DSP Robotics Support

martinvicanek wrote:Thanks for the feedback, guys. Here is a variant with offline processing. Processing is done in blocks and results are cumulated, so the final readings are RMS values. Unfortunately it takes some time to process an entire track: on my computer it is about the time you would need to stream it. May be there is a smarter way to do it?

Oh, and I added a color input to the Ruby display code.

Hi Martin,

I have been trying to study aspects of the 'SpectralEnvelope-2' that you kindly posted. Even with your, envious, design and function , I've been looking to see how to implement a type of reverse design.

What I mean, you already analyze and hold each of the 40 bands in an array ... is the possibility there to [in a separate module], feed those array values [static-final] back thru the myriad of filters that could act upon a different audio source ? I've been probing points of possibility, but maybe I'm looking for something that can't be done with this.

I have a working model based on a 10-band splitter [probably designed by You] ,that I use to handle both processing. [expanding the 10-band to 40 does not seem possible for me, at least in this lifetime ]
Also the ability to stream/analyze off-line is fantastic.

Not that it would help, I'd be happy to post what I have using the 10-band, just to show that it does exist, it's still in my crazy layout look [just short of nightmare for anyone else ].

thank-you for everything

RJHollins, yes, I realize that a synthesis filterbank is needed to achieve your goal. Since it is for mastering, you would probably want linear phase? I have some ideas about a possible design, however there are a few things I want to finish first, so it may take a while. Can you wait?

Absolutely.

Just this point [to better realize the intended goal]. The accuracy needs are to replicate the overall frequency spectrum and transpose that into EQ bands. This EQ'd signal is then the 'model' that a calibration noise [or sweep] would be used to pass through this 'snapshot'. It is not envisioned to actually process the signal for mastering ... but as a static response that is referenced for gain structuring down the chain.

The idea might sound more complicated than it aut be. This is an experimental idea, so I'm trying to stay open on the process. The closest model is using a MATCH-EQ result in a non-traditional way. I'm looking at FS to provide the 'match' function AND to be able to play audio THRU this 'match', affected by this captured curve.

Still ... Thank-you once again !

What you're describing I believe is simply a vocoder where the bands do not "release."

Forgive my ignorance. I'm just trying to understand what you're aiming to accomplish. But I don't understand what the frequency content has to do with the overall gain/amplitude of the signal in your application.

Why not just find the peak amplitude of the overall signal?

Perfect Human Interface wrote:What you're describing I believe is simply a vocoder where the bands do not "release."

Forgive my ignorance. I'm just trying to understand what you're aiming to accomplish. But I don't understand what the frequency content has to do with the overall gain/amplitude of the signal in your application.

Why not just find the peak amplitude of the overall signal?

Hi PHI,

Maybe a 'vocorder' is the beast needed ... i don't know.

What I do know from testing ... I do need a 'capture EQ-curve'. Why?

This captured curve will then be the model that a calibrated PINK NOISE will feed through. That new output, totally influenced by the capture-curve, will be fed into a series of actual EQ's used for the mastering process. TRIM are applied at each EQ stage to match cal'd level.

Keep in mind, I'm using a series of EQ's [not 1 algo, like might be normal]. This 'stack' of EQ's could easily be 8 or more. Each EQ is a single band ... so not as dramatic as it might appear.

In practice, each of these, say, 8 EQ bands will affect the signal output, so gain staging is required at each instance.

NOTE: I'm looking to replace the practice of playing the actual audio file. The 'captured-curve' will BE the temporary replacement. The 'calibrated' noise/tone will use -18dB RMS as reference.

My testing has envolved using FabFilter's Q2 equalizer with its MATCH EQ feature. The match is using the original music source matched to the REF PINK noise. The resultant curve was then used as the static model.

Testing has shown an accuracy of 1/2 dB per EQ instance [using 8 separate bands]. Not bad.

2. Gain staging through the entire EQ chain takes maybe 15 secs. Again ... much faster than the tried, traditional method [as you suggested, and as we've done for many years].

I'm looking to improve this process. These early test/research show very promising results, so I would like to proceed with a FlowStone solution. The initial difficulty ... I'm not exactly sure of the name of the programming techniques ... a Captured EQ ... a Vocorder ... I just don't know, and why I started this thread in hopes the FS 'GANG' can relate and explain.

If this idea sounds like a 'vocorder' candidate ... I'm interested in learning more about this. My experience is only from the recording side of this.

Can a vocorder capture a source envelope and hold it? I know it's a filter, and that a secondary signal can be passed through it.

Right now I have a 10-band on the FS test-bench. Each band is tracking its level during 'LEARN' mode. It holds the levels. I'm thinking I may need to raise the resolution [using narrower bands]. I do want to be mindful of phase concerns with these bands. Things like 'OVER-LAP', and 'GAPS' need to be minimal to non-existent.

Martin has provided some very interesting tools to this point. I just needed to remind that this 'Capture' will not actually be used in the processing. It's ONLY use is to mirror the source audio as a static snapshot, so that the REF tone can pass into each EQ with the same band energies as the source audio has.

Please ask if I can clarify or further confuse this

Thanks!

OK, I'll be the one to hold my hand up!

After you've captured the source track's eq curve this will feed into a bank of matching filters, like a "frozen" or held vocoder.
Then you pass pink noise through the filter bank which is now set up to mimic the reference track's eq curve.

What I don't understand is what you then do with the processed (filtered) pink noise which is carrying the eq curve of the reference track.

Cheers

Spogg

Ahh ...

With the Filtered EQ curve passing the CAL REF NOISE ...

We measure the LEVEL coming out of each EQ.

TRIM that EQ output level back to UNITY GAIN [what goes IN = OUT].

Proceed through next successive EQ, bringing its OUTPUT to UNITY.

etc to end of Chain.

2. Now we can dis-able both the NOISE source AND the snapshot FILTER.

3. Continue mastering work.

NOTE: throughout the EQ process, making change/adjustments to EQ frequency, Q, and/or Gain [+/-], the chain again leaves UNITY.

We can then engage the NOISE/FILTER combo, and do a re-balance.

This can take place dozens of times. Engineers' discretion when to re-balance. The key feature ... you eliminate visually monitoring each of the EQ's [with many GUI windows opened], scrolling to THE section of the audio source to play/loop and balance. Because you are NOT having to still listen to the track, you also get a momentary 'ear break', which can help you return with focus.

Yes, we still have GUI's to open, but they are VU with PEAK readout and TRIM. [BTW ... there is another FS project that I'm thinking to make that more efficient ... later time].

Hope this helps .... I think me trying to describe it makes it seem much more complex than it is. I used this exact concept [with FabFilter MATCH EQ] function. I had a chain of 8 Manley - Massive/Passive EQ bands, all routed and chained as a Plogue BIDULE preset.

I use my FS designed MIDI controller plugin [a single GUI interface], that communicates all commands to the 8 EQ's in the chain. I basically have these individual bands in the background, as I never need to look at them. [This controller was actually part 1 of this overall concept. [mucho credit to several FS Guru's and participants] Actually, the MIDI controller is a type of RACK design. It can handle up to 16 external plugins [thru 16 MIDI channels]. Bands, whether PEAK or Shelves .... Clean or full dynamic harmonics can be decided and activated as the engineer needs. There are also some special features added, that I've learned after 35+ years working, to enhance the workflow. [could not even begin to do this had it not been for FlowStone].

Again ... thanks for reading all this. I really do appreciate it.

Thank-you

I kinda think I've got it. Let's see...

You use the system proposed to calibrate a different eq system. Then you are finished with the FS proposed system, relying on your newly calibrated eq chain. I assume this is because there is an advantage to using your other eq chain over just using the FS filter bank.

Yes/no?

Cheers

Spogg

I think ya got it !

I may need to put together some charts and graphs to illustrate

The actual EQ's that I use for Mastering are very different from an algorithmic design. The difference can be like comparing a Synthesizer to a multi-dynamic Sampler of a specific, REAL instrument. Not a mathematical, modeled emulation. Basically, the EQ's are multi-layer samples of actual analogue hardware. Not just the frequency/gain/Q properties, but also harmonic structure ... and they are dynamic. Technically .... they are multi-sampled convolutions. And yes ... a 'super-computer' is really handy.

As to the FS project.

First, we must dispel previous or normal practice.

A typical procedure ... calibrating, say, a tape machine.

A 1K tone is generated from the Console with meters reading 0dB. At the tape machine, we adjust INPUT TRIMS to match 0 dB. Or look at a mixdown session to tape. We send a 1K, 10K, and a Low freq [30, 60, 100], and print this at the head of the master tape. We can then use these recorded 'test tones' to playback on a different machine at the Mastering facility for cutting to vinyl or prep for disk. Basic stuff back in the day.

Now we look at an effect chain. There could be compressors, EQ's, EFX ... whatever. Units like compressor can usually be Cal'd via a 1K tone. But an equalizer is a different beast.

Set 'FLAT', an EQ can be made UNITY balancing the OUTPUT to the same level as the INPUT.

However ... if the EQ has a band set at, say, +10 @ 100 Hz, and that is engaged ... a 1K test tone will not be affected by a band set at 100 Hz much ... the OUTPUT may change slightly.

Now we play a Kick drum or Bass through that EQ [with a +10 @ 100Hz] and the output from the EQ will be much greater. This will now impact the next FX in the chain.

So we look to a different 'calibration' source. Full spectrum. We've 2 basic choices: PINK or WHITE noise. I've chosen
PINK for several reasons.

Seems that this would solve all problems. Just use PINK noise ... send that through the EQ chain. A +10dB band gain will respond accordingly ... re-balance each OUTPUT thru chain .... done.

Now play the actual audio source. Hmmm .... now the levels don't seem to react the same. Reason ... the actual audio source has a different spectral envelop than the test PINK noise used to set Unity Gain Structure.

So ... what if we SHAPE the PINK noise to match the actual audio source ?

If the audio source was down -10dB at 100 Hz, and we are adding +10@100 ... this is much different than adding +10dB to the 'flat' PINK noise [I'm calling it 'flat' for audio purposes]. ok?

But WHY are we going through all this ! .... they ask [snickering in the background].

TIME. Recall ... We are not using a single, selt contained, ALL-Bands equalizer with built in Gain Compensation.
We may have 8 or more separate units to balance. OR ..... we may be using actual outboard HARDWARE

Traditionally, we played the actual audio source [maybe a loop], and watched how the meters responded, waiting for that decisive PEAK ... and adjust level. Then continue on to the next unit in the chain. Etc, etc ... We did it like this for decades.

That's not the end ... make some changes to the EQ settings ... the gain structure changes.

Small changes to EQ settings are not the end of the world ... music and it's levels USE to be DYNAMIC

Things are a bit different when it comes to Mastering. YES ... we still use all the techniques available. We can DRIVE an EQ to get some added harmonic action happening. Other times .... we need Pristine. It all depends on the project/client.

Also ... the Mastering EQ's have special DRIVE function, so that from UNITY GAIN, we can internally control DRIVE on each unit.

Great ... now back to saving TIME.

By using a 'Captured Spectral Response' of the actual audio source as a FILTER. We can then send PINK through this Captured FILTER, and efficiently step through the chain and make UNITY gain between each unit. No looking for and looping. Simply:

1. Turn DOWN the monitors
2. Engage the capture Filter
3. Apply PINK.

Crazy idea ... yeah. Refinements to come ... most welcomed

This is part 2 of a project strategy. Part 3 is in the works The working title, .... 'Deja-VU'.

Again with the LONG posting. I don't mean to do it like that ... it just seems to happen

I know some are fearful .... but please, ask questions ... offer solution ideas ... i'll try harder to be brief in response ... HEY ... i heard that !

really .... thanks Guys for being here. If I had just one other person around me that had a clue ... the BURDEN youse Guys bare would be lessened.

thanks.

Thanks for expanding on the details RJ.
I know nothing about commercial mastering so this has been very interesting for me.

It would seem to me that, if I am following this correctly, that the accuracy or faithfulness of the modified pink noise eq is quite critical to the usefulness of the procedure. In this case would it be accurate to say that 24 bands would be much better than say 10 or 16?

As a side issue doesn't this all get messed up if you then use RIAA eq to cut the groove into a master? Or is this step just SO last century?

Cheers

Spogg