Live's resampling quality aka Sample Rate Conversion

Timur · Post by **Timur** » Tue Mar 18, 2008 1:56 pm

Don't take it personal, but I think the quality of Live's Resample Engine (SRC) is sub-par, at least when resampling very low sample-rate clips. I don't know if it's bugged or not, but either way it could probably be improved in a future release.

Here is a recording made with the inbuild microphone of my Nokia 6230i mobile phone that saves its files to 8 KHz / 16-bit AMR format. I did those recordings to test the capabilities of the mobile, decided that it was time to take vocal lessons again after that recording.

My Creative X-Fi's SRC serves as reference, since it's hardware SRC is superb with more than 50% of the X-Fi's processing power aka transistors solely being dedicated to only that one task. All files were played trough the X-Fi at 48 KHz (=6 x 8 KHz) and digitally recorded directly through it's Control Panel ("What you hear") at 48 KHz / 24-bit. Winamp was using the AudioBurst FX output plugin at 24-bit, Live was using the X-Fi ASIO drivers at 24-bit. For capacity/bandwidth reasons they were then encoded to MP3 via Lame at 386 kbit CBR. Warp and Fade were disabled for the Live samples.

1. Nokia AMR Player: nokia_amr.mp3 - 3.27MB

2. Winamp AudioBurst FX, inbuild Resampling OFF (aka X-Fi SRC): nokia_x-fi.mp3 - 3.23MB

3. Winamp AudioBurst FX inbuild Resampling LOW (Characteristics SHARP, Noise Floor -86 dB aka worst possible setting, lowest CPU-load): nokia_audioburst_low.mp3 - 3.25MB

4. Live HI-Q: nokia_live_hi-q.mp3 - 3.28MB

5. Live Default: nokia_live_low.mp3 - 3.26MB

Live's Default Resampler is kind of a nice low-fi effect, I'd call it a bitcrusher or something like that and maybe its harsh character is even intented!? But even the HI-Q is worse than even the lowest possible setting of the AudioBurst FX Resampler (offers the options LOW, MEDIUM, HIGH, EXTREME) without seeming to save any much CPU-load in comparison. To my ears Live's Resampler exhibits distortion and noise even in HI-Q when compared to the others resamplers and when compared to the original (via AMR Player through Audiophile 24/96 at 8 KHz).

So if this is a bug, then please correct it. If this is your real current implementation then I suggest to rename the default Resampler into LO-Q, the now HI-Q one into Default and last but not least please put a real HI-Q resampler into Live.

Timur · Post by **Timur** » Tue Mar 18, 2008 2:06 pm

Here is some graphical evidence of that something is wrong with Live's Resampler:

Original file / X-Fi / AudioBurst FX:

Live HI-Q:

Live Default:

Timur · Post by **Timur** » Tue Mar 18, 2008 4:33 pm

So which applications do others use for quality resampling?

I put this link in another thread already. That website compares several hard- and software solutions:

Timur wrote:Nice webpage that allows graphically comparison of any different SCR algorithms (both hardware and software). Most astounding to me is the change of Apple's Core Audio SRC from Tiger to Leopard, obviously they changed it for the better (eventhough the Sine Sweep shows higher peaks its average level of artefacts seems to be lower).

http://src.infinitewave.ca/

sabiwa · Post by **sabiwa** » Tue Mar 18, 2008 4:49 pm

What is this?

Timur · Post by **Timur** » Tue Mar 18, 2008 5:11 pm

Some would answer: Timurism!

But if this is a serious question, this is about the quality of Live's inbuild sample-rate conversion. Whenever you load an audio clip into Live that uses a different sample-rate than the rest of your Live Set (or better than the sample-rate that you set via Preferences) the sample-rate has to be converted (like when loading a 44 KHz sample into a 48 KHz set).

If you do not use the HI-Q button within clip-view then Live will use its Default conversion (which is the last audio-example of my original post and graphically shown by the last graph). If you use the HI-Q button then it uses a supposely high(er) quality method. But unfortunately this still seems to lack audibily when being compared to other methods that seem to be just as CPU friendly. Also the frequency-graph of even the HI-Q method shows that frequencies are added to the original signal that should not be there (like in my example there should not/cannot be any frequencies above 8 KHz because the original file is cut at 8 KHz, but with Live's algorithms there are alot of them).

rozling · Post by **rozling** » Tue Mar 18, 2008 5:17 pm

sabiwa wrote:What is this?

A CENTRE FOR ANTS???

How are they supposed to learn if they can't even fit inside the building?

Timur · Post by **Timur** » Wed Mar 19, 2008 1:16 am

Yes, Ants must be the answer!

Pitch Black · Post by **Pitch Black** » Wed Mar 19, 2008 1:40 am

Timur · Post by **Timur** » Wed Mar 19, 2008 1:58 am

What would he use for sample-rate conversion? A curling iron?

dcease · Post by **dcease** » Wed Mar 19, 2008 2:08 am

did you run a phase test? lol!
as i've stated before, i don't trust that damn resample... maybe i'll try the hi-q, now... lol! i never gave it a thought, if it is all live stuff it's the same rate anyhow. i mangle loops, so i don't worry bout a bit of whatever happening to the file

eh, i hardly use it for track mixdown, prolly cause i'm convinced it sux, if your charts are accurate, which i believe they are, this may well be true.

but others use it, and like it, so i dunno

Timur · Post by **Timur** » Wed Mar 19, 2008 2:23 am

I have to underline that the charts were screenshot at the very end, I did not see them myself before. I just noticed how bad Live's playback of those files sounded when I imported them and started diggin deeper from there. It's the ears that judge first, but apart from that I do find it disturbing that obviously Live adds lots of frequencies above 8 KHz to a sound clip that was recorded at only 8 KHz. What does it do when you add a 44 KHz clip to an 88 KHz set? Do you realize that this will raise the RMS energy/volume of the clip and thus lead to a couple of problems with metering, compression and limiting?

Anyone eager to test that?

Nod · Post by **Nod** » Wed Mar 19, 2008 2:34 am

Timur wrote:So which applications do others use for quality resampling?

Have you tried Audiomove - cross platform and free?

http://www.lcscanada.com/audiomove/

Definitely up there with the payware likes of Voxengo's R8Brain Pro at $130

( EDIT: File under ignore - think I've misunderstood your request Timur

)

b0unce · Post by **b0unce** » Wed Mar 19, 2008 2:38 am

http://www.audiofile-engineering.com/waveeditor/

Tone Deft · Post by **Tone Deft** » Wed Mar 19, 2008 2:42 am

Timur wrote:Some would answer: Timurism!

LOL! totally.

timur - can you characterize how loud the artifacts are? I just got home from work, don't want to think for a while. I think you resurrected that site that compares SRCs (sample rate conversion) for various DAWs.

the important thing with analyzing this stuff is to know how loud the artifacts can be, SRC is a really difficult thing to pull off quickly, with linear phase and low distortion. I draw the line at -60dB for stuff I care about in making tunes. for a rigid measure of an audio system, under -100dB is 'mission accomplished' although the AD1896 (mentioned below) is around -130dB.

btw your sound card isn't using transistors really, it's either the Analog Devices AD1896 SRC or a similar chip. you can argue that all silicon chips have gates and transistor is a generic type of gate, but anyway, it's all about the AD1896, my guess.

careful jumping to conclusions too.

Timur · Post by **Timur** » Wed Mar 19, 2008 2:45 am

b0unce wrote:http://www.audiofile-engineering.com/waveeditor/

That one was suggested by Tarekith as well (at least he seems to use it). Thanks.