Best sound quality - run 96khz samplerate, do not oversample.

[aciddose]

decimation requires the same computation regardless of the fraction involved.

it's only in specific situations such as half-band decimation that optimizations can be used.

[mystran]

decimation requires the same computation regardless of the fraction involved.

it's only in specific situations such as half-band decimation that optimizations can be used.

Well, you can certainly use the poly-phase approach with any rational ratio, but I don't see how it's relevant anyway. If a plugin runs (say) 10% CPU at 44.1kHz then as long as the resampling (back and forth) uses less than another 10% you'll typically end up saving CPU when host is running 88.2kHz or higher.

I'm not suggesting we start resampling for 48kHz to 44.1kHz and back anyway. I'm suggesting it might be sensible to downsample to a "sensible" range when the host rate is much higher than what is useful (just like we upsample when the host rate isn't high enough).

[AudioGuy720]

Let me add my two cents, based on the readings of people who are smarter than me.

From Bob Katz' mastering audio book I gather that *as a delivery format* and using higher quality converters during downsampling there is no audible difference between high sample rates (88.2kHz+) and lower ones (44.1kHz/48kHz).

The most audible problems are with passband ripple (stuff you can hear), alias distortion (harmonics you can hear) and artifacts created by too steep of a transition band (which causes echoing). So as long as your downsampler doesn't suck 44.1 kHz is just as good as 96 kHz.

However as a recordable format it seems higher sampling rates do offer benefits. Not because we can hear supersonics but because digital errors (which cause audible distortion at lower sample rates) are spread into portions of the spectrum where we cannot hear it. Heck, even upsampling 16-bit/44.1 kHz or 48 kHz recordings (from DAT tapes, for example) is beneficial.

I'd venture to say most new plugins (2006 and above) do upconversion. In other words if your recorded material is at 44.1 kHz it will upsample 4x to 176.4 kHz. Then as the last step an anti-alias filter (pretty much a high quality high cut filter) is applied before downsampling back to 44.1 kHz (or whatever) and any aliasing/distortion isn't heard. But since you may not know how well a plugin oversamples it's probably best to start with a higher sample rate.

According to Dan Lavry 192 kHz sampling (and above) is not only overkill but is detrimental to recording. Accuracy decreases and distortion increases. And running hardware that is 192 kHz natively at a lower rate doesn't help either. 96 kHz seems to be the sweet spot. And from my experience using an old 16-bit/48 kHz system that destroys a lot of newer prosumer gear that is 24-bit/96 kHz, converters and the circuitry around them are more important to quality than marketing numbers.

Higher sampling also allows you to slow down audio with much higher quality. That is, if your rate is say 96 kHz and playback is 44.1 kHz. Try it!

According to Jim Johnston, who invented the science behind perceptual coding, 50 kHz is the ideal sample rate--damn close to the now-standard 48 kHz.

As far as Dan Lavry is concerned though, 60 kHz is ideal.

My personal suggestion is record at 24-bit 48 kHz if your computer isn't the greatest. But if you have a bad ass machine 24-bit 96 kHz. Why not, you paid for all that power/speed?

[bezusheist]

not to get OT, but could some one recommend a (very) good SRC converter for downsampling because i find the one in Logic to suck ass pretty bad...

[aciddose]

I'm suggesting it might be sensible to downsample to a "sensible" range when the host rate is much higher than what is useful (just like we upsample when the host rate isn't high enough).

sure, if you don't actually want to produce what the user asks for that's fine. there are some plugins out there already which i know for a fact do this.

in most cases though you're not going to find that operating at the native rate will cost more than decimating the input and interpolating the output. just as you won't find many cases where it's less expensive to interpolate the input and decimate the output at a lower native rate.

yes, plugins should take note of the rate they operate at. there were plugins for years that didn't care and would use per-computed coefficients for 44.1k. those can't be used at any other rate. when over-sampling it makes sense to allow the user to adjust the factor, or aim for a specific minimum rate.

i don't think it makes sense to set a maximum rate though except in cases where the processing is extremely expensive, more so than the re-sampling.

if we're talking about generators that have many voices then there is no decimation taking place (assuming no input) and interpolation will certainly be faster than running at the native rate. that's potentially a place where it could make sense. still, you'd be over-riding what the system/user says. if it were made an option, that might be ok. otherwise the plugin would only be made incapable / lame.

[AudioGuy720]

not to get OT, but could some one recommend a (very) good SRC converter for downsampling because i find the one in Logic to suck ass pretty bad...

iZotope SRC (available in their RX Advanced or since you're on a Mac http://www.audiofile-engineering.com/waveeditor/ . I'm PC-only and wish this program were on Windows because I'd love a cheaper way to get their SRC like Wave Editor offers.

Another good one (Windows only) is http://www.voxengo.com/product/r8brain/ and the Pro version as well.

To compare different converters check out http://src.infinitewave.ca/ . Be aware that iZotope has some type of involvement with the site but most people agree iZotope's is top notch. For now I'll just use the one built into my DAW, LOL! Don't let the charts fool you though because when I demoed RX Advanced the minimum phase setting to be the most pleasing to the ear.

[bezusheist]

not to get OT, but could some one recommend a (very) good SRC converter for downsampling because i find the one in Logic to suck ass pretty bad...

iZotope SRC (available in their RX Advanced or since you're on a Mac http://www.audiofile-engineering.com/waveeditor/ . I'm PC-only and wish this program were on Windows because I'd love

Another good one (Windows only) is http://www.voxengo.com/product/r8brain/ and the Pro version as well.

To compare different converters check out http://src.infinitewave.ca/ . Be aware that iZotope has some type of involvement with the site but most people agree iZotope's is top notch. For now I'll just use the one built into my DAW, LOL!

yeh, ,,,i was cheking the SRC comparison site earlier but i dont really understand it...
i will try the iZotope out,,,
i tried the AE wave editor before for other reasons, now i need to check back with that for the SRC...
thanks...

edit: oh f@k iZotope over $1000? seriously? Wave Editor it is then...

[AudioGuy720]

not to get OT, but could some one recommend a (very) good SRC converter for downsampling because i find the one in Logic to suck ass pretty bad...

iZotope SRC (available in their RX Advanced or since you're on a Mac http://www.audiofile-engineering.com/waveeditor/ . I'm PC-only and wish this program were on Windows because I'd love

Another good one (Windows only) is http://www.voxengo.com/product/r8brain/ and the Pro version as well.

To compare different converters check out http://src.infinitewave.ca/ . Be aware that iZotope has some type of involvement with the site but most people agree iZotope's is top notch. For now I'll just use the one built into my DAW, LOL!

yeh, ,,,i was cheking the SRC comparison site earlier but i dont really understand it...
i will try the iZotope out,,,
i tried the AE wave editor before for other reasons, now i need to check back with that for the SRC...
thanks...

edit: oh f@k iZotope over $1000? seriously? Wave Editor it is then...

Haha, yep I meant to warn you about RX Advanced costing $1,000+. I forgot to finish my sentence (ends with "I'd love") but it's now edited.

[stonestreet]

A bit OT but something that I am curious about. Twenty odd years ago I bought a Denon CD player (I still have it but it does not work anymore) It says on the front '4x oversampling' This is long before I started messing about recording sounds digitally into a computer. I always thought that the oversampling was like reading the information four times prior to spitting it out to the amp and correcting any errors in the process.

The CD player in my car is part of the car linked in to car visual display. I am not sure if I can change it to a choice of my own. It is crap and jumps the playing of the cd when going over the slightest of bumps. So I would like the idea of oversampling as in my thought.

Now can anybody satisfy my curiosity about the oversampling in this context. Is it the same oversampling that you guys are discussing? Cheers

[aciddose]

A bit OT but something that I am curious about. Twenty odd years ago I bought a Denon CD player (I still have it but it does not work anymore) It says on the front '4x oversampling' This is long before I started messing about recording sounds digitally into a computer. I always thought that the oversampling was like reading the information four times prior to spitting it out to the amp and correcting any errors in the process.

The CD player in my car is part of the car linked in to car visual display. I am not sure if I can change it to a choice of my own. It is crap and jumps the playing of the cd when going over the slightest of bumps. So I would like the idea of oversampling as in my thought.

Now can anybody satisfy my curiosity about the oversampling in this context. Is it the same oversampling that you guys are discussing? Cheers

yes. oversampling is when you sample more often. this doesn't require you to sample from the original source. any signal made up from samples can be re-sampled to another rate using interpolation.

when you convert samples which are actually impulses with an infinitely thin duration to an analog signal, you need to settle on some way to represent those as it's impossible to have infinitely thin impulses. (which would be composed of an infinite amount of energy in an infinitely thin slice of time, which obviously is nonsense except in cases like the big-bang, assuming we're even nearly right with that sort of thinking...)

what you can do is replace the impulses with flat levels drawn between them. you can draw a ramp between the two points which is basically a sort of filter applied to the flat "steps". you can apply other sorts of filters afterward.

there is a great difficulty in producing accurate filters in analog electronics. the solution is to ease the requirements of the analog filter by doing most of the work digitally.

in order to accomplish this, you interpolate the data from the samples and apply a digital filter.

this is what is generally referred to by the term "oversampling".

edit: actually, i didn't bother to mention that when reconstructing the analog signal the impulses aren't really infinitely thin. they're sinc impulses at the sampling frequency which are spread out over an infinite period of time in both the past and future. obviously i left that out because the "infinitely thin" explanation would make your eyes glaze over just as much as any other, so whatever. they really are "infinitely thin" in some ways, but the reconstruction requires that you filter those to remove any content above half the sampling rate - that sort of filter happens to be a sinc at the sampling frequency.

[kmonkey]

not to get OT, but could some one recommend a (very) good SRC converter for downsampling because i find the one in Logic to suck ass pretty bad...

Voxengo r8brain

[stonestreet]

@aciddose

Thanks for your reply. You are correct in your prediction about my eyes glazing over. That is a good thing.

So was the printing of '4x oversampling' on the fascia of the player a marketing issue? Do all 'modern' CD players include oversampling as standard?

Cheers.

[bezusheist]

not to get OT, but could some one recommend a (very) good SRC converter for downsampling because i find the one in Logic to suck ass pretty bad...

Voxengo r8brain

thanks, but is this Windows only?...i am on Mac (i should have specified)...there are a few other Voxengo plugs i wish were Mac compatible...maybe some day...

[aciddose]

@aciddose

Thanks for your reply. You are correct in your prediction about my eyes glazing over. That is a good thing.

So was the printing of '4x oversampling' on the fascia of the player a marketing issue? Do all 'modern' CD players include oversampling as standard?

Cheers.

yes, all modern systems use oversampling as a standard. i's pretty much impossible to find a 1980s style dac used in any modern audio hardware.

it's actually more complicated than that though, see "delta-sigma modulation".

[stonestreet]

@aciddose

Thanks for your reply. You are correct in your prediction about my eyes glazing over. That is a good thing.

So was the printing of '4x oversampling' on the fascia of the player a marketing issue? Do all 'modern' CD players include oversampling as standard?

Cheers.

yes, all modern systems use oversampling as a standard. i's pretty much impossible to find a 1980s style dac used in any modern audio hardware.

it's actually more complicated than that though, see "delta-sigma modulation".

Cheers, most informative.