Wave's WIR Impulse response convertion to WAV

[Burillo]

nor does it make a lot of sense on the face of it given that convolution is a simple process where adding noise to the extent you claim would only occur with a significant programming error

...or simply using lower precision math. 30db of noise floor sounds like 5 bits of resolution (sans dithering, of course). given how long ago the Waves one was coded, it wouldn't be so far out of the ordinary if they "optimized" their algos this way.

[simon.a.billington]

btw- Is Waves IR still limited to 6 sec max IRs length like it was when I did my testing?

Yeah

To be honest, though, that’s a freaking long reverb tail!!

[simon.a.billington]

nor does it make a lot of sense on the face of it given that convolution is a simple process where adding noise to the extent you claim would only occur with a significant programming error

...or simply using lower precision math. 30db of noise floor sounds like 5 bits of resolution (sans dithering, of course). given how long ago the Waves one was coded, it wouldn't be so far out of the ordinary if they "optimized" their algos this way.

The algorithms would be 64-bit now. Waves updates this type of thing behind the scene and trickles it through a Central update.

Although the convolutions themselves will most likely just be 44.1 or 48kHz, quite possibly 16-bit.

[Burillo]

nor does it make a lot of sense on the face of it given that convolution is a simple process where adding noise to the extent you claim would only occur with a significant programming error

...or simply using lower precision math. 30db of noise floor sounds like 5 bits of resolution (sans dithering, of course). given how long ago the Waves one was coded, it wouldn't be so far out of the ordinary if they "optimized" their algos this way.

The algorithms would be 64-bit now. Waves updates this type of thing behind the scene and trickles it through a Central update.

Although the convolutions themselves will most likely just be 44.1 or 48kHz, quite possibly 16-bit.

well, they're 64-bit now, but the person reporting the 30db noise floor difference has said that he tested it way back when. it is therefore entirely possible that back then internal math in Waves IR1 did not use 64-bit math but instead used less precision.

[kvaca]

nor does it make a lot of sense on the face of it given that convolution is a simple process where adding noise to the extent you claim would only occur with a significant programming error

...or simply using lower precision math. 30db of noise floor sounds like 5 bits of resolution (sans dithering, of course). given how long ago the Waves one was coded, it wouldn't be so far out of the ordinary if they "optimized" their algos this way.

The algorithms would be 64-bit now. Waves updates this type of thing behind the scene and trickles it through a Central update.

Although the convolutions themselves will most likely just be 44.1 or 48kHz, quite possibly 16-bit.

well, they're 64-bit now, but the person reporting the 30db noise floor difference has said that he tested it way back when. it is therefore entirely possible that back then internal math in Waves IR1 did not use 64-bit math but instead used less precision.

well, I dont have the latest version of waves plugins so cannot test it, but I doubt they have changed much or anything in code...because the MAIN thing which prevented me to use IR1 was that stupid 6 sec limitation, which is-as confirmed here-still valid /half of my private IRs are longer than 6 sec!/
looking at manual on their site today it says 92-106 db noisefloor which is exactly the same as it was cca 10 years ago...and compared against -144 db of Pristine Space/which is older code!!/ is like comparing CD vs DVDaudio or SACD...

[Burillo]

first of all, does the plugin itself have that noise floor, or do the impulses? the plugin may very well have very low noise floor as far as DSP is concerned, but the impulse may not have noise floor that low. i do have latest Waves IR1, so i can run a few tests today.

second of all, -144db of noise floor sounds like 24-bit audio. -92-106dB likely indicates 16-bit with dithering (-96dB nominal, but you can get down lower if you use dithering). does Waves IR1 support 24-bit files at all? if it doesn't, that would explain higher noise floor.

(by the way, i don't think comparison with DVD-A/SACD is apt here, those things are gimmicks, while processing noise floor is a real issue)

[kvaca]

first of all, does the plugin itself have that noise floor, or do the impulses? the plugin may very well have very low noise floor as far as DSP is concerned, but the impulse may not have noise floor that low. i do have latest Waves IR1, so i can run a few tests today.

second of all, -144db of noise floor sounds like 24-bit audio. -92-106dB likely indicates 16-bit with dithering (-96dB nominal, but you can get down lower if you use dithering). does Waves IR1 support 24-bit files at all? if it doesn't, that would explain higher noise floor.

(by the way, i don't think comparison with DVD-A/SACD is apt here, those things are gimmicks, while processing noise floor is a real issue)

yes,it was plugins noisefloor, as I have used 32bit float audio for all testing then...if you can test it using dirac and high resolution IRs I would be curious for the results

btw-noisefloor of Waves IRs is usually even worse than noisefloor of IR1 plugin, probably because they have used relatively short sweep testtones...but that doesnt mean they are crappy IRs, just a tiny bit noisier...
btw2-IRs own noise after convolutin translate into ugly artificial reverb and do NOT worsen plugins noisefloor...

[Burillo]

i don't think i have any "high resolution" impulses (assuming ones included with IR1 aren't) - do you know where i can get at least one for free?

[kvaca]

i don't think i have any "high resolution" impulses (assuming ones included with IR1 aren't) - do you know where i can get at least one for free?

you can use ANY short HR /32bit/ audio instead of IR /f.i.drumshot etc/, beware: max 6 sec long

after convolution with dirac timealign and normalize both files and use compare feature of your waveeditor of choice/I use Wavelab/, if you dont have any then do phase reverse of one of the files and mix them up - and measure the results against original audio...happy testing

[Burillo]

wouldn't just shooting a sine wave through it reveal the noise floor?

[Burillo]

shooting a sine wave through IR1 reveals what appears to be 24-bit (-144dB) dithered down into pretty much inaudibility (and that's with three impulses active - direct, early and tail - single impulse noise floor is even lower):

IR1-noise.jpg

So, apparently, you're wrong about IR1's high noise floor - it's perfectly fine.

[kvaca]

shooting a sine wave through IR1 reveals what appears to be 24-bit (-144dB) dithered down into pretty much inaudibility (and that's with three impulses active - direct, early and tail - single impulse noise floor is even lower):


IR1-noise.jpg

So, apparently, you're wrong about IR1's high noise floor - it's perfectly fine.

yes, assumed that difference between both files /=residual noise/ is -144 db its OK...
but...from your picture above there is still clearly visible your testtone, which reveals your testing as wrong /or unfinished/...if you really used pure sine tone instead of IRs there should be almost no significant sine tone visible in differential audio /see my testing method above/

sorry for editing this post-Ive overlooked important thing

[Burillo]

shooting a sine wave through IR1 reveals what appears to be 24-bit (-144dB) dithered down into pretty much inaudibility (and that's with three impulses active - direct, early and tail - single impulse noise floor is even lower):


IR1-noise.jpg

So, apparently, you're wrong about IR1's high noise floor - it's perfectly fine.

yes, assumed that difference between both files /=residual noise/ is -144 db its OK...
but...from your picture above there is still clearly visible your testtone, which reveals your testing as wrong /or unfinished/...if you really used pure sine tone instead of IRs there should be almost no significant sine tone visible in differential audio /see my testing method above/

sorry for editing this post-Ive overlooked important thing

i didn't edit anything out, nor does it matter if i do - the noise floor is not going anywhere, as the noise is not part of the signal anyway - it's byproduct of processing. so no, the test is not wrong just because i didn't follow your convoluted test method.

[kvaca]

shooting a sine wave through IR1 reveals what appears to be 24-bit (-144dB) dithered down into pretty much inaudibility (and that's with three impulses active - direct, early and tail - single impulse noise floor is even lower):


IR1-noise.jpg

So, apparently, you're wrong about IR1's high noise floor - it's perfectly fine.

yes, assumed that difference between both files /=residual noise/ is -144 db its OK...
but...from your picture above there is still clearly visible your testtone, which reveals your testing as wrong /or unfinished/...if you really used pure sine tone instead of IRs there should be almost no significant sine tone visible in differential audio /see my testing method above/

sorry for editing this post-Ive overlooked important thing

i didn't edit anything out, nor does it matter if i do - the noise floor is not going anywhere, as the noise is not part of the signal anyway - it's byproduct of processing. so no, the test is not wrong just because i didn't follow your convoluted test method.

I have measured differences /residual noise from fft processing/ but you seem have measured impulses/testones/ own noise floor...

[Burillo]

shooting a sine wave through IR1 reveals what appears to be 24-bit (-144dB) dithered down into pretty much inaudibility (and that's with three impulses active - direct, early and tail - single impulse noise floor is even lower):


IR1-noise.jpg

So, apparently, you're wrong about IR1's high noise floor - it's perfectly fine.

yes, assumed that difference between both files /=residual noise/ is -144 db its OK...
but...from your picture above there is still clearly visible your testtone, which reveals your testing as wrong /or unfinished/...if you really used pure sine tone instead of IRs there should be almost no significant sine tone visible in differential audio /see my testing method above/

sorry for editing this post-Ive overlooked important thing

i didn't edit anything out, nor does it matter if i do - the noise floor is not going anywhere, as the noise is not part of the signal anyway - it's byproduct of processing. so no, the test is not wrong just because i didn't follow your convoluted test method.

I have measured differences /residual noise from fft processing/ but you seem have measured impulses/testones/ own noise floor...

First of all, test tone was generated inside REAPER by a JS tone generator, and did not have any detectable noise floor whatsoever (i checked - complete silence down to -180dB).

Second of all, if the "residual noise from fft processing" was as high as you initially claimed, it wouldn't have mattered if my analysis is impulse's own noise, or it's a byproduct of processing - the noise level would have still be higher than it is.

Finally, you yourself have (correctly) pointed out that noise in the impulse itself would not have manifested as noise in the signal, because the impulse response only filters, it doesn't add anything. In ither words, what we see on the graph is residual noise from processing, not noise from impulse response.

So, no, the result still holds true - i can't know anything about IR1 from a number of years ago, but right now IR1 gives out a perfectly clean signal down to dithered 24-bit noise floor.