Soundcloud killed my Songs - mp3 distortion test

[john caldwell]

Your iTune 128 x LAME 128 null is very interesting. There is so much difference across HF and MF bands that I wondered about a time offset between the two files. Could there be a time offset explaining the dramatic results after subtraction?

I've abandoned Soundcloud because of poor fidelity, but I never looked as deeply as you have as to what part of their process degrades fidelity so.

Thanks for offering this.

John Caldwell

[wheeldo]

Hi John,

The iTunes conversion added a little more silence to the start of the track so, while I lined up the first audible samples of each track in Logic as best I could, this might have amplified the differences a little.

But the difference between the two 128k tracks is still incredible without the null test. I could post them separately later if you like?

[KGS]

You should be able to measure the delay. Use another track with no silence in the beginning at all. Convert it through iTunes and then measure the delay.

[wheeldo]

Thanks for that KGS!

Here's the re-done 128k comparison, along with the separate iTunes and LAME compressed mp3s so you can A/B them.

Does anyone have any idea what it is that causes this difference and, more to the point, how I can get around it?!

[john caldwell]

Originally Posted by KGS

You should be able to measure the delay. Use another track with no silence in the beginning at all. Convert it through iTunes and then measure the delay.

What if the conversion algorithm truncates the file from the beginning? I have no knowledge what so ever if this occurs, but do we have assurance that all these processes simply elongate the files?

Again if there are time offsets, the null can be difficult to interpret - at least for HF events. As we move down the frequency spectrum, offsets would need to increase in duration to produce areas that won't null, of course.

John-

[wheeldo]

For anyone who's interested in this, I've just discovered one of the main things LAME conversion struggles with is extreme panning. The tracks I've tested are all double-tracked acoustic songs with one guitar hard left and one hard right, so I tried converting a mono version of the same mix using the same settings - it still sounds like a 128k mp3, but it's no worse than most, and is far, far less 'crunchy'!

So if your mixes are sounding terrible on soundcloud, it might just be worth winding the panning in, or even uploading a mono version for streaming purposes.

[JoleFIN]

There's one bootleg remix of our song that has been spread widely for a few years, and I accidentally set it for sale for a few days (wanted to have Youtube sync licensing done with the help of the distributor in the first case) ... I had it removed since it was in mono 64kbps. Lo and behold, it still sold 89 copies within the few days it was for sale. The "normal" people / enduser won't differentiate between different mp3 codecs or kbps ratios, if they don't differentiate between mono and stereo.

[ludeers]

Hi wheeldo,

I had very big problems with my synth sound in Soundcloud. I searched through the internet for more than month, tried every advice available (headrooms, highcuts, dithering, various formats, even started to change the timbre etc), but nothing seemed to help. At LAME 192 kbps mp3 synth sounded good, but at 128 kbps it was completely ruined no matter what I tried to do.

Few days ago I noticed in Soundcloud forum that they use Joint Stereo when they are transcoding the tracks. At the first moment I didn't pay attention to that, but after reading your thread - that panning could cause the problems, I started to analyse my synth timbre from very beginning.

Everything seemd ok, until I opened the Multimeter in Logic and in Stereo Image noticed, that the main stereo density of each note played on synth each time appeared in opposite side of the centre (left, then right, then again left, right etc).

In ES 2 synth I immediately unclicked the Polyphonic button and instead clicked Mono, a little bit sculpted the timbre, then bounced it, uploaded on Soundcloud and guess what? Everything at last was ok! No more that irritating crunchy sound on high frequencies!

So, if anyone has problems with Soundcloud sound quality, please check your stereo sounds and preferrably use mono (not only synths, but also drums, percusioins etc.).

[LimpyLoo]

Originally Posted by wheeldo

For anyone who's interested in this, I've just discovered one of the main things LAME conversion struggles with is extreme panning. The tracks I've tested are all double-tracked acoustic songs with one guitar hard left and one hard right, so I tried converting a mono version of the same mix using the same settings - it still sounds like a 128k mp3, but it's no worse than most, and is far, far less 'crunchy'!

So if your mixes are sounding terrible on soundcloud, it might just be worth winding the panning in, or even uploading a mono version for streaming purposes.

Thanks for this observation. My stuff is pretty pan-heavy, and I feel like it takes quite a beating from conversion.