Sample rate?

[Sifunkle]

Feel free to scroll to the last paragraph for the main point of this OP, otherwise have fun with my ramblings

Forgive me if I haven't thoroughly sifted through other threads, but I typed this up in the "Newbie" forum before deciding it was better suited here. Which should tell you something about the kinda stupid question to expect And sorry if it's not strictly about mastering per se.

Before I'd done much research, I used to just think 'Well I'm aiming to make demo CDs, so I'll just stick to 16 bit and 44.1 kHz for everything, the whole way from guitar to discman'.

Now that I'm (very) slightly more educated, I realise that was probably a silly idea, so I'm now planning to record everything in 24 bit depth. I'm still a bit confused about what sample rate to record at though.

I'd like to keep it fairly low so I don't chew through my hard drive (I'm a poor student), but I'd also like to have a shot at some basic (ie. amateurish) mastering, given that I won't afford (artistically or financially) professional mastering any time soon.

I understand that for mastering, I'll want to have a high sample rate. Would it be correct to assume that if I've recorded all my tracks in 44.1 kHz, there's no point bumping up to 96 for the track I'll attempt mastering on (as there's only 44.1 thousand samples per unit time, therefore extra samples will just be blank)? Or will that bump actually work, because the samples of all the different tracks will be timed ever-so-slightly differently to each other so that there's actually more than 44.1 kHz present in the mix?

If you want the question phrased simply, without all my pondering: Can I record my tracks at 44.1 kHz, mix, then bump up to 96 kHz for my finished mix, prior to mastering?

Thanks in advance,

Si

 

[Mo Facta]

Uy yuy yuy. There are a few things to know here. I'll try my best to explain them

Firstly, if the production is going to end up on CD, there is no advantage to upsample anything to 96kHz if you're already at 44.1. I would highly recommend staying at 44.1 throughout all phases. However, THAT BEING SAID, the main advantage to a session/project with a higher sample rate is that processing like reverbs will sound better or any other effect that generates time-based artifacts.

Many engineers prefer working at a higher sample rate from the get go, like 96kHz, and then downsampling to 44.1 during mastering. Many prosumer converters also sound better at 96kHz so that's a good case for going with it.

But honestly, I would work in 44.1 and be done with it. Don't concern yourself with all of this sample rate trickery as it will almost certainly bring you more cons than pros.

In terms of "bumping up" to 96kHz before mastering, if you really want to do it, you have to use a sample rate converter and the one that I can recommend is the Voxengo r8brain. It is probably the best software SRC you can get. Just be aware that upsampling to 96kHz does not add any more time-based resolution to the audio than is already inherent but merely allows the mastering software to operate at that rate and thus any processing that is applied will be in that domain.

This may or may not give you better results.

Cheers

 

[sjjohnston]

There are real live mastering guys here who might know some stuff that's not obvious to "ordinary people," but:

The primary effect of sample rate is that you have to limit the frequency range of your material so it doesn't exceed (a little less than) half the sample rate. Thus, when you record something at 44.1k, your AD converter must apply a lowpass filter that cuts off everything above about 20k. If you "downsample" something that's already in digital form, you'd apply a similar filter (in software) at that point. The possibly bad (or possibly imperceptible) downsides are (a) you lose the harmonic content over 20k that was present in what you originally recorded, which may not be audible anyway (or, depending who you ask, may be in some strange and subtle way nobody exactly understands); and (b) no lowpass filter is perfect, so it'll have some effects other than simply blocking everything over 20k, which effects may or may not be readily perceptible (and if they are, may or may not be negative).

If you "upsample:" (a) whatever supersonic (over 20k) information was present in the original signal is long gone and won't magically reappear and (b) whatever damage the lowpass filter previously did in the audible range isn't going to be undone either, magically or otherwise.

About all you'd do would be to add the capability to add new supersonic-frequency-range information to your recording. I suppose that might happen if you were to use some mastering tool that adds harmonics to what's already there (like a BBE maximizer that runs beyond the audible range). Whether and why you'd want to do this - particularly if you're going to wind up taking it back out when you convert it back to CD form - is mysterious to me. Perhaps a real mastering engineer knows something here that I don't.

Another factor: when you bump up the sample rate, the recording won't have blanks in the new samples; rather, it'll interpolate values based on the ones you have.

If you were, say, to go from 44.1k to 88.2k, it would just insert a new sample between every existing pair, and "guess" at the value using an interpolation algorithm. At the simplest, such an algorithm would simply insert a value halfway between the two on either side of it; more clever algorithms take account of the waveform and the fact it's not just linear. In any event, so long as it stays somewhere in between the values on either side, it shouldn't make a lot of audible difference, as the resulting error would exist in the super-20k frequency range anyway.

If you go from 44.1k to 96k, it's not simply a matter of inserting new samples, but you've got to replace the data you have with new, made-up-by-an-algorithm, data that fits the sample rate. The resulting resampling may not do a lot of harm, but I'm having trouble figuring out how it's going to do much good.

 

[rfahey86]

Keep in mind the threshold of human hearing is around 21Khz

 

[sjjohnston]

Keep in mind the threshold of human hearing is around 21Khz

For most people it's a good bit lower than that (more like the teens at best, for people who aren't themselves teens), but there are persistent beliefs/theories/religious tenets holding that the super-sonic frequencies have some effect on perception ... for all I know, they might be right, and - in any event - I'm not in a position to prove they're wrong.

 

[miroslav]

I believe that if you up-sample before you edit/mix...it might help improve the quality of any FX or processing you will be adding/applying to your 44.1 Hz tracks, though the actual tracks will not benefit from the up-sample.

 

[Sifunkle]

Ahh, that's definitely helped my understanding: I think the main conceptual problem I had was that I hadn't linked the sample rate (frequency) as being relevant to the actual frequencies of the sounds I'd be hearing (eg, obviously capable of producing a 16 kHz sound, but not 100 kHz), but that's evident now thanks to your responses

However, THAT BEING SAID, the main advantage to a session/project with a higher sample rate is that processing like reverbs will sound better or any other effect that generates time-based artifacts.

I believe that if you up-sample before you edit/mix...it might help improve the quality of any FX or processing you will be adding/applying to your 44.1 Hz tracks, though the actual tracks will not benefit from the up-sample.

In light of that, I think I'll probably just stick to 44.1 throughout the whole process, as all I think I'm likely to use at mastering will be EQ, compression and a limiter. Unless there are any as-yet-unvoiced reasons I shouldn't?

Relating to super-audible frequencies, my hypothesis based on my current understanding is that that they would influence the audible frequencies; I imagine it would be similar to using an LFO to change a synth tone. But perhaps in most cases, the effect would be imperceptible, as the modulation would (obviously) be at a frequency beyond our limit of resolution. But perhaps if you had recurring phase cancellation or something, it might become perceptible? I'm just guessing here, haven't really read up on it too much. Ignore me as the ignorant newb I am

And the range of human perception of sound I had in mind was 20 Hz to 20 kHz (roughly), with the upper end declining with age and damage (I'm regretting some of the loud shows I've been to now + liberal use of earphones as a teen ). Is that about right?

Also, according to...

The primary effect of sample rate is that you have to limit the frequency range of your material so it doesn't exceed (a little less than) half the sample rate. Thus, when you record something at 44.1k, your AD converter must apply a lowpass filter that cuts off everything above about 20k.

... does this mean that there's potentially around 1-2 kHz of frequencies inaudible to humans on a standard CD? Maybe my dream of 'music for dogs' isn't doomed to failure...

Thanks for your replies

 

[Bobbsy]

Hmmm. That's an interesting thought.

There's no doubt that a lot of effects can sound better on tracks recorded at a higher sample rate but I'm less convinced that these advantages also apply after up-sampling. Since the up-sampling process is already using clever algorithms to "fill in the dots" for the samples that aren't actually there, this might introduce exactly the same issues as applying the effects to a 44.1kHz files in its native state.

This could be an interesting experiment when I get some time--though that might not be for a few weeks.

Bob

 

[Sifunkle]

I expect whether the effects would be noticeably different after up-sampling would depend on both how the effect was mathematically applied and on the exact algorithm the converter used. If an effect is applied based on the relationship between one sample and the next, you might see zero difference at different sample rates if the converter is linear, as the converter would have just inserted another point onto the 'line of best fit', not actually changing the relationship at all.

I suppose as was pointed out earlier, a better converter would do better modelling of the initial file in order to fill in the gaps, and effects would be much more noticeable. Although I'm sure many effects are applied differently than what I've suggested. This is just supposition, if anyone more informed wants to correct me, go for it

To be honest, I think I'd completely underestimated the finesse that converters are capable of. Doesn't surprise me, I'm really not a tech-head But I suppose there's good money to be made in producing converters that can offer much better quality, so I will try and be more appreciative of the sophistication of such things.

 

[Mo Facta]

a

I believe that if you up-sample before you edit/mix...it might help improve the quality of any FX or processing you will be adding/applying to your 44.1 Hz tracks, though the actual tracks will not benefit from the up-sample.

+1

Cheers

 

[Mo Facta]

I expect whether the effects would be noticeably different after up-sampling would depend on both how the effect was mathematically applied and on the exact algorithm the converter used. If an effect is applied based on the relationship between one sample and the next, you might see zero difference at different sample rates if the converter is linear, as the converter would have just inserted another point onto the 'line of best fit', not actually changing the relationship at all.

I suppose as was pointed out earlier, a better converter would do better modelling of the initial file in order to fill in the gaps, and effects would be much more noticeable. Although I'm sure many effects are applied differently than what I've suggested. This is just supposition, if anyone more informed wants to correct me, go for it

To be honest, I think I'd completely underestimated the finesse that converters are capable of. Doesn't surprise me, I'm really not a tech-head But I suppose there's good money to be made in producing converters that can offer much better quality, so I will try and be more appreciative of the sophistication of such things.

The playback of the converters just represent whatever is going on within the DAW. The sample rate conversion "algorithm", as you put it, is not housed in the converters.

The whole point of this is whether or not the OP will gain resolution on processing and FX during mastering if he up-samples his mixes to 96kHz from 44.1 before hand. The interpolation (or filling-in-the-blanks) will be done by the sample rate conversion software (such as the Voxengo r8brain) and not by the DA converter. The DA will simply play back whatever it is fed.

The sonic aesthetics of the DA converter affecting the resulting audio is another discussion altogether.

Cheers

 

[sjjohnston]

Additional thoughts:

My previous notes are based on the assumption that upsampling would never involve an AD converter at all, as it's simply a conversion from one digital format to another. I don't know if that wasn't sufficiently obvious that it didn't require stating. I suppose you could, if you were so inclined, upsample by doing a DA conversion, then recording back through an AD converter at a higher sample rate. If your goal is maintenance of the signal, this seems silly; the only reason to do it would be a sort of "effect," akin to recording a digital signal to tape, then back to digital. I've not heard of anyone doing this and, in any event, it's a whole other subject than the original question.

In his later post, the "converter" the OP referred to wasn't (if I'm understanding his post) the AD converter, but the sample-rate converter that's in the software. Of course an algorithm is something that exists in software (if you want to get all semantic, software is algorithms).

High-frequency content can affect sonic frequencies in some situations: one obvious one is that if you attempt to do an AD conversion on an analog signal with content whose frequency is more than half the digital sample rate, you'll get aliasing. That's an effect you generally don't want, and exactly why AD converters have to apply a lowpass filter to the analog signal before recording.

I'm not seeing the argument that the following process:
Original analog signals -> 44.1k mix -> 96k mix -> mastering -> 44.1k mix
can do anything good in terms of fidelity or harmonic content. It is more likely to introduce unwanted distortion and loss of fidelity. Of course, introducing distortion and loss of fidelity may be what you're going for (that's basically the entire purpose of a guitar amp in recording). It could be a more subtle "lo-fi" effect than the usual bit-rate reduction, reamping, intentional distortion, extreme EQing, etc. Other than that, it doesn't seem to be a good idea:

The only plus:
- The 96k master can introduce new super-sonic-frequency information that's not in your 44.1k mix. That, I think, is the only way the processing you do to the master would be "higher quality," which is a somewhat vague phrase standing by itself.

On the other hand:

- The 44.1k -> 96k conversion can introduce fresh quantization issues, and can't really do anything other than degrade the fidelity of the signal. Going 44.1k -> 88.2k would be better. The degradation might well be imperceptible, though no more imperceptible than the supposed improvement in "quality" of the mastering effects referred to above.

- The 96k -> 44.1k (or 88.2k -> 44.1k) conversion at the end will take all the super-sonic information introduced by the mastering processes back out again.

- The lowpass filter applied (digitally) to the make the 96 or 88.2k -> 44.1k conversion will introduce new infidelities that you'd avoid if you never did the up/down sampling.

Final note: This is quite different from this hypothetical process
Original analog signals -> 96k mix -> mastering -> 44.1k mix
which could be at least slightly preferable, as it avoids the problems created by the lowpass filter at original recording (or at least moves them way up out of the audible range) and preserves the supersonic information in the original signal.

Indeed, in terms of fidelity:
44.1k-all-the-way > 44.1-to-96k-to 44.1k
for some of the exact same reason that
96k-to-the-last-step > 44.1k-all-the-way

I think those "greater than" symbols represent, as a practical matter, very minor differences in actual perceivable sound. At the end of the day, I don't think you really would do great harm by the up- and down-sample: you would just produce marginal, barely-if-at-all perceivable infidelities.

Of course, the same is true of the comparison between recording at 96k from the get-go vs. using 44.1k from the beginning to the end. That's why I'd opt to record the thing at 44.1k in the first place, and just keep it at that rate.

 

[NYMorningstar]

"Additional thoughts:" All good stuff, thx SJ

 

[miroslav]

I expect whether the effects would be noticeably different after up-sampling....

I think you may have missed my point.

I'm not saying up-sampling improves existing FX/processing (or recorded tracks for that matter)...
...I'm saying that if you up-sample your tracks, then the FX/processing that you apply to those tracks AFTER you up-sample, will now by applied at a higher rater.
IOW...your tracks don't benefit (44.1 kHz)...but you up-sample them to say, 88.2 and THEN apply FX/processing, which means all those reverb tails and EQs and compression is running at 88.2, so at least the FX/processing will be of higher quality...which may or may not make a difference to the final mix, depending on how much FX/processing is being added and how "fine" it needs to be.
My attitude has always been that if you have HD space and CPU power...you lose nothing from running at higher rates, but you could(?) gain a little something.
I usually track at 88.2 kHz...it costs me nothing to do that, regardless if *I* can really hear any difference or not.
I view it as taking out an insurance policy.

 

[PDP]

My attitude has always been that if you have HD space and CPU power...you lose nothing from running at higher rates, but you could(?) gain a little something.

Same here, I've also heard that at 96k the higher frequency samples are thicker (or there are more of them) and thus the highs sound less "grainy"

Not sure if thats true, but like Miro is saying, I have the DAW power to run it so...
Why not?

 

[Sifunkle]

In his later post, the "converter" the OP referred to wasn't (if I'm understanding his post) the AD converter, but the sample-rate converter that's in the software.

That's exactly what I meant. Being the amateur I am I probably used the wrong terminology and confused things. I'm more of a concepts than a details guy

I think you may have missed my point.

I'm not saying up-sampling improves existing FX/processing (or recorded tracks for that matter)...
...I'm saying that if you up-sample your tracks, then the FX/processing that you apply to those tracks AFTER you up-sample, will now by applied at a higher rater.

Oh, I got what you meant, sorry to have misled! My later post was more of a ramble on some new things that were drawn to my attention (sample rate conversion algorithms, and above-audible frequencies) not really in the context of my initial question. I probably just confused things, sorry.

Anyway, in light of what seems to be several conversions with a negligible perceptible impact, and because I don't have supremo CPU and HD space, I think I'll just stick with 44.1 kHz all the way. At least until I get better at everything else; at this stage, slight improvements from mastering probably aren't gonna do much for the overall quality of my stuff

 

[Mo Facta]

FWIW, I find that high end converters have less of a margin of quality between sample rates.

Cheers

 

[TobyC]

A very good compromise is to work at 24bit/44.1kHz and dither to 16bit as a final stage. I know several plugins and DAW's (GuitarRig4 and Sonar X1PE) have options that allow them to process at higher resolutions within the software but bring their output back to 24/44.1. This is designed to maintain quality without the need for upsampling the project to 88.2 or 96 kHz without all the associated overhead.

Hope this helps.

 

[Farview]

Some plugins up-sample to process and down-sample back to the project rate all by themselves.

You just need to use the sample rate that sounds the best on your converters/interface. The difference in sound has much more to do with the design and implementation of your interface that it does the sample rate you are running. As was pointed out, the super high end converters don't sound any different at different rates, only the cheaper ones do. One of the designers of the most respected converters in the world once said "If different sample rates sound different, your converters are broken".

Besides, if recording at high sample rates made a huge difference, there wouldn't be any debate. No one argues that a Prius is faster than a Lamborghini...

 

[Mo Facta]

As was pointed out, the super high end converters don't sound any different at different rates, only the cheaper ones do. One of the designers of the most respected converters in the world once said "If different sample rates sound different, your converters are broken".

This is because of the chosen analogue components that manufacturers use in their products. Usually it's the first thing they skimp on because of budget constraints and accordingly it's the first thing that'll make your interface sound crap.

FWIW, many manufacturers use the same converter chips, especially pro-sumer stuff.

Cheers