Rendering in T is not as perfect as I thought

[leeturne]

I posted the same problem on this board a couple of months ago. It definitely truncates the audio and there is a signal breakup and degradation. The post is here: http://www.kvraudio.com/forum/viewtopic ... 931#942931

[semiquaver]

OK I'm really freaked out about this - I had the same (or considerably worse) results with and without dithering at 16 and 24 bits. I'm off to another DAW to make a comparison.

[semiquaver]

I just made a comparison in DP and heard the expected tails with and without dither and they are definitely different than in T. I'm guessing something is screwed up in T's dithering.

For one thing noise is pllainly audible in rendered silence - if I'm not mistaken dither should NOT be heard in rendered silence as its amplitude is less than the least significant bit. So the implementation of dither in T is defenitely screwed up and this needs to be moved right to the top of the list.

[IIRs]

You are mistaken: dither modulated the LSB and techniques such as noise-shaping exist to make it less audible.

[Lunch Money]

Nobody has ever claimed that T's dither is perfick. Have you tried disabling T's dither and using MDA dither or a different dither instead?

Greg

[gavriloP]

That is definitely audible with headphones and loud volume.

I ran some tests too because I was sure it was about lowering bit-depth(without dithering it usually sounds pretty much like that) but I WAS SHOCKED!

I used sfz and piano soundfont with sfz's best guality resampling setting and after that I used big GlaceVerb preset. I hit single fast low and high notes and everything sounded nice in Tracktion. Also the 24bit TapeIt recording was pristine (wave editors meters showed clearly how it faded into the -130db and below to total silence). I am using Echo MIA 24bit/48kHz. Then I exported the audio from Tracktion with various settings (and yes, the master output volume was at 0db not the default -3db). I exported 16 bit & 24 bit with and without Tracktion's own dithering (from output settings in Tracktion's Setting page) and I also exported it as 32bit.

So the results:
TapeIt sample's fadeout sounded nice and clean(just as it should at 24 bit)it went to total silence smoothly. But then I listened those Tracktion exports and also checked the meters on wave editor(old soundforge). There were troubles in each samples! That same truncation noise on fadeouts beginning after -60db or something like that. It was even more evident with bright high piano notes. Also every sample was cutted into the total silence quite soon after that (about at -75db). The dithering seemed to have practically no effect. 16 and 24 bit versions both sounded pretty much the same which is quite alarming. In 32bit version there was no audible noise but it too went to total silence somewhere at -105db(there might be noise there, but it wont bother anyone, I think).

So what does this mean? It really sounds like somekind of gate effect or bit truncation (which of course are in some sense same thing cause you leave some bits out when you gate and mute that near silent data) It was worrying that both 24 and 16 bit sounded so much alike (in other words, there wasn't any real gain to use 24bits, which there should be ). Also that dither proved to be pretty useless. Normally I can live with this noise if I use some constant sampled background noise or dither with noise shaping but if I want to work with not so hot material with plenty of headroom or boost some silent material there are huge problems! It practically makes the 24bit recording useless for me and this is scary.

I hope there is some solution to this other than TapeIt because even though it works well, there is one major concern with it, I cannot do any songs that won't run in realtime. I have been using NFR Tracktion since December and I really hope this will be fixed.

-Mikko

[Lunch Money]

Don't suppose you read my post?

[Mr. Slater's Parrot]

I can hear the difference, for sure, on my monitors (HR824s). I need to turn the volume up a bit. But the problem is audible, absolutely, with my set up.

[semiquaver]

Platinum - I just rendered some silence and then normalised the result. The noise is not a single bit flipping around - there are at least 4 bits of depth and there is DC offset as well. I am not an expert in these matters and so could be mistaken but this doesn't seem right.

And whether or not it is what you would expect a cursory comparison of very quiet reverb tails will showed a clear difference between DP and T not to Ts advantage.

[pdxindy]

Platinum - I just rendered some silence and then normalised the result. The noise is not a single bit flipping around - there are at least 4 bits of depth and there is DC offset as well. I am not an expert in these matters and so could be mistaken but this doesn't seem right.

And whether or not it is what you would expect a cursory comparison of very quiet reverb tails will showed a clear difference between DP and T not to Ts advantage.

both 24bit no dither and you find a difference? or is just dither different?

[semiquaver]

I did T at 16 and 24 bits with and without dither and DP at 16 bits with and without dither.

Both DP tails sounded more like what I would expect than Ts tails. Likewise just examining very quiet material in T I found a kind of distortion I didn't expect. My guess is that the dithering is not right but who knows.

[semiquaver]

OK did some reading and will return to tests. yes dither should be broadband and not less than the LSB - my bad. The crucial point in this thread it that the renders don't sound like Tapeit. I can't verify this on Mac but it should be looked at.

[gavriloP]

Nobody has ever claimed that T's dither is perfick. Have you tried disabling T's dither and using MDA dither or a different dither instead?

Greg

I think this is not just about the quality of dither. If the noise from 32bit floats (original Tracktion) to 24bit is equal to resampling 32bit to 16bit then there sure is something wrong, or am I missing something here? Of course there is no problem if I "master on the fly" before I export song from Tracktion, but this is normally not the case. Now every process to song after exporting is revealing noise that shouldn't be there, in theory at least.

Mikko

[semiquaver]

as far as I can tell T1 adds dither to every 16-bit render. I have tried rendering silence with the dither unchecked in the settings page on every output and I get noise in the rendered clips.

[semiquaver]

I don't hear dither noise in the 24 bit renders - but the tail is identical either way. MJy guess: perhaps dither is being added to the truncated 16-bit file rather than mixed with the 32-bit file before truncation. Because I don't hear the expected benefit from the dithering.