Don't let Live resample your audio!

[Funkstar De Luxe]

ugh, can't open the .rar file.

Tested it here, working fine. Make sure you are using the latest version of Winrar. Or you can just make some tone sweeps yourself, very easy. You can do it with the demo of Adobe Audition

[Hercules Robinson]

[quote="Funkstar De Luxe"]
As you can see, the 96k-44.1k is butchered. Very audible artefacts.. quote]

The 48k signal can be encoded fine at 96k with no aliasing, as your picture shows. However, the maximum a sr of 44.k can hope to encode with no aliasing id 44.1 / 2 = 22050 Hz. meaning you can say night night to your 48 k signal. You need a sample rate of at least twice the maximum frequency you want to encode digitally (Nyquist and Shannons law of sampling). Basically, this means at the least you need at least two sample points (snapshots) of a frequency to encode it, one at the +ve amplitude (say 1) and one at the -ve amplitude( say -1), as the sinewave repeats one period you need at least to record its top and bottom part.

Basically what your doing with that example is downsampling. As your downsampling a signal of 48k encoded at 96k with a sr of 44.1 k your gonna get aliasing, you need to anti-alias filter that signal first with a low pass filter with a cutoff of 22050 Hz to avoid aliasing artifacts being introduced.

In my limited knowledge i think thats the case.

[Sten]

Wikipedia has a pretty decent technical explanation of what appears to be happening. Looks like ableton uses a simple resample in this case.

http://en.wikipedia.org/wiki/Dithering

[Funkstar De Luxe]

As you can see, the 96k-44.1k is butchered. Very audible artefacts.. quote]

The 48k signal can be encoded fine at 96k with no aliasing, as your picture shows. However, the maximum a sr of 44.k can hope to encode with no aliasing id 44.1 / 2 = 22050 Hz. meaning you can say night night to your 48 k signal. You need a sample rate of at least twice the maximum frequency you want to encode digitally (Nyquist and Shannons law of sampling). Basically, this means at the least you need at least two sample points (snapshots) of a frequency to encode it, one at the +ve amplitude (say 1) and one at the -ve amplitude( say -1), as the sinewave repeats one period you need at least to record its top and bottom part.

Basically what your doing with that example is downsampling. As your downsampling a signal of 48k encoded at 96k with a sr of 44.1 k your gonna get aliasing, you need to anti-alias filter that signal first with a low pass filter with a cutoff of 22050 Hz to avoid aliasing artifacts being introduced.

In my limited knowledge i think thats the case.

Yes, that is an example of reducing sample rate. My point is that Ableton is not doing a very good job of it.

[lunabass]

Basically what your doing with that example is downsampling. As your downsampling a signal of 48k encoded at 96k with a sr of 44.1 k your gonna get aliasing, you need to anti-alias filter that signal first with a low pass filter with a cutoff of 22050 Hz to avoid aliasing artifacts being introduced.

In my limited knowledge i think thats the case.

Your spot on but the point is that Ableton should have a sufficient anti-aliasing filter to stop this...if this is actually the issue at all.

Many moons ago someone posted a link to a site showing graphs of many DAW's and the accuracy and efficiency of their anti-aliasing filters. Live's graph looked like one of the worst. The graphs showed that frequencies were being allowed through above the Nyquist frequency.

I just cant seem to find the link. It was quite a techy website...anyone?

[Tone Deft]

Many moons ago someone posted a link to a site showing graphs of many DAW's and the accuracy and efficiency of their anti-aliasing filters. Live's graph looked like one of the worst. The graphs showed that frequencies were being allowed through above the Nyquist frequency.

I just cant seem to find the link. It was quite a techy website...anyone?

that would be interesting.

this isn't it but looks like an interesting site
http://www.dawbench.com/

[longjohns]

@ funkstar -

really liked the artwork on your web page

[Funkstar De Luxe]

@ funkstar -

really liked the artwork on your web page

Thank you very much. Very kind of you to say so.

[lunabass]

that would be interesting.

Ok so I've worked it out. I was thinking of the pass band test results on the http://src.infinitewave.ca/ website.

Now I'm assuming (and it is a big assumption) that this is a test of the anti-aliasing filters and shows the difference between the ideal filter and the actual tested filter...can someone with a little more knowledge on this confirm or deny?

Unfortunately the Ableton Live results are no longer up there. Which either means they were floored or that they only have results from current versions and have not tested version 6.

Either way I do remember that the results were quite bad

[bensuthers]

you know, its not a problem if you always work at the sample rate that you intend using.

[lunabass]

you know, its not a problem if you always work at the sample rate that you intend using.

i agree...thats why i just use 44.1k. i've done plenty of hearing tests between 192 and 44.1k and i just cant hear the bloody difference!

[Tone Deft]

funkster this is heavier than I first realized, nice work, thanks for putting up with my doubts. this post of yours

Also, I am not slating Ableton, most DAWs are totally bad at resampling. This is why it should be done by a dedicated application. As far as native DAW resampling goes, Ableton is about mid table. You can see more test here http://src.infinitewave.ca/

speaks VOLUMES.

kinetic, this is also a great link

http://www.bias-inc.com/products/peakPr ... ePaper.pdf
Some more DAW results. The Peak SRC seems also to be very good (at least good enough for me).

veeeeeery eye opening.

Yes Peak does do a good job. this link has most daws tested http://src.infinitewave.ca/

more coolness.


at home now (obviously work's slow lately) I'd like to hear the artifacts out of curiosity.


I'm going to flip flop, the thread title is accurate, if you're going to do sample rate conversion, don't do it in Live.

can I assume this happens with dragging in samples from sample libraries? that seems like a time when this would be an issue for most users most of the time.

I'd like to hear the differences, take an original file, sample rate convert it then, run one of them through the utility plug in fully inverted to subtract out the signal and hear just the noise.

props funkster and others who have shed light on this, thanks.

[Angstrom]

that would be interesting.

Ok so I've worked it out. I was thinking of the pass band test results on the http://src.infinitewave.ca/ website.

Now I'm assuming (and it is a big assumption) that this is a test of the anti-aliasing filters and shows the difference between the ideal filter and the actual tested filter...can someone with a little more knowledge on this confirm or deny?

Unfortunately the Ableton Live results are no longer up there. Which either means they were floored or that they only have results from current versions and have not tested version 6.

Either way I do remember that the results were quite bad

I'm not so sure that the Ableton (passband) tests were bad - it was simply that the test previously determined the slope of the filter at nyquist. As I remember Ableton was distinguished by having no filter (!) A shame they don't list them anymore.

anyway, I took a look at the snazzy new infinite wave tests to compare to your sweep test results for Ableton.

take a look at the sweep tests for Sequoia (94 -> 44.1 conversion), certainly if Abletons sweep tests were bad, then you would expect something much better from Sequoia ?
I think Sequoia costs around $3,000
Sequoia Sweep

like, you know - not perfect at all.

so, if Sequoia looks like that, well perhaps we best not get too upset about how ableton looks eh?
Actually one of the best up there appears to be Sonar !
unless you go for something totally dedicated like Weiss Saracon

Anyway - among the DAWs Ableton is no-where near as bad as some of the competitors if infinitewave is anything to go by.

this means: It is not a case of "Don't let live resample your audio" but more like - "Don't let any DAW re-sample your audio".
because none of them do it well. (Edit - christ, check out the goldwave plot - it's like a carpet!)

[Tone Deft]

also of note is that any artifacts in purple are seriously pushing the dynamic range of your sound card. you can amplify a signal later and those artifacts will be amplified, but -100dB down is certainly 'in the noise'.

the meat of the distortion is the whitish areas, that's around -40dB and up, plus some gain later, could get into the more noticeable area.

[Angstrom]

Just so everyone knows - this is exactly the sort of thread that re-ignites the "Live has a bad sound" sort of myth. Just watch.

People will vaguely remember it in a months time, and mention it to someone who repeats it as a truth.

Future person : "I read somewhere that Ableton aliases frequencies when it renders, that's why I use Logic. Ableton is a toy. There were a load of tests published somewhere, but I can't find them now. There were graphs and everything. Live is just for Loops."



Just to show how inaccurate that perception would be


Logic


Ableton



yet I can guarantee you that someone will recount this thread as providing proof that Ableton is inferior. Just wait.