Resampling software isn't "bad", though it is an additional step of digital processing.

Going from 96 to 48 is quite simple. Apply a filter with a pass band to 20 kHz and a stop band at 24 kHz, on all the data, then throw away every other sample. Or, more accurately, do the calculation only for every other sample. Going from 88.2 to 48 or 96 to 44.1 is more complicated. You apply the appropriate filter at constantly shifting phases.

The thing is, you've recorded frequencies that you cannot reproduce at your delivery sample rate. Either you can get rid of them when you initially capture the audio with an analog filter, or you can sample higher frequencies and throw them out digitally. Either way, those frequencies get thrown out - even if you apply the standard processing of compression, EQ and reverb.

The exception, in my book, is when you will re-pitch your sounds. In that case, those higher harmonics can make it into the audible spectrum. So, if you're recording for sound design, definitely record at higher rates. You might play your broken glass sample straight in today's project, but you might pitch it down for a slow motion explosion in tomorrow's film.

I guess you could record everything at 96 kHz or 192 kHz and do your whole project at that rate, sampling down only in the final step. That would give you a high-frequency master. Personally, I wouldn't bother if you're using mid-fi gear. The Hollywood pros with their $1,500 mics and Sound Design mixers and recorders don't do this for film, so why bother when using an Rode NT2 into a Zoom H4n? At 192 kHz, you'll cut your real-time plug in count in four. Ouch!

I like to think about audio like this: there are primary issues and secondary issues. Primary issues are things you can hear and you can identify the source of the problem, like distortion, the wrong mic for the situation, or a high noise floor. Spend your money and time eliminating these primary issues. Secondary issues are the lack of polish on the apple. These issues might be improved with things like gold plated, oxygen free cables, super-stable word clocks, premium A/Ds, and 192 kHz sampling. Don't sweat these things until you've got the primary issues sorted.

If you own a studio and are marketing yourself as the best-of-the-best, then by all means develop a 192 kHz workflow with all the esoteric tricks. That attention to detail may or may not result in a better sound, but you can probably bill for it. In other words, it offers a tangible benefit - higher hourly rates.

Unless you have top equipment, top monitors, and a top room - and aren't re-pitching - I question the tangible benefit of higher sampling rates for audio with video.

But back to my earlier point. I'm talking theory. If motivated, do the tests with your own equipment and ears. Do blind tests. If the higher sampling rate offers a tangible benefit, go for it! :)

I agree .. but everyone should try 24/96 at least once.

Cheers.

But wouldn't 64/768 be so much better?????????????

Seriously, I think what I'm curious about is whether going from 96k to 48k actually throws away every other sample or does something more sophisticated like averaging the two samples or interpolating between them. Could be!

And I know enough about sampling to know that my math is too rusty to really understand it anymore - I've been in meetings with the guys who design the signal processing algorithms for disk read write channels and I understand enough to know what I don't understand.

If you were to simply throw away every other sample, you would get aliasing. It's similar to throwing away two of three lines of an image to create DSLR video. It works great on low frequency stuff, but gets ugly with high and out of band frequencies.

From 96 to 48, you low pass filter by passing everything below 20 kHz and cutting everything above 24 kHz. You can then throw away every other sample without penalty. (Or, better yet, only apply the filter calculations for as many samples as you need in the end result.)

Great thread!

My take is that a way to look at this is aquisition, post production and delivery.

In my opinion, 24/96 would most benefit the post production arena as the more information an EQ or effect has to work with the more impact it will have. So the way I see it is beef up the aquisition specs to feed the processors more information so they can create the most natural sound in a digital environment.

Then let the output ditherers cull the large amount of info down to the delivery specs.

But one can deliver 24/48 on DVD & BD so you are really only lowering the sample rate.

Do a test to see if you can hear a difference in post and in delivery.

All A>D converters use oversampling. The required low pass filtering is done digitally. It also seems the use of digital gain is quite common in cheap equipment.
High end audio recording is done using Direct Stream Digital (DSD), a bear of a thing to work with. Arguably only of benefit if delivery SACD which is surprisingly enough quite alive and kicking in Japan. I read the Sydney Opera House has become one of the world's premium recording venues for classical music since Vladimir Ashkenazy took over as conductor of the SSO.

As many have said none of this is very relevant to the H4n.

Yes, if you can hear a difference on your equipment. I can hear a difference on Edirol R-44s, even when editing the material on my laptop monitoring via the headphone output.

I have not done any A/B comparison to see if it "trickles down" to lower bit-rate exports of the audio. But since my time is spent working with the original files, I'm happier when it sounds better.