Sample Rate Conversions - hosts compared!

[nuisance sonore]

I use it when mastering, all the time (except lately, after hearing the difference between r8brain pro and free I can't arse myself to convert anything until I can afford r8brain pro! ). I first upsample the file from whatever format it was in to 96kHz and then apply all the processing I need. Then I downsample it to 44,1kHz for CD burning.

Cheers!
bManic

Ok, but what is the logic behind this method? Looks to me that, by upsampling, you are artificially adding data which wasn't there in the original to begin with, which would add nothing to the precision of the calculation but would just make the calculation longer to do, and then downsampling again would remove data from the files, again adding nothing to the precision and possibly taking away from.

I'm not trying to argu here, just trying to understand. I understand that 24 bit calculation would be more precise than 16 bit; I also understand that if you start with material recorded at, say, 88.2KHz, you have more data to process (more samples for every second of recording) than the same at 44.1 KHz so precision (if precision is to be how close it is to the original) should be better. Where I get lost is artificial upsampling, which in my head would imply more destruction of the original than actualy improving it.

Anybody care to explain (in terms anybody can understand)?

[Hink]

I think it's worth noting, and possibly marking on the graphs, the -96dB limit of 16-bit data, and the -144dB limit of 24-bit data. While some of these conversions are obviously higher quality than others, it looks like almost all of them are "good enough for rock-n-roll".

great post and so true...rock-n-roll imo is about imperfections...I too do everything at 16 bit. 44.1k...

[headquest]

I would be interested to see a similar scientific analysis and comparisson of bit-rate conversions/dithering, which most of us probably do more regularly.

Given the apparent significant quality differences when it comes to sample rate conversion, I suspect it is safe to assume that similar disparity exists in the case of bit-depth. While I doubt we could hear the difference in sample rate conversions in an identifyable way, combined with other elements of file conversion and including dithering, these are issues that should concern us all.

I found a discussion of these results on the Sonar forum, where leading Cakewalk developer Ron Kuper expressed surprise... he seemed to be unaware of the relative quality of Sonar's sample rate conversion! Contributers in that thread seemed to acknowledge that the results may be significant and need looking into. It was also suggested that the much-discussed 64-bit mix summing in Sonar 5 may well be compromised by poor sample rate conversion.

http://forum.cakewalk.com/tm.asp?m=669053&mpage=1&key=

Of all the multitrack audio recorders included in the research, Audition is the only one that clearly got good results for sample rate conversion. Sequoia (Samplitude) and Pro Tools got comparitively good results too, while Sonar and Logic are quite a shock. Could this be one factor in a bigger picture which might explain why Audition/Pro Tools/Samplitude have a reputation for "sounding better"?

[jupiter8]

Could this be one factor in a bigger picture which might explain why Audition/Pro Tools/Samplitude have a reputation for "sounding better"?

No. Ignorance is.At least in the Samplitude and Audition case. In the Protools TDM case the 48 bit integer instead of 32 bit float would be the explanation.

[bmanic]

I think it's worth noting, and possibly marking on the graphs, the -96dB limit of 16-bit data, and the -144dB limit of 24-bit data. While some of these conversions are obviously higher quality than others, it looks like almost all of them are "good enough for rock-n-roll".

great post and so true...rock-n-roll imo is about imperfections...I too do everything at 16 bit. 44.1k...

And everybody does rock n roll?

.. ..

[headquest]

Could this be one factor in a bigger picture which might explain why Audition/Pro Tools/Samplitude have a reputation for "sounding better"?

No. Ignorance is.At least in the Samplitude and Audition case. In the Protools TDM case the 48 bit integer instead of 32 bit float would be the explanation.

You seem very sure of that...

Having recorded in Sonar 5, Live 5, Tracktion 2 and Audition 1.5/2.0 I have definitely found it easier to get good results in Audition.

I would guess there are many reasons for that... but I don't think we can so completely dismiss the sample rate research results as a possible factor...

[christianmusicmaker]

Great find Headquest. The results are interesting.

As for the Cakewalk thread...it looks like the Cakewalk forums are down...very rare, but definitely down.

[Hink]

I think it's worth noting, and possibly marking on the graphs, the -96dB limit of 16-bit data, and the -144dB limit of 24-bit data. While some of these conversions are obviously higher quality than others, it looks like almost all of them are "good enough for rock-n-roll".

great post and so true...rock-n-roll imo is about imperfections...I too do everything at 16 bit. 44.1k...

And everybody does rock n roll?

.. ..

where exactly did I say that?

[bmanic]

Where exactly did I say that you said that? I stated a question.

.. another round?

Cheers!
bManic

[bmanic]

I use it when mastering, all the time (except lately, after hearing the difference between r8brain pro and free I can't arse myself to convert anything until I can afford r8brain pro! ). I first upsample the file from whatever format it was in to 96kHz and then apply all the processing I need. Then I downsample it to 44,1kHz for CD burning.

Cheers!
bManic

Ok, but what is the logic behind this method? Looks to me that, by upsampling, you are artificially adding data which wasn't there in the original to begin with, which would add nothing to the precision of the calculation but would just make the calculation longer to do, and then downsampling again would remove data from the files, again adding nothing to the precision and possibly taking away from.

I'm not trying to argu here, just trying to understand. I understand that 24 bit calculation would be more precise than 16 bit; I also understand that if you start with material recorded at, say, 88.2KHz, you have more data to process (more samples for every second of recording) than the same at 44.1 KHz so precision (if precision is to be how close it is to the original) should be better. Where I get lost is artificial upsampling, which in my head would imply more destruction of the original than actualy improving it.

Anybody care to explain (in terms anybody can understand)?

How about you use your ears and do a simple test? Take a piece of music at 44,1kHz. Now take an equalizer plugin, maybe try the default one in your host. Boost with a high shelf as much as you can from 10kHz up. Now do the same after having converted the music to 96kHz with r8brain pro (demo allows conversion up to 1 minute in length). Then downsample the file back to 44,1kHz, again with r8brain pro (using minimum phase is preferrable). Then compare the two files. The original 44,1kHz that you boosted and the "oversampled" one. ok?

Cheers!
bManic

[Hink]

Where exactly did I say that you said that? I stated a question.

.. another round?

Cheers!
bManic

is this "who's line is it anyways" questions only game?

[jones-y]

Anybody care to explain (in terms anybody can understand)?

I've heard it said that digital filters sound better when run at higher sample rates. I can't be bothered to go through all that though... So I don't know for myself that this is true.

I do, however, know that my soundcard sounds better at 96k than at 44.1k. I inquired why, and the answer that I was given was that the anti-aliasing filter doesn't have to work as hard or have as sharp a cutoff.

[pdxindy]

it looks like almost all of them are "good enough for rock-n-roll".

A guitar recorded at 44.1 sounds better in relation to 88.2 than a vst synth. Software synths sound clearly better at the higher sample rate because you lose the aliasing artifacts.

When I listen to my softsynths at 96k they sound distinctly cleaner than at 48k. I almost always work in 48 or 96 for video.

[pdxindy]

I use it when mastering, all the time (except lately, after hearing the difference between r8brain pro and free I can't arse myself to convert anything until I can afford r8brain pro! ). I first upsample the file from whatever format it was in to 96kHz and then apply all the processing I need. Then I downsample it to 44,1kHz for CD burning.

Cheers!
bManic

Ok, but what is the logic behind this method? Looks to me that, by upsampling, you are artificially adding data which wasn't there in the original to begin with, which would add nothing to the precision of the calculation but would just make the calculation longer to do, and then downsampling again would remove data from the files, again adding nothing to the precision and possibly taking away from.

I'm not trying to argu here, just trying to understand. I understand that 24 bit calculation would be more precise than 16 bit; I also understand that if you start with material recorded at, say, 88.2KHz, you have more data to process (more samples for every second of recording) than the same at 44.1 KHz so precision (if precision is to be how close it is to the original) should be better. Where I get lost is artificial upsampling, which in my head would imply more destruction of the original than actualy improving it.

Anybody care to explain (in terms anybody can understand)?

How about you use your ears and do a simple test? Take a piece of music at 44,1kHz. Now take an equalizer plugin, maybe try the default one in your host. Boost with a high shelf as much as you can from 10kHz up. Now do the same after having converted the music to 96kHz with r8brain pro (demo allows conversion up to 1 minute in length). Then downsample the file back to 44,1kHz, again with r8brain pro (using minimum phase is preferrable). Then compare the two files. The original 44,1kHz that you boosted and the "oversampled" one. ok?

Cheers!
bManic

This makes sense. I do the same thing with photography. I upscale 8bit per channel images to 16bit and when moving pixels, double the image size. Then do significant editing, the downsample/scale.


With images, it is better to stick to double, so I guess a conversion from 88.2 to 44.1 is faster and cleaner than 96 to 44.1 - but that is just a guess

[bmanic]

Yes, I think that might be the case. Aleksey can probably confirm this. I just think that the current crop of superior samplerate converters don't need to go exact double. Why I keep mentioning 96kHz is just because I'm so used to working at 48kHz. I have tried to hear a difference when doing 44,1 -> 88,2 -> 44,1 versus 44,1 -> 96 -> 44,1 but I've failed the ABX test miserably.

Cheers!
bManic