Wow! It’s been a while since I last updated the ol’ Stereo Bus. But I haven’t forgotten about it. I’m settling into San Francisco nicely, keeping myself very busy. Unfortunately, space is too expensive for me to open up the studio that I planned to open here, but the good news is I’ve started playing bass guitar for the band Cuban Cigar Crisis. CCC is gearing up for recording an album Summer 2010 and I’m already doing research into the production process. I have decided that we should go with a high sample-rate recording this time around. I’ve been working pretty much exclusively at 24-bit/44.1khz but now DAWs are fast enough and hard drives are cheap enough that there’s no good reason not to step it up. 192khz is simply overkill. Most DACs simply upsample to achieve that rate and response can actually be poorer as a result. This leaves the choice between 96khz and 88.2khz.
What’s Your Target?
It all comes down to your target. What is the destination sample rate? Obviously if you plan on releasing high-definition DVD audio, you’ll want your material to be tracked at 96khz. Most of us are targeting standard CD-Audio and simply want to squeeze more definition out of the sound.
Analog vs. Digital Mastering
So, if you ultimately plan on landing at 44.1khz, the question becomes how you’ll end up there. Mastering engineers who use entirely analog processing will pipe your mixes out into the analog domain, throw some compressors and EQ into the fray, and re-convert the result using their high-end DACs. This isn’t a sample-rate conversion. It’s a re-digitization of the sound. Therefore there isn’t a strict correlation between the origin sample rate and the destination sample rate besides how well the fidelity of the source material translates into the digital domain. This is probably a factor of what sample rate the AD converters the mastering engineer is using ‘prefers’. So don’t worry about it. In this situation, 96khz may be the best choice.
If the engineer is going to be mastering in the digital domain, you have to factor in sample rate conversion.
Sample Rate Conversion
Digital mastering will require a reduction in sample rate from 96khz or 88.2khz to 44.1khz. The translation from 88.2khz to 44.1khz is exactly 1/2. It’s a nice, clean mathematical conversion. To go from 96khz to 44.1khz, on the other hand, we have to divide by some nasty number. 88.2khz, in this case, will almost certainly sound ‘sharper’ and cleaner.
A good analogy is digital photography. Try converting a high-res image to 960px and 882px, respectively. Then save both down to 441px. No matter what dithering algorithm you choose, you’ll probably be as surprised as I was at the results. This is actually a pretty apt analogy to audio.
Check out these images:
Lossless PNG converted from 88.2 to 44.1
Lossless PNG converted from 96 to 44.1
Notice how the 96 > 44.1 image is substantially blurrier, particularly in the high-contrast areas around the white spots of my kitty’s fur. The 88.2 > 44.1 image is significantly sharper. This conversion was made using standard bicubic dithering
So, basically, I’m sold. The sonic differences between 96khz and 88.2khz are minimal, yet the sample rate conversion to 44.1 is much smoother with 88.2khz.
What do you think?
![[del.icio.us]](http://images.del.icio.us/static/img/delicious.small.gif){kind=small}
6 Comments
Very nice. The math nerd in me wants to point out that the 88.2 -> 44.1 conversion comes out cleaner because of the direct halving. 96 -> 44.1 is a 2:1 ratio whereas the 96 -> 44.1 is approximately 2.177 to 1 ratio. It makes sense you’d lose clarity in an uneven division like that. Even the cheapest of components should be able to accurately divide by 2.
Digitally dividing by a decimal extending out to 30 places probably isn’t as easy to engineer as it sounds. No pun intended.
That’s actually exactly my point! Even 1/2 ratio comes out cleanest :)
Very nice. The math nerd in me wants to point out that the 88.2 -> 44.1 conversion comes out cleaner because of the direct halving. 96 -> 44.1 is a 2:1 ratio whereas the 96 -> 44.1 is approximately 2.177 to 1 ratio. It makes sense you’d lose clarity in an uneven division like that. Even the cheapest of components should be able to accurately divide by 2.
+1
Absolutely right. So unless you have super high-end filters, stick with even sample rate conversions!
Is the conversion from 88.2khz really smoother to your ears than 96khz? I’ve done tests, using iZotope sample conversion, which is the best I have ever used, and I can’t tell if it’s any smoother or not. Theoretically, possibly, but audibly….?
Yeah, MBIT+ is a great dithering algorithm. I doubt the sound would be smoother perse. I associate smoothness more with bit depth than sample rate. If anything I think the dithered conversion would be smoother, as the photo is smoother, but the undithered conversion would be clearer. But I suspect the ultra-high frequencies might be less smeared going from 88.2 to 44.1.
I don’t have any gear that produces tones above 22khz, so I doubt I’d hear much. The high sample rates are definitely worth it to reduce any resonances or harmonics introduced by a low-pass filter, but you’ll end up with artifacts if you have to dither the result. The point is that for 88.2 to 44.1 you just throw away half the samples and you get exactly what would have been recorded at 44.1, if I understand it correctly.