In a bid to try and end the question of "Does Atomix have poor sound quality as compared to XXXX software?", I've done a analysis of a simple frequency sweep. Be aware though that I am in NO way bashing Atomix or trying to put it down in any way. I love Atomix, I love to use Atomix, and the creators have put a great deal of hard work behind the programming of Atomix to make it the best DJ software available. Programming software is not as simple as A + B = C, so before anyone tries to use my analysis to bring down Atomix, lets see YOUR programming skills crank out something better.
Now, with that said, onto my analysis. Goto http://roadriders.no-ip.org/4images
There will be a couple of categories listed near the top. The first I want to deal with is the "Reference Audio" category.
For the base wave file, I used a TDK Sound Reference Check wav ripped directly from the cd using EAC (Exact Audio Copy) with 0 errors. Its a standard, slow frequency sweep from 20Hz to 20kHz. The frequency analysis was done in Cool Edit useing a Triangular FFT (for more a exact graph) using both a Linear and Non-Linear view. The linear view shows the frequency range from 0Hz to 22kHz, divided evenly every 1kHz. The non-linear view shows mainly from 0Hz to 14,482Hz, with the remaining Hz up to 22kHz compressed within the final bars. The graph is not divided evenly, but shows abit more detail of the mid to lower frequency range.
The wav was then compressed using Lame version 3.92MMX codec, at 320kbps, joint stereo, and a Q level of 0. The exact command line used was "-b 320 -m j -h -q 0". The mp3 file used for the remaining tests is available at http://roadriders.no-ip.org/Reference_Audio_Frequency_Sweep.mp3
The mp3 file was then played back using Winamp with the 2.8a Fraunhofer based codec and the DiskWriter 2.0c output plug-in. The MAD Plug-in was also used with the DiskWriter plug-in. The same Lame codec was used to decode the mp3 it created. Atomix was set to software mixing and was used at 100% pitch, and then 102% pitch to playback the mp3 file and record. There was NO analog recording involved in any of the test. All files were read and directly written to wav, so there was no digital to analog to digital loss involved in any of the test.
The test was simple. Each codec would write out a wav file and it would be analysed in Cool Edit using both Linear and Non-Linear views. Each codec has pictures of the results of the analysis, along with a quick comparison to the Reference wav file. The pictures speak for themselves.
Lame provides almost an equal output to the original wav file. I'd have to say about 99% faithful to the orignal. Winamp with the Fraunhofer codec is a very close 2nd, with the MAD plug-in is a close 3rd. But alas, Atomix is a very distant 4th.
If you notice in the comparison of the 100% pitch, the resulting Atomix audio loses some db's across almost the whole frequency range. Some areas lost 3db! And forget about the 102% pitch comparison. It almost doesn't resemble the original file at all.
So there you have it. Straight and to the point. If your ears can't tell, well now your eyes can. Enjoy.
Grimm
PS - Its a slow connection, so if alot of ppl hit my site at once, it will bring the site down to a crawl.
Now, with that said, onto my analysis. Goto http://roadriders.no-ip.org/4images
There will be a couple of categories listed near the top. The first I want to deal with is the "Reference Audio" category.
For the base wave file, I used a TDK Sound Reference Check wav ripped directly from the cd using EAC (Exact Audio Copy) with 0 errors. Its a standard, slow frequency sweep from 20Hz to 20kHz. The frequency analysis was done in Cool Edit useing a Triangular FFT (for more a exact graph) using both a Linear and Non-Linear view. The linear view shows the frequency range from 0Hz to 22kHz, divided evenly every 1kHz. The non-linear view shows mainly from 0Hz to 14,482Hz, with the remaining Hz up to 22kHz compressed within the final bars. The graph is not divided evenly, but shows abit more detail of the mid to lower frequency range.
The wav was then compressed using Lame version 3.92MMX codec, at 320kbps, joint stereo, and a Q level of 0. The exact command line used was "-b 320 -m j -h -q 0". The mp3 file used for the remaining tests is available at http://roadriders.no-ip.org/Reference_Audio_Frequency_Sweep.mp3
The mp3 file was then played back using Winamp with the 2.8a Fraunhofer based codec and the DiskWriter 2.0c output plug-in. The MAD Plug-in was also used with the DiskWriter plug-in. The same Lame codec was used to decode the mp3 it created. Atomix was set to software mixing and was used at 100% pitch, and then 102% pitch to playback the mp3 file and record. There was NO analog recording involved in any of the test. All files were read and directly written to wav, so there was no digital to analog to digital loss involved in any of the test.
The test was simple. Each codec would write out a wav file and it would be analysed in Cool Edit using both Linear and Non-Linear views. Each codec has pictures of the results of the analysis, along with a quick comparison to the Reference wav file. The pictures speak for themselves.
Lame provides almost an equal output to the original wav file. I'd have to say about 99% faithful to the orignal. Winamp with the Fraunhofer codec is a very close 2nd, with the MAD plug-in is a close 3rd. But alas, Atomix is a very distant 4th.
If you notice in the comparison of the 100% pitch, the resulting Atomix audio loses some db's across almost the whole frequency range. Some areas lost 3db! And forget about the 102% pitch comparison. It almost doesn't resemble the original file at all.
So there you have it. Straight and to the point. If your ears can't tell, well now your eyes can. Enjoy.
Grimm
PS - Its a slow connection, so if alot of ppl hit my site at once, it will bring the site down to a crawl.
Posté Sun 05 Jan 03 @ 3:54 am
Grimm
Damn site logged me out before I hit the post button. >:|
But the analysis was done by me.
Grimm
But the analysis was done by me.
Grimm
Posté Sun 05 Jan 03 @ 3:56 am
abnorm
Thanx for that great work!
Posté Sun 05 Jan 03 @ 4:05 am
ikke
I have to agree, the work you did is great. Where did you learn all this stuff?!?
On the other side I have to disagree on some parts of your conclusions, which is just my idea, and maybe I'm completely wrong.
I think the comparison between Atomix playback and WinAmp isn't completely fair. Winamp just plays back the original signal (and that's its task), but Atomix also has to do a lot of calculations (the DSP-side of the work) too. You can't use numbers on a pc with infinite numbers behind the ',' , so everytime you do a calculation on a given signal, even if it's a very basic operation (and most DSP-related calculations aren't really basic :) they are, like every formula, a bunch of lots of basic calculations), but you still loose information with every * or / you do.
And: Atomix needs to do all its operations in "real-time" (well, not in realtime, but +-). So you can't use the best calculations aviable (with as less loose of information as possible), because those are too slow.
I don't know whether I'm completely right (I don't know anything about the inside of Atomix, I only can guess :)), but that's just the way I see it.
Altough: great test, Grimm!!! It's not really a sinecure to do any soundcheck :)
Greetz, Ikke
PS. Once again I don't want to offence anyone in this post, or minimize anyones work :)
On the other side I have to disagree on some parts of your conclusions, which is just my idea, and maybe I'm completely wrong.
I think the comparison between Atomix playback and WinAmp isn't completely fair. Winamp just plays back the original signal (and that's its task), but Atomix also has to do a lot of calculations (the DSP-side of the work) too. You can't use numbers on a pc with infinite numbers behind the ',' , so everytime you do a calculation on a given signal, even if it's a very basic operation (and most DSP-related calculations aren't really basic :) they are, like every formula, a bunch of lots of basic calculations), but you still loose information with every * or / you do.
And: Atomix needs to do all its operations in "real-time" (well, not in realtime, but +-). So you can't use the best calculations aviable (with as less loose of information as possible), because those are too slow.
I don't know whether I'm completely right (I don't know anything about the inside of Atomix, I only can guess :)), but that's just the way I see it.
Altough: great test, Grimm!!! It's not really a sinecure to do any soundcheck :)
Greetz, Ikke
PS. Once again I don't want to offence anyone in this post, or minimize anyones work :)
Posté Sun 05 Jan 03 @ 11:29 am
flex
Nice work dude, this has been a long time in coming and hopefully some changes in opinions and the software will be made now.
Flex
Flex
Posté Sun 05 Jan 03 @ 2:11 pm
Phenryll
good work :)
Posté Sun 05 Jan 03 @ 2:36 pm
Dj T-Bone
indeed, very nice!
hopefully teamers will start with the pitch algorythm ...
this degrades sound quality the most ... ;)
i've noticed some dj progs use FMod for playing the audio maybe you could test that engine ...
hopefully teamers will start with the pitch algorythm ...
this degrades sound quality the most ... ;)
i've noticed some dj progs use FMod for playing the audio maybe you could test that engine ...
Posté Sun 05 Jan 03 @ 4:06 pm
ikke
I worked with FMod for some time, but I don't like it. You're too restricted. I think you refer to that "Kramixer" prog (which is in fact nothing more but a gui for fmod).
But why not :) taking a look at it isn't a murder :) I don't really like xaudio either :D
Greetz, Ikke
But why not :) taking a look at it isn't a murder :) I don't really like xaudio either :D
Greetz, Ikke
Posté Sun 05 Jan 03 @ 6:09 pm
Dj T-Bone
dj decks uses it too i believe ...
mp3tunes also ...
and sound quality isn't that bad althought in kramixer you also have that hissing pich sound ... and in mp3tunes not
and it's twice Fmod ... :/
mp3tunes also ...
and sound quality isn't that bad althought in kramixer you also have that hissing pich sound ... and in mp3tunes not
and it's twice Fmod ... :/
Posté Sun 05 Jan 03 @ 7:35 pm
ikke
It's not only the sound quality which counts, but also the useability, isn't it?
Just my .02, but we're getting OT :)
Greetz, Ikke
Just my .02, but we're getting OT :)
Greetz, Ikke
Posté Sun 05 Jan 03 @ 8:29 pm
Grimm
ikke - I thought about the alterations that the Atomix code does to the audio output as well. I ran all the test with no Effects of any kind running. The gain was left where it laid, as was everything else. The only change I made was to the pitch. One might expect a small change in audio quality when something is changed. But when all is level and normal, the audio sould not suffer and loss in quality.
And yes, I too believe that its not only the sound quality that counts, but the useability as well. But at least for now the question of poor audio quality from Atomix can be laid to rest for a while.
Grimm
And yes, I too believe that its not only the sound quality that counts, but the useability as well. But at least for now the question of poor audio quality from Atomix can be laid to rest for a while.
Grimm
Posté Sun 05 Jan 03 @ 9:38 pm
Grimm
PS. Hopefully the programmers will take this into account and dedicate themselves to improving the sound quality for the next version. I know many will complain about not getting XXX effect, or XXX feature. But I hope most of us will agree that audio quality sould be thier #1 priority, not making Atomix work and look more like your favorite deck which already has the quality to begin with.
Grimm
Grimm
Posté Sun 05 Jan 03 @ 9:42 pm
Thibaud Dafaux
totally agree
GnoedeL
GnoedeL
Posté Mon 06 Jan 03 @ 6:42 pm
ikke
Of course you're right. With useability, I wasn't talking about the useability of Atomix itself, but the Audio reader classes, like FMod :)
Greetz, Ikke
PS. Personally I don't hear any bad sound, but that's just because I never used Atomix with a real amplifier. My JBLs on my desktop perform great. But that's just my point of view, and its a lucky were not all equal :)
Greetz, Ikke
PS. Personally I don't hear any bad sound, but that's just because I never used Atomix with a real amplifier. My JBLs on my desktop perform great. But that's just my point of view, and its a lucky were not all equal :)
Posté Mon 06 Jan 03 @ 8:28 pm
Damocles
Was the master tempo on ? Would be interesting to see the difference between master tempo on and off.
Posté Tue 07 Jan 03 @ 4:39 pm
Grimm
Master tempo was off. I would expect the results to be similiar to the 100% pitch setting if master tempo was on. I'll run a quick test and post the results.
Grimm
Grimm
Posté Tue 07 Jan 03 @ 11:30 pm
Grimm
Ok. The pics are up at the site. As expected, the 100% pitch with Time Stretch On gave similiar results with Time Stretch off, so the picture was not uploaded.
However, the results with 102% pitch and Time Stretch on were different then what I expected. With Time Stretch on, the output audio didn't wildly add in noise as with Time Stretch off. The cutoff at around 20kHz is still there, unlike the output with Time Stretch off that had added in noise all the way up to the end of the graph. However, there is still alot of loss and gain through most of the frequency range.
Grimm
However, the results with 102% pitch and Time Stretch on were different then what I expected. With Time Stretch on, the output audio didn't wildly add in noise as with Time Stretch off. The cutoff at around 20kHz is still there, unlike the output with Time Stretch off that had added in noise all the way up to the end of the graph. However, there is still alot of loss and gain through most of the frequency range.
Grimm
Posté Wed 08 Jan 03 @ 12:10 am
Damocles
Interesting. I would have expected the sound quality to be worse with Time Stretch on. Seems like it's the opposite ! Thanks again for performing those tests !
Posté Wed 08 Jan 03 @ 11:21 am
Grimm
Well, it depends on how you look at it.
The basic way that Time Stretching works, is by overlapping or dropping samples at regular intervals. Say you have a song 100 seconds long and want to perform a Time Stretch on it of around 10%. In overlapping, the program would play 1 to 9 seconds, and then overlap seconds 10 and 11, then play 12 to 19 seconds, then overlap seconds 20 and 21, etc etc. Some programs might actually just cut out a second, so it would go 1 to 10, then 12 to 20, 22 to 30, etc. Some programs do a mix of both dropping samples and mixing. If done properly, Time Stretching can sound near perfect, as if the audio was originally done at that tempo. Done poorly, and the audio can sound like crap.
To date, Atomix does some pretty damn good Time Stretching. I've yet to find ANY other program that can Time Stretch in *real time* as good as Atomix. Even many audio editing programs that dont do it in real time sound as good as Atomix, except for TimeFactory. TimeFactory for me gives near perfect time stretched audio. Nearly no hiccups or missed beats.
But, Atomix audio output still suffers abit when Time Stretching, although as you have noticed, not anywhere near as much as without Time Stretching.
Grimm
The basic way that Time Stretching works, is by overlapping or dropping samples at regular intervals. Say you have a song 100 seconds long and want to perform a Time Stretch on it of around 10%. In overlapping, the program would play 1 to 9 seconds, and then overlap seconds 10 and 11, then play 12 to 19 seconds, then overlap seconds 20 and 21, etc etc. Some programs might actually just cut out a second, so it would go 1 to 10, then 12 to 20, 22 to 30, etc. Some programs do a mix of both dropping samples and mixing. If done properly, Time Stretching can sound near perfect, as if the audio was originally done at that tempo. Done poorly, and the audio can sound like crap.
To date, Atomix does some pretty damn good Time Stretching. I've yet to find ANY other program that can Time Stretch in *real time* as good as Atomix. Even many audio editing programs that dont do it in real time sound as good as Atomix, except for TimeFactory. TimeFactory for me gives near perfect time stretched audio. Nearly no hiccups or missed beats.
But, Atomix audio output still suffers abit when Time Stretching, although as you have noticed, not anywhere near as much as without Time Stretching.
Grimm
Posté Wed 08 Jan 03 @ 12:10 pm
trip_likeido
The time stretch off produces that hiss ... the time stretch enabled sounds gargled and produces those hiccups and lost bass on the occasional beat.
I prefer disabling it. I find that studio programs such as Sound Forge and Acid do the best pitch shifting and time stretching.
I prefer disabling it. I find that studio programs such as Sound Forge and Acid do the best pitch shifting and time stretching.
Posté Wed 08 Jan 03 @ 4:16 pm
