AZID - Cracks in the output by using dynamic compression

[phynix]

hi!
i have a little problem with the dynamic compression-method of azid. when it's activated, the encoded files contain cracks - not many, ~5 in the whole file i would say, but when i encode it without this option, the output is just perfect. i have "auto find maximum gain" activated, too. what do i do wrong?? if you can help me, i would be pleased to hear from you!

p.s.: excuse my bad english :rolleyes:

[MaTTeR]

Hmm...I've always used Azids Dynamic Compression and never heard any pops. We really need more info from you before we can help.

What GUI are you using for the encode process? DD's GUI? Commandline?

Your encoding to MP3? If so, what version of LAME? What LAME settings are you using?

Need more info! :-)

[phynix]

ok - here is the full story:

first i rip the vob-files with smartripper and extract the ac3-stream with vstrip -> with only the demux-function activated and only one ac3-substream selected. then i start dds azid/lame GUI [v 0.3] and encode the ac3-stream into wav-format [azid v1.7.1]. the only activated functions are: LFE to LR channels: -3db; dynamic compression: normal; auto find maximum gain; print current settings/Bitstream information/Downmix matrix; Quit on Bitstream Error. i later encode the file to mp3, but the errors occure in the wav-file too, so it can't be lame that causes the problem, also it can't be vstrip, because when i encode the ac3-file without the use of the dynamic compression, there aren't any errors in the output.

i hope this info is enough, if you need more, plz ask me again :) !

[DSPguru]

Originally posted by phynix
auto find maximum gain

try setting a manual gain value.

[DarkAvenger]

This bug has already been reported to Midas some weeks ago. Unfortunately he didn't fix it yet.

[ChristianHJW]

I never had probs like this, maybe because i output into a 24 bit format from Azid ??? Or maybe i'm already deaf ....

BTW : Is Azid's 24 Bit output floating point ? Can LAME read 24 bit floating point ( it certainly can read Azid/SSRC 24 bit output ) ?

Really looking forward to DSP Gurus 32 bit floating point implementation from Azid to Lame. I just hope he doesnt forget about SSRC, for obvious reasons ;) ....

Not that i would believe 32 bits resolution are necessary at all .... but if you dont have to deal with intermediate WAV files, thanks to BeSweet, you dont have to care about file size at all .... so i fully agree with DSPGuru's point of view :

Why not using 32 bit ??? Quality freak as i am, using 32 bit handling for resampling and/or normalizing gives me this undescribable feeling that the absolute optimum was achieved ..... how wonderful this is ;) .... and pls. LigH, dont be cruel now and remind me about the brutal things that are done in most studios ...

[phynix]

i tried to set a gain-level manually, but there were exactly the same errors as with auto find-function. i have varied almost all settings in the used programms, but the result didn't get better. hopefully midas will fix this bug soon...

thx for your help!

[phynix]

hmm... when i select 24 or 32 bit output, wmp can't read the file and when i encode it to mp3, the file contains very loud murmur.

[MaTTeR]

Originally posted by phynix
hmm... when i select 24 or 32 bit output, wmp can't read the file and when i encode it to mp3, the file contains very loud murmur.

I use 24bit output with 2 passes to find the gain. I then encode it with Lame without any issues. Don't use WiMP, use another player for your audio/video needs. I'm pretty sure LAME can't except 32bit WAV files so stick with 24bit if you have the space :-)

[phynix]

Do you mean, you encode it with azid to a 24bit-floating-point-wav an then perform the 2-pass processing with SSRC?? i didn't use SSRC by now, i will try that!

by the way, what is WaveBooster for?? doesn't it do something very similar to dynamic compression in azid??

[MaTTeR]

Originally posted by phynix
Do you mean, you encode it with azid to a 24bit-floating-point-wav an then perform the 2-pass processing with SSRC?? i didn't use SSRC by now, i will try that!

by the way, what is WaveBooster for?? doesn't it do something very similar to dynamic compression in azid??

Thats correct. Use DD's GUI, think the latest is beta 4. You might also try HeadAC3he 0.15. I've been using it more freqently as the speed and output quality is great.

I've never used WaveBooster so I can't really help you there.

[phynix]

i don't understand it, a bitrate higher than 16 just don't work :mad:! it is not only, that wmp can't play it, but the output is at very bad quality - as i said - very loud murmur and the original sound of the input is hardly recognizable, like recorded in a snowstorm. i have resampled the file to 16bit, but the quality didn't become better... this question is probably stupid, but do i need a special codec for floating-point wavs??

[ChristianHJW]

... what OS are you using ?? Its in fact possible that your OS doesnt support 24 bit floating point WAVs to be stored on the HDD.

Try latest BeSweet and BeSweet GUI, according to DSPGuru it will process sound internally in 32 bit mode ( highest possible quality, even a bit exagerated maybe, but why not ) and doesnt create intermediate WAV files, the blocks are fed from Azid directly to Lame or SSRC/Lame ...

@Midas :

If you're still around Mate, we know you didnt have the time to look into the problem of overflow errors for the auto find max gain feature recently, but wouldn it be possible to simply reduce the amplification factor for 2nd pass by lets say 3 dB ( factor found in 1st pass divided by 1.414 ) if 24 bits output was chosen in Azid ?? This would remove the problems instantaneously and with 24 bit output we're still on the safe side in terms of SNR ....

@DSPGuru :

Do you have access to Azid code ??

[phynix]

im using winXP pro. the methode you described sounds interresting - i'll try it.

thx for the hint!

[phynix]

the result is the same - when i open the mp3 with winamp, it tries to write a wav-file to c:, the lenght of it is ~3x as long as the input-file, there isn't any sound and the file is "played" fast forward. i tried to encode the ac3 to a 32 bit wav [with dyn compression] resampled it to a 16 bit wav [2-pass] and let lame encode it to a vbr mp3.

i don't know what's still wrong - perhaps winXp is such a os that doesn't support it...?

[MaTTeR]

Originally posted by phynix
the result is the same - when i open the mp3 with winamp, it tries to write a wav-file to c:, the lenght of it is ~3x as long as the input-file, there isn't any sound and the file is "played" fast forward. i tried to encode the ac3 to a 32 bit wav [with dyn compression] resampled it to a 16 bit wav [2-pass] and let lame encode it to a vbr mp3.

i don't know what's still wrong - perhaps winXp is such a os that doesn't support it...?

I use XP just like others on here and it works fine. Your WinAmp settings are screwed up. Somehow you have it setup to decode an MP3 to a WAV file on your hard disk. Go to the WinAmp prefs and select the output and then make sure that DiskWriter isn't highlighted. You should have it set to Wave Out.

[phynix]

:) thx, i think i got it finally!
your right, it was winamp that caused that problem with the >16 bit wav-playback.

just one more question, then i stop bugging you: could you tell me, if these setting are allright??

DDs BeSweet GUI [0.4 b]:

Azid [1.7.1]
++++++++++++

LFE 2 LR: -3 db
dynamic compression: normal
auto find max. gain
file type: 32 bits floating-point wav

SSRC [1.28]
+++++++++++

Sampling rate to: 44100hz
Perform 2 pass processing

Lame [3.89 b]
+++++++++++++

Mode: Joint Stereo
VBR: Old Routine; qual.:8; disable writing xing tag; minimum allowed br: 0; max. allowed br: 190

[MaTTeR]

Your welcome. Glad your up and running now.

Your settings look good for BeSweet but..I'm not so sure about the LAME/MP3 encode settings your using. Maybe someone else can advise you on these but they look low quality to me. Do a listening test and see what you think.

[ChristianHJW]

... matter is correct, V8 in VBR will result in very low bitrate files ( about 100 kbps ), R3MIX experts recommend standard ABR with default ATH instead of VBR for these low bitrates ...

[DSPguru]

Originally posted by ChristianHJW
@DSPGuru :

Do you have access to Azid code ??
i wish i had..

[LigH]

Sorry for being late :cool:

I also know these occasional cracks in the audio - a few times over the whole movie, best hearable of course where the surrounding sould is weak. I would guess anything... but I would prefer to look at those cracks - if it is only a little step from one sample to the next (the it could be a jump in the internal dynamic compression factors), or several crazy samples (then I would expect a decoding error).

The same problem I found with CoolEdit Pro: Read a high-resolution wave, save it, read it again... and you see many cracks in the audio. I'm not sure where they come from. I don't trust CEP anymore, I'd like to write an analysing program, a visualisation - but I can't find a good 2D graph Delphi component (or I'm too dumb to install them properly - they always crash).

To the question about the required resolution: IEEE single precision float values always have valid 24 bits, the so called "mantisse", no matter how near they are around 0. / 24 and 32 bit integer waves have at maximum so many valid bits - at their extremes!

Where a 24 bit integer wave has only e.g. 4 bits in very weak scenes, a 32 bit integer wave has 12 bits, a 16 bit integer wave would have 0's only - a float wave would have all 24 bit of its mantisse available. But tell me - can you really hear such weak sounds, below 96 dB? I doubt that! :D

[phynix]

i don't know at wich db these cracks occur, but they are on places where the audio is pretty quiet, not on very loud scenes - i mean i have heard, that they only refer on one channel, left or right.

@ matter & ChristianHJW: the quality is mostly between 96 and 128 - but thats ok, because i always do 1-cd-rips and i think its better to have a good picture and instead a worse audio, than to rip a high-qual mp3 and having a even lower picture-quality that 1-cd-rips have anyway.

[MaTTeR]

Originally posted by phynix

@ matter & ChristianHJW: the quality is mostly between 96 and 128 - but thats ok, because i always do 1-cd-rips and i think its better to have a good picture and instead a worse audio, than to rip a high-qual mp3 and having a even lower picture-quality that 1-cd-rips have anyway.

If that's the quality your looking for then you should be ready to RIP. 128k should be ok depending on the movie track, never used anything below that though. Definitely look into using ABR instead of VBR, I think you'll like the results. I have to say this though, please don't post these flicks all over the net if they have poor audio quality for the rest of us :-) Nothing worse than me spending a day to download a movie only to find someone skimped on the quality. Cheers!

[DSPguru]

Originally posted by LigH
But tell me - can you really hear such weak sounds, below 96 dB? I doubt that! :D

and does dithering sounds moRe logical to you than just using the original dynamic range ?

[ChristianHJW]

Originally posted by phynix
@ matter & ChristianHJW: the quality is mostly between 96 and 128 - but thats ok, because i always do 1-cd-rips and i think its better to have a good picture and instead a worse audio, than to rip a high-qual mp3 and having a even lower picture-quality that 1-cd-rips have anyway

These bitrates are o.k., i am using the same range when doing 1 CD rips ... what matter and myself were trying to tell you, dont use VBR, go for ABR ....

[ChristianHJW]

Originally posted by DSPguru
and does dithering sounds moRe logical to you than just using the original dynamic range ?

... finally i am starting to believe my understanding of what dithering does is correct ... trying to avoid empty blocks ( 0000 ) because otherwise the resampling process gets f...

LigH, you have to admit there is absolutely no argument against using 32 bit floating point if there is no intermediate file to be created .... who cares if Azid is handling a 32, 24 or 16 bit word to SSRC and/or lame, as long as the output is a very high quality 16 Bit MP3 or MP2 !

Using 32 bit seems to be over exagerated, i agree that 24 bit floating point is much better than human ear can do, but why not using 32 bit and being absolutely sure that even under bad circumstances the absolute optimum was achieved ....

[LigH]

Dear ChristianHJW,

I just agree with your opinion: During a process, the best possible quality should be used, only the final "mixdown" can then be reduced to the "human recognition range". This is true for audio arranging (why else do people like e.g. Cubase VST/32, why did Steinberg introduce "TrueTape" saturation simulation?) and it will be true for related but simpler conversions.

The AZID.DLL exports float values by decoding, therefore they should be used as long as possible, until you have the final waveform which you want to compress e.g. to MP3, to maintain the maximum quality during the optimisation process. (Remember where it started? Some people - like me - were wondering if it were possible to make sounds louder while avoiding digitizing noise of amplified 16 bit waves...)

And dithering - of course, why not! It is just simple: Those too less magnificant bits which are cut off while converting to integer are used to describe the probability of the new least significant bit(s) to be set - voila, triangular dithering. (For the case anyone remembers: Making the "PC beeper" speak or sing "Mushroom", this was completely done with dithering - the piezo speaker could only be set to on or off, but switching fast enough allowed the membrane to float between these two states. You can still find source code on my page.)

[DarkAvenger]

Anyone still knows Turrican or Katakis? The digitized sounds on the C64 were played the same way..

[phynix]

@ matter & ChristianHJW: hmm, most of the divx-rips around have a similar quality, but i'm going to test the abr-methode. if the filesize is not much higher than that of the vbr, i will change to it.

@ matter: you can calm down, i think you wouldn't download german divx-moviez anyway ;).

@ all: thx again, you really helped me with this :)!