Fellow sound developer Dan Dean and myself decided to tested the speed of Acoustic Mirror on different Pentium processors. Dan has a P4 2.5GHz machine and I have a P3 666MHz . For the test we decided to use the Myerson Concert Hall (Bruce Richardson developed this impulse for Sonic Foundry) and the Beatles song \"Lucy In The Sky With Diamonds\". It took 47.4 second on the my P3 and 20.9 seconds on Dan\'s P4. The P4 2.5GHz turns out to be 2.26 times faster even though the clock is 3.75 time faster and the processor is a next generation CPU.
I used the same reverb impulse #2900 (from the Pure Space Film and Sacred Reverberation Impulse CD) which is 11.7 seconds long on a Mozart Overture (261 seconds long). All the sound files on the Mac and PC were converted to 32 bit resolution. On a 867MHz Dual Processor Power PC G4 running OS X the convolution process in Bias\'s Peak took 183 (70% of real time length of the sound file RTLSF)seconds and SoundHack took 193 seconds (74% RTLSF) the 666MHz Pentium III running Acoustic Mirror took 68 seconds (26% RTLSF) by extrapolation a P4 2.5GHz machine should take 30 seconds or (11.5% RTLSF) I guess I will have to upgrade soon in order to find out !
At this point in time I do not know if the sound files are identical however they all sound great.
Maybe you should try this with Altiverb in both Peak (as a VST bounce) and Spark? (Logic 6 offline bouncing might be interesting too?)
As far as I\'m aware, both Peak and Soundhack\'s convolution are not optimized for realtime use, so their calculations might not be that efficient.
However, Acoustic Mirror does run in realtime and might therefore use more efficient calculations (or better shortcuts! [img]images/icons/tongue.gif[/img] )
There\'s no OSX version of Altiverb yet, so running it on a dual machine won\'t likely improve it\'s speed.
Altiverb will run in real time with 24bit resolution (an amazing computational feat) however it cannot do 32bit floating point calculations. For creative sound sculpting floating point definitely gives you the highest signal to noise ratio. In a sense all the programs are real time because they process a 4 minute sound file in less than 4 minutes.
What do you mean when you say that Acoustic Mirror is real time ?
I do not have a full version of Altiverb yet (only the demo version) so I cannot fully test this program.
That\'s an interesting observation, I hadn\'t looked at it that way (the convolutors being in a sense realtime).
I didn\'t know that Altiverb was 24 bit internally..I figured that it was 32 bit since that is what all the VST hosts (Logic, Cubase)internal bit depth is, as far as I\'m aware at least; I do know that the IR limit is 24 bit for Altiverb.
As far as I know Acoustic mirror runs in realtime when you run it as an fx in SF6, at least with short impulses and the quality set up to 5, dependent on your cpu speed you might be able to run longer IR\'s in realtime.
I think programs such as vegas or acid can run it as a realtime plugin as well (like altiverb).
I heard that Nuendo has a convolution plug in as well...no idea how well that works though.
Bruce pointed out that AMirror, is a hog with buffers, so in some applications it wotn run in realtime. Par tof this is probably because it is DirectX. I can get it to run in realtime in Vegas only. Sound forge only in \"preview\" mode and it tends to stop whenever it wants.
There\'s apparently a new VST Impulse response app, with some latency, as well it only uses 16 bit files.
In general 32 bit floating pont in probably better for one pass processing, due to dithering issues.
but when it comes to impulses, its more than probably if you\'re dealing with the same algorithm.
Impulses and impulse response effects deal with signal to noise ratios more than anything else, more than reverbs in most cases, since the s/n ratio of the actual impulse response adds to the s/n ratio of how the effect works.
low level ambience characteristics are key to the way an impulse works if I\'m not mistaken. Especially since its linear.
When dynamic impulse response works with long inpulse responses it will make an extreme difference.
..... atleast extreme depending on what you\'re listening too. If it where the same recording and same impulse at different bit depths I\'d bet it would make a difference
now which would be better....... might depend on ears, which is more representative of actual sound or characteristic of the cue/mix/effected wave file,...of which \"better\" could probably be easily determined by mathematics......whichs ultimately leads to \"which is better\" heard or mathematically, and with sound depending on an actual recorded/created source.....its your ears that matter.
with emphaisis on \"your\"
I think we are beyond the \"which bit depth\" is better argument.
All Ernest is doing is trying to keep the comparison\'s in their \"best\" state mathematically, for consistany\'s sake. I mean, if one does 32 bit floating point, and he doesn\'t use it, once could always argue it would be better at 32-bit FP......
which is obviously not the main point, since the algorithm is key here, but no one will think that first..... I mean Altiverb is obviously better than A Mirror big [img]images/icons/rolleyes.gif[/img] added there [img]images/icons/smile.gif[/img] since I believe no one has ever put a single impulse against a single impulse here.
I\'ve found some problems with Acoustic Mirror, once of which may be its impulse creation set up, or \"generic\" impulse creation in general, and Ernest may have found a way around it
and I for one am encouraging him to do more, and hope he can get all the impulse response/effects guys to give him tools to do his tests, since I think he\'s one of the guru\'s at this stuff.
In fact he\'s probably too involved in \"sound\" for half of us here, including me (which obviously isn\'t a stretch)