Kronos Piano 44100Hz - SoundCloud's distortions issue

[popmann]

SRC and dithering a full scale sample can cause distortion.

You should not be recording anywhere NEAR full scale with 24bit. Meaning...the track should come no where near zero. Right?

So, you need to be careful bringing it to full scale. It should happen prior to dither, for sure. Assuming you've recorded NOT full scale...try doing the SRC...THEN dither/reduce to 16bit.

[synthjoe]

SRC ... a full scale sample can cause distortion.

Hi popman,
Could you please point me to the source of this theory? I'm unaware of this - SRC should not cause distortion even to the largest signals when applied properly (surely not worse than digitisation itself).
Dither will indeed cause distortion, not only when applied to a full scale signal (in fact, dither is a randomized pattern of distortion applied to the signal - perceived as 'noise'). Interestingly enough, however, this will result in a more pleasing sound to the ear due to removing periodical occurrences of digitisation artifacts (ringing, metallic sound, etc.).

[firstlovedan]

Wow! Great song dude! To be honest, I agree with Gregc. I listened to it on my iPad 3 and 1 found no discernible distortion until the volume was pushed a tad. I heard absolutely no digital artifacts whatsoever. The compression was a tad bit harsh, but again, no distortion. Great Job! My wife and I both listened, and enjoyed your song.
Firstlovedan

[midinut]

That is a great song. If you don't mind me taking this slightly off-topic for a sec, in the YouTube video of The Autumn, how did you get the "movie from panning zooming in and out of pictures" to happen. I really like that style and have some pics I'd like to try that technique with myself just for fun. Thanks in advance.

[mihmix]

Wow! Great song dude! To be honest, I agree with Gregc. I listened to it on my iPad 3 and 1 found no discernible distortion until the volume was pushed a tad. I heard absolutely no digital artifacts whatsoever. The compression was a tad bit harsh, but again, no distortion. Great Job! My wife and I both listened, and enjoyed your song.
Firstlovedan

First of all thank you all for your feeback! I'm really happy to share my music! And thanks to Korg for brilliant instrument!

Now it's really odd... I've just checked and you're right! I listened the soundcloud record on my iPhone and it has no any distortions!
It means that problem is in Windows audio decoders? I listened the record on several different PCs with windows. They all play with distortions.
I use Kronos as the soundcard and I have just checked all devices settings in windows - 24 bit and 48kHz. Now I do not understand why apple sounds well and MS Windows does not??? What am I doning wrong?

That is a great song. If you don't mind me taking this slightly off-topic for a sec, in the YouTube video of The Autumn, how did you get the "movie from panning zooming in and out of pictures" to happen. I really like that style and have some pics I'd like to try that technique with myself just for fun. Thanks in advance.

I used Sony Vegas. It's easy. Put your photo on the video track, set frame positions in two differents time points. And it's done!

[midinut]

[ [/quote]
I used Sony Vegas. It's easy. Put your photo on the video track, set frame positions in two differents time points. And it's done![/quote]

Thanks. You're right, it does sound too easy. lol
I'll check into Vegas. Just starting to dabble with video.
Recently got a Canon T4i and have been getting to know it. It does great HD video, but unless you're on a tripod and have a steady hand. I've got some great pics that it would be great to take the viewer through as you have done. Can you do multiple points or are you limited to just two time points? Again, not trying to derail the thread.

[popmann]

SRC ... a full scale sample can cause distortion.

Hi popman,
Could you please point me to the source of this theory? I'm unaware of this - SRC should not cause distortion even to the largest signals when applied properly (surely not worse than digitisation itself).
Dither will indeed cause distortion, not only when applied to a full scale signal (in fact, dither is a randomized pattern of distortion applied to the signal - perceived as 'noise'). Interestingly enough, however, this will result in a more pleasing sound to the ear due to removing periodical occurrences of digitisation artifacts (ringing, metallic sound, etc.).

http://www.gearslutz.com/board/masterin ... ually.html

Just one a quick google came up with. I've been doing this long enough I don't remember where I learned it specifically...or even details, which doesn't matter to me...solution is to order ITB mastering correctly: any audio processing at full rez, SRC, peak normalization, add dither, reduce bit depth.

[roblof]

SRC ... a full scale sample can cause distortion.

Hi popman,
Could you please point me to the source of this theory? I'm unaware of this - SRC should not cause distortion even to the largest signals when applied properly (surely not worse than digitisation itself).
Dither will indeed cause distortion, not only when applied to a full scale signal (in fact, dither is a randomized pattern of distortion applied to the signal - perceived as 'noise'). Interestingly enough, however, this will result in a more pleasing sound to the ear due to removing periodical occurrences of digitisation artifacts (ringing, metallic sound, etc.).

SRC will cause quantisation artifacts since the signal won't land on the same points on the analog waveform.

If a waveform is normalized to 0dBFS before SRC it is possible that the top of the waveform can get distorted due to not being able to fully translate the numbers as there exist no numbers above 0dBFS. I.e. Clipping the peaks.

This is related to intersample modulation (ISM). Sometimes also called intersample peaking.

Unless you have a limiter or processing that is ISM-aware you should NEVER normalize to 0dBFS

This plays havoc with most lossy codecs including mp3. You should normalize to -3dBFS.

Even with ISM-aware limiters and converters the praxis is to normalize somewhere between 0.1 to 0.5dB below 0dBFS.

Also, keep in mind that adding dither to the 0dBFS signal will make the signal clip into digital distortion at the peaks as well since you're adding a signal to an already maxed out waveform.

So why does it mostly sound bad on windows computers? Because the windows audio mixer defaults to processing the signal with eq, loudness, compression and speaker processing. If your audio material is peaking 0dBFS and also affected by ISM then the builtin audio mixer will clip.

This difference is mostly noted when mixing a song in a DAW and when listening to it on the same computer with the same speakers, it sounds different.

The reason is that the DAW uses asio and therefor bypasses the builtin windows audio mixer and therefor creating different results when listening.

Btw, Soundcloud offers mp3@128kbps for streaming. However, if the upload author click the download checkbox then anyone can download the audio in its original uploaded format. Best of both worlds...

[synthjoe]

To some it might seem that this dialogue is not related to the original topic - please read the last paragraph of this post (before the edit), if in doubt.

http://www.gearslutz.com/board/masterin ... ually.html

Just one a quick google came up with.

Thank you for this source. Another prime example why 'googling' should always be paired with critical evaluation - see below why.

SRC will cause quantisation artifacts since the signal won't land on the same points on the analog waveform.

If a waveform is normalized to 0dBFS before SRC it is possible that the top of the waveform can get distorted due to not being able to fully translate the numbers as there exist no numbers above 0dBFS. I.e. Clipping the peaks.

This is related to intersample modulation (ISM). Sometimes also called intersample peaking.

Indeed. Now, let's examine what I've wrote:

SRC should not cause distortion even to the largest signals when applied properly (surely not worse than digitisation itself).

'Properly' in this context means that you use an SRC algorithm that does not involve low-pass filtering or similar processing to simplify SRC (e.g. the 'least common multiple' principle is used) and is performed on a signal that was digitised intact (i.e. no clipping). Both what roblof wrote and what popmann googled talks about cases when an already clipped signal is SRC'd or some processing is applied to the normalised signal before SRC. What I've talked about is SRC pure. If a non-clipped signal is SRC'd, the distortion inherent to the lower sample rate will apply, but no more (i.e. 'not worse than digitisation itself'). The 2003 AES Nielsen and Lund paper in the thread googled by popmann states exactly this:

If the digital samples were derived from an analog source without overload and not processed in the digital domain, there would be no potential overload problem at the conversion to analog, or to another sample rate.

Of course, if clipped at digitisation (i.e. inter-sample peaking) or processed in a domain where signals above 0 dbFS are allowed (this might be the case for digital signal processors, filters, etc.), then the signal is already distorted pre-SRC and this can indeed be exacerbated by further digital processing. Now, all this just further underlines what both of these two fine gentlemen (roblof and popmann) intrinsically suggested: avoid clipping (record and mix PEAKING at -3...-6 dBFS at most) and use normalisation only once and at the final stages of your process. Distortions originating from ignoring this rule will be exacerbated by further processing or encoding (SRC, MP3, etc.).

EDIT:

...any audio processing at full rez, SRC, peak normalization, add dither, reduce bit depth.

I'd suggest swapping the last two - dithering should be the very last step applied at the final bit depth. If recuding bit depth, dither should be applied again (and preferably non-dithered material should be used for the bit depth reduction). Methinks...