3ll3d00d

Wonder Woman - corrected vs pink noise - https://imgur.com/a/KptYn

Wonder Woman (2017, BD / Atmos) - JRiver equivalent filters

NB: I refer to it as S below to differentiate from the original post, jriver calls this Q though.

measurement to verify shape -> https://imgur.com/a/05Ixe

UHF / distance (relative to flat response; see note*):

High Shelf: f0 = 13900.0; S = 0.442082; gain = -3.501

HF slope (see note **):

High Shelf: f0 = 3140; S = 0.9997; gain = -2.0
High Shelf: f0 = 7750; S = 0.552885; gain = -2.0
EQ Peak: f0 = 1950; Q = 1.75; gain = +0.33

MF shelf:

Low Shelf: f0 = 1000; S = 0.720521; gain = -1.25

LF shelf:

Low Shelf: f0 = 265; S = 0.785708; gain = -4.0

Nuttall

I made the  image a bit bigger and included pink noise spectrum as a reference curve for comparison - https://imgur.com/a/yxK3V

If there is some other reference curve that would be better then let me know & I can update it. I'll get to the after pic tonight probably.

This is the before C channel for Wonder Woman, I ran it through speclab with no decimation and 4096 fft length to give ~11Hz frequency resolution

https://imgur.com/gIHjYdF

OK, I'll pull that out for you.

I do know how convert to jriver format filters. Are there any particular scenes (or segments of the film) I could use for this?

I have wonder woman so can extract this for you with and without your correction on top if you like. Just post what you want to see and I will get to it over the next few days.

have you tried Capt America Civil War? if not, I will give it a go.

The LFE channel is a straight line rolloff on a linear scale on both avg and peak from 30Hz down so I guess that's a good sign (for a track that should yield some useful content)? 

https://imgur.com/87rzdnR

On 08/11/2017 at 9:15 PM, SME said:

I am also thinking a bit about automation here.  I think the tricky thing about using something like MSO to find filters is that we want to only make coarse shape adjustments.  The finer peaks and dips in the peak and/or average graphs should be retained because they arise from specific effects that are meant to have characteristic sounds.

That still may not be the most time consuming part though.  That might be the headroom check.  

The reason I suggested using MSO is because you have complete control over the parameter range for each filter and also complete control over how many filters, and what type, it can use. Combine this with smoothing of the response if necessary and I would have thought you'd end up with a smooth filter shape.

re headroom, is it the summed result that is the concern? 

On 09/11/2017 at 4:13 AM, maxmercy said:

Anywhere from 4-12 hours per film, with about 1/2 that time just waiting for processing to happen as I think about how to do the next trial solution. 

that seems a long time to wait for processing, what are you using for that?

Removing the drivers will help alot IME as the weight will be more evenly distributed which makes them much less of a bear to manage (and also means the driver is safe if you do have an accident with the box)

If the approach is based on a target curve which is based on other known good bass films then t seems like this should be amenable to automation, at least as a first cut anyway.

For example, create a minimum phase representation of the actual response and a target curve, load into MSO and create a configuration that has a load of adjustable Q shelf filters & some PEQ available to it then give it some time to work out what combination of filters produces the target response. 

FWIW I used jriver convert format to generate the input wav file and then these are run through my app in a way that is equivalent to the approach used around here in speclab (which decimates it to 1000Hz and then runs it through the fft with a nuttall window of either 0.5Hz or 1Hz resolution, I forget which). I used a 0.5Hz resolution in my case, this is pretty much all done by scipy (for the fft bits) and librosa (for the resampling)

I thought I'd double check so used eac3to to generate the wavs instead (though they both use the same underlying dts decoder anyway - https://github.com/foo86/dcadec), same result. 

If you get a different result then it would be interesting to know how you're performing that extraction and conversion.

It is L R C LFE RL RR SL SR (which is pretty standard afaik)

8 hours ago, SME said:

Assuming I can't get a good GOTG2 BEQ before next weekend, I don't suppose anyone would be willing to post channel-by-channel PvAs for GOTG2 somewhere so I can take a crack at one? 

try this https://imgur.com/a/mjxeS

I notice (in that speaker shootout thread on avs) you commented about certain aspects of your room response being SBIR rather than modal in nature. I was curious to know what you meant by this. I suppose I am thinking of 2 specific question. What makes you say it is SBIR rather than modal? what practical difference has this made to your treatment strategy?

it accepts biquads -> https://sourceforge.net/p/equalizerapo/wiki/Configuration%20reference/#filter

I tried another one and got similar results, hacked together a multichannel wav with a sine wave at a different frequency for each channel and ran that through jriver then analysed it, result is a mono wav with a missing channel 4, i.e. no LFE channel in the output. I'm quite sure this used to work so will report over at jriver forum to see what is going on.

edit: see https://yabb.jriver.com/interact/index.php/topic,111023.0.html  appears to be some random defect in this particular installation of jriver :|

edit2: turns it out it is a bug in jriver - https://yabb.jriver.com/interact/index.php/topic,111023.msg767472.html#msg767472

-20.2 dB for all channels except the LFE.  -10.2 dB for the LFE channel then combine into a single mono.wav file of the same Bit depth.

that is handled via jrss downmixing (+ the additional single level adjustment)

Any idea why your PvA looks so different from what JSS posted above?

 

Yours appear to be higher resolution (longer window?) but looks smoother, so I don't think it's a resolution issue.  Indeed, the curve shape and levels both look quite a bit different.  Could the difference have to do with how the down-mixing is being done before analysis?

dunno, I downmix using jriver as per http://www.avsforum.com/forum/113-subwoofers-bass-transducers/1333462-new-master-list-bass-movies-frequency-charts.html#post23468771

I just doublechecked the settings, recreated the mono wav and ran it through speclab and my app. I get the same result each time.

1. Level - This is measured by digitally bass-managing the 5.1 or 7.1 signal.  The Level is a composite number, and is calculated by the average of the [1] highest peak in dB (maximum 126dB for 5.1, 128dB for 7.1), [2] the average/RMS dB level of the track (125ms integration time) and [3] the RMS peak level (loudest single 1/8th of a second of the film) in dB.  Full modulation of the waveform is considered to be 0dBFS.  The ratings are as follows:

 

5 Stars - >112.5dB composite

4 Stars - >110dB composite

3 Stars - >107.5dB composite

2 Stars - >105dB composite

1 Star - <105dB composite

 

@maxmercy how do you calculate [2]? I don't understand how the integration time relates to a single dB level for the entire track. The time window sounds like it relates to the length of each slice in the spectrogram but then how do you turn that into a single number?

I haven't seen it yet, it's on my pile of things to watch, sounds like one to watch sooner rather than later though

I recommend "Kubo and the Two Strings".  The bass in this is the real deal.  Trust me.

 

Edit: It's a pretty good movie too.

I have that one, this is what I get for it - http://imgur.com/a/Bin6h

fwiw I finished off a new version of my app which includes the speclab style wav processing I mentioned a few posts back - see https://github.com/3ll3d00d/vibe/releases/tag/0.4.0for the windows exe (or http://vibe.readthedocs.io/en/latest/install.html#installing-the-analyser for linux). It should be easy enough for anyone to pick up and use.

let me know if you hit any issues, don't think anyone else has used it yet I could easily add some options to make it easier to use for this case though, for example 

auto detecting files dropped into a certain directory

providing some config option for tweaking the fft length

providing a quicker way to see peak/average on one chart (rather than having to select peak then select average each time)

fwiw the vibration app I wrote does the same sort of analysis as speclab for the target curve function - http://www.avsforum.com/forum/155-diy-speakers-subs/2681865-rpi-based-diy-vibration-meter-3.html#post52214001- and the relevant bit of user guide is http://vibe.readthedocs.io/en/latest/uiguide.html#wav-upload

It is currently tuned to produce output with ~1Hz resolution though, it decimates to 1kHz sample rate and then performs an analysis based on a 1024 point fft. Bit easier to use than speclab as you just drag and drop wav's into it, no options exposed to configure it though.

I think that sort of thing is most commonly packaged as a VST