplex, no room tone. Somewhat better than PZMs I'd say.

As you say, all these devices are great for comms, heck I could throw thoat and bone conduction mics into the suggestion box as well :)

Bob.

Oh yes, the Polycom conference room mics. We have one of those in the office and it is impressive. Those aren't so much noise reduction as echo & reverberation reduction. Very sophisticated, very good at what they do. I'm not sure how well they'd perform in a live, noisy environment. Of course, that's not what they're intended for. I have had them get confused and "lose the thread" of the speech going on if there's a sudden ambient noise.

Just a thought---maybe the customer wants to hear the noise?
JJK

While poking around for noise-canceling mics for broadcast, I came across this. You can invert the phase of one capsule to turn it into a true bidirectional noise-canceling array. Handles major SPLs, too. (Just noticed it's an omni, so don't know how well it would work for isolation).

http://www.bhphotovideo.com/c/product/425114-REG/Sanken_COS_22_COS_22_Double_Capsule_Lavalier.html

Just exactly how loud are these freezers? If they aren't outrageously loud and since it is a steady-state noise, Izotope RX or Sound Forge Noise reduction could go a long way toward reducing or eliminating the freezer noise.

Despite what anyone says, taking two microphones at different positions and trying to invert the phase and cancel the noise simply will not work except at very low frequencies and even then it will require careful level and phase matching in order to achieve any effective cancellation.