Upsampling and your Workflow

[OpusOfTrolls]

If you are working with plugins, absolutely. Less aliasing, less EQ curve distortion, and better time/phase resolution in the highest frequencies.

[wado1942]

It really depends on what I have to do.
First, I must clear the air and say that few converters work better at higher sample rates and some of them actually perform worse. Therefore, if I'm working OTB, no upsampling... end of story. If I'm working ITB and it's a quickie job or just doesn't need much processing, I usually don't upsample. If it's a job where I need to spend more time on it and/or need a lot of processing, I tend to double-sample.
Here's the thing, anything you add that affects dynamics, like compression, expansion, clipping etc. introduces false harmonics. These false harmonics can exceed the Nyquist rate and create aliasing. In that case, upsampling can drastically reduce this affect. I have reason to believe that some delay based processes like EQ and reverb (especially impulse response reverbs) are more precise as well. A lot of people will disagree with me, but I tested some processes at 48KHz and 192KHz, like an impulse reverb, and instantly heard the difference. EQ, it's harder to tell, but being able to have precision of better than one output sample may lead to more natural EQ if the code is written to take full advantage of the resolution.
Simple processes like fades, group level adjustments & basic edits don't benefit from higher sample rates.

[engmix]

Originally Posted by wado1942

It really depends on what I have to do.
First, I must clear the air and say that few converters work better at higher sample rates and some of them actually perform worse. Therefore, if I'm working OTB, no upsampling... end of story. If I'm working ITB and it's a quickie job or just doesn't need much processing, I usually don't upsample. If it's a job where I need to spend more time on it and/or need a lot of processing, I tend to double-sample.
Here's the thing, anything you add that affects dynamics, like compression, expansion, clipping etc. introduces false harmonics. These false harmonics can exceed the Nyquist rate and create aliasing. In that case, upsampling can drastically reduce this affect. I have reason to believe that some delay based processes like EQ and reverb (especially impulse response reverbs) are more precise as well. A lot of people will disagree with me, but I tested some processes at 48KHz and 192KHz, like an impulse reverb, and instantly heard the difference. EQ, it's harder to tell, but being able to have precision of better than one output sample may lead to more natural EQ if the code is written to take full advantage of the resolution.
Simple processes like fades, group level adjustments & basic edits don't benefit from higher sample rates.

Interesting perspectives. Thanks for the details Wado. I've certainly heard differences in tracking records at 88 verses 44k. I'm thinking that has to be on the converter end. I tend to use my Lavry 4496's. I feel like i've been making records for too long to be dealing with a placebo effect, then again i am human, so who knows.

[Joelistics]

I always play back (from Pro Tools) at the sample rate the client delivers, but capture at 96khz (into Pyramix). For me it's not just the plugins that sound better at higher sample rate but the analog chain itself too. I have done a lot of A/B testing and it's quite obvious that 96k sounds best for what I'm working with (even after downsampling to 44khz afterwards). The silky highs... yummie...
You'll get varying opinions on this so best to make your own tests and use your ears..

[Table Of Tone]

Although I used to always capture at 44.1/24, it's making more sense these days to capture at 96/24, especially with the MFiT program.

Let's face it, that's the way it's going!

If you're running any plugs or hardware digital DSP, post ADC, they can also sound a little more natural at the 96.
Use of a really good SRC to 44.1 can still get you a better sounding result in the end for CD, etc, and you've also still got your 24k nyquist if you need to down sample from 96 to 48 for video audio.

I don't really see any point in going higher than 96 though.

There is a sweet spot and it also depends on your converters!

[Huntley Miller]

I wouldn't upsample the source, but I've always captured my chain at 96/24 and the digital processing after that benefits from the higher sample rate.

[engmix]

Any particular reason for running at 96 verses 88k. Is this something you have all arrived at via listening tests, or is there some science to it, because i would think that the conversion from 88 to 44 would be a bit more graceful.

[Huntley Miller]

Originally Posted by engmix

Any particular reason for running at 96 verses 88k. Is this something you have all arrived at via listening tests, or is there some science to it, because i would think that the conversion from 88 to 44 would be a bit more graceful.

I have a custom A/D that only runs at 96k. The whole 88 to 44 SRC being better than 96k to 44 based on less math is a myth.

[wado1942]

The whole 88 to 44 SRC being better than 96k to 44 based on less math is a myth.

Well, it's essentially the same amount of math, but the differing results are no myth.


For the record, I try to capture at 88.2KHz or 96KHz as much as possible. I'm a firm believer in high resolution audio. I can't tell a difference between capturing at 96KHz and 192K, even though some plugins do seem to be a little cleaner at the higher rate. It's just too taxing to work at that rate for virtually no benefit. However, if the source is 44.1KHz, destination is 44.1KHz and I'm working OTB or don't need much processing, I just leave it as is.

[Huntley Miller]

Originally Posted by wado1942

Well, it's essentially the same amount of math, but the differing results are no myth.


For the record, I try to capture at 88.2KHz or 96KHz as much as possible. I'm a firm believer in high resolution audio. I can't tell a difference between capturing at 96KHz and 192K, even though some plugins do seem to be a little cleaner at the higher rate. It's just too taxing to work at that rate for virtually no benefit. However, if the source is 44.1KHz, destination is 44.1KHz and I'm working OTB or don't need much processing, I just leave it as is.

Neat picture. I have no idea what it's meant to illustrate.

[engmix]

Originally Posted by Huntley Miller

Neat picture. I have no idea what it's meant to illustrate.

It appears to show that one has more aliasing than the other, but not sure which is which.

[Huntley Miller]

Originally Posted by engmix

It appears to show that one has more aliasing than the other, but not sure which is which.

It doesn't give any information about the test signal or which SRC is used. Is this the one that came with his Mom's Dell?!?

[Babaluma]

I used to always transfer and capture at the source file's rate, and then SRC to the destination format if needed, but recently I've been experimenting with upsampling everything to 96, processing (D/A/D), and capturing at 96, again before SRC to desitination format if needed.

Nothing conclusive so far, in fact it seems quite arbitrary. Some tracks seem to benefit from upsampling, whilst others don't. More tests needed!

[lucabasano]

I used to upsample to 96 before analog chain and capture 44.1/24 bit,indeed cause weiss or zsys eq I use before the da running best,not for other reason.

luca

[Table Of Tone]

Originally Posted by engmix

Any particular reason for running at 96 verses 88k. Is this something you have all arrived at via listening tests, or is there some science to it, because i would think that the conversion from 88 to 44 would be a bit more graceful.

96/24 wav's are automatically accepted for MFiT.
Other sample rates are only considered.

I'm guessing Apple are just trying to keep everything stored at 96/24 as a possible future default format?

It really depends on the SRC in question as to whether 88.2 goes down to 44.1 better than 96 does.