DSP Robotics Support

Hi Guys,

Please bear with me as I'm thinking over a new project.

Most important ... can it be accomplished with FlowStone ?

Before boring too many details [i know, you're welcome] ...

1. Is it possible to 'capture' the spectral envelope of an audio track? Programmed via FlowStone ?!?!?

Basically, playing an audio wave file into an analyzer that would freeze/hold that overall response. IDEALLY to use the resultant 'shape' as an EQ shaping.

I know we have all kinds of Capture/Match EQ plugins all over the market. I've purchased plugs like FabFilter Q-2, Voxengo Curve EQ, iZotope, etc . But I have a very special project idea/research concept that I need to explore [and would be happy to share the overall concept with you Guys], but I first need to get some preliminary information together.

I've been following, heck, every thread on this forum. I'm following the 'convolution' project with much interest.
As a 30+year audio engineer, I quite familiar with the process, but I'm not versed on the technicals/terminology of the inner workings.

I'll hold here to these initial questions.

I certainly appreciate all enlightening.

Thank-You

Spectral envelope of a track:
1. Do you mean an average over an entire track? Or some shorter interval?
2. What resolution are you aiming at: Octaves? Semitones? Something in between?

martinvicanek wrote:Spectral envelope of a track:
1. Do you mean an average over an entire track? Or some shorter interval?
2. What resolution are you aiming at: Octaves? Semitones? Something in between?

1. It may need to be PEAK level [though RMS might need to be tested.

The length of the audio source should cover the average song length. [3-4 min]. That would be a starting point.
It may be that a 30 sec sample would be enough time to stabilize the response.

2. resolution. A least 1 oct. Might need as fine as 1/3 oct.

Without any practical testing, I'm having to consider a broad range. I don't mean to be so wishy-washy.

If this is not to confusing ... let me try this

1. I do NOT want to make a 'match EQ' for the specific purpose of of imposing a source curve to a destination. We already have plenty of those. Not interested in that.

Actually ... I'm looking at a type of 'Gain-Staging' concept.

Here's the deal.

Working with plugins like NEBULA, the libraries are arranged into specific bands for each preset. If you need 6 bands, this requires 6 plugins. It can easily be more based on the project.

Because of the plugin nature, this means opening/closing many windows as adjustments are made. Very bad for workflow/concentration.

My MIDI controller can talk up to 16 Nebula plugins, all from a single GUI interface. Works great [thanks to much help from here]

The next big issue .... proper 'gain-staging' through the entire chain.

I already use Ian Shepards 'PERCEPTION' plugin, that monitors the top and bottom of the chain and provides automatic , perceived', gain compensation. That part is fine.

I use KLANGHELM's VU meters between each NEBULA instance to both monitor level, and adjust TRIM. Again, all good there.

The time consuming issue. I do NOT want to play a track, and open/close windows while adjusting for 'Unity Gain' from one plug to the next.

What I have been using is a modified form of 'PINK NOISE', generated from the MIDI controller. Using this, I just

1. stop playback
2. send 'PINK' at a calibrated -18db [RMS] through the entire chain. [made up of various bands, shelfs doing cuts/boosts ... whatever.
3. adjust each VU TRIM to read -18. [actually, meters are ref'd to -18, so the display is 0dB].

Generally, this works fairly well. Across 8 eq plugins, the level playing the actual source track may be 1 to a few dB off. Not to bad across 8 plugins. But sometimes, it can be more significant.

Before anyone criticizes ...

I'm looking to speed up the process of gain-staging. Nothing stops from doing it the traditional way ... however, even using the current PINK technique has more than proven its usefulness ... SONICALLY [Nebula is level sensitive].

Typically, I can balance the entire effects chain in 15 seconds or less this way.

But I looking to improve the concept.

The thought.

What if I was able to capture the sonic envelope of the audio source, that has the overall EQ curve, so that WHEN the track is NOT playing ... I send the PINK signal THROUGH the captured EQ at a defined reference [-18dB RMS], whereby each of the EQ's are seeing the source curve at levels that simulate the actual source track.

I've been experimenting with this concept using FabFilter Q2 'Match' function. The resultant EQ curve is inserted after the PINK generator, then followed by the efx chain.

So far, the testing is proving to be quite effective.

Because of these results, I was looking to see if this could be accomplished using FS [rather than FabFilter].

I'm currently ignoring the fact that I've never tried to program anything of this nature ... if I even could.

I don't even know the feasibility using FS ! At my age ... what else kinda trouble would I want to get into

I hope some of this makes a little sense. Heck ... I appreciate just having this read.

[aside]. It's possible to question the use of the PINK cal noise. Normally, PINK noise tends to have bouncing LEVELS. The design I'm using is fairly stable, level wise.

I've thought about the type of 'chirp' tone that is used in REW for 'shooting' out a room. Back in the daze, we didn't have all the digital technology available, and often used OSC and noise generators for calibrating tape decks. But also we used these tones to quickly balance a chain, or fine tune a recall session from documentation and polaroid photos

Back then ... everything was hardware ... with real knobs, buttons, and meters. Today its multiple windows and a mouse. [i hate it] but that's the way of today. I'm just looking to improve the system and workflow. All of this comes together to help focus on the main part of the job.

LISTENING to the MUSIC. That's what I get paid to do. Just looking for a better solution.

Hey ... thanks for reading and replying ... to everyone. I know several of you Guys that really do some amazing things programming and knowledge. There is no other place or people I can even begin to talk about stuff like this. So again, thanks for being here ... and all the kind help.

Sincerely.

hmm ... I see I've scared everyone away.

darn it

Here is a 10 octaves/40 bands filter bank that might be useful. I have connected it as a live display, however it should be possible to apply it offline. Have fun!

Thank you as well.

More Martin Magic!

Love it.

Cheers

Spogg

This IS really nice Martin. Thank-you.

I obviously need to put time in and study the details, not only the 'hows it work', but also how you organize/layout a design. You always do seem to provide 'multi-levels' of learning, which is most appreciated.

This project concept of gain staging, is something that permeates the audio recording forums. It is regularly asked about. There have been utilities that aid in addressing this ... but I still see another level this might be taken to.

My experiments continue to show positive results with various software I mentioned before.

I currently mastering 3 new projects that cover a wide range of musical styles from DixieLand, to Big Band, to restoration of a vintage jazz performance that I've been commissioned to restore. During which I take a moment to do this 'experimental' test concept.

I've only started to look and test the 'Spectral Envelop' schematic.

With it 40 bands, I'm fairly convinced that this would provide more than enough spectral analysis and band gains.

I'm now considering a first step to capture these bands into separate arrays.
2. then AVERAGE out each band.
3. then post that to the display just to get a visuals of the result of this first step.

I've no idea what limits I may hit or run into. This is completely new territory for me. So with bare minimal experience, and a push to experiment, I can probably crash the best of computers

In the back of thought, I'm thinking of the FFT capture example dealing with Reverb convolution.

Bottom line thought. Can we capture, into a single, static instance, the energy and spectral envelop that represents the original audio source.

Then use that static 'snapshot' as a filter.

Feeding a reference, broad band noise, like PINK, through that filter would then pass either the MAXIMUM [peak] band levels [or the Ave RMS level ... depending on design], whereby the successive FXchain would respond as if it dealing with the original source.

The big difference. This 'CAPTURE' would be MAX level [static], rather than streaming the entire source JUST to determine this MAX level.

The hope is that we would have MAX determined.

The Reference noise would feed through the chain starting at a known level [-18dB RMS], and with VU's inserted between processing plugins, we would TRIM for UNITY GAIN ... what goes IN comes OUT.

During production, as we make gain changes via EQ, the outputs of each plugin will change the gain structure [as expected].

We then re-activate the 'capture filter', and send the Ref-Noise to re-balance.

This is a normal procedure that can involve many many re-balances during the session. It could easily be in the 100's of times to do this ... for each song being master, or even each track of the song depending.

I'm leaving out the boring details, but as routine 'gain staging' is ... it is very important. I'm just looking for a new/different approach for workflow efficiency.

This is a concept that I'm looking at. With decades doing this as a full-time Profession, I don't pretend to know it all. Every session can be another learning experience.

I'm passionate about music, sound, and my work. FlowStone has made capabilities for me to try ideas/concepts. I don't know FS's capabilities ... I DO know my limitations.

I welcome ideas, comments, observations ... and strategies that would further this concept into a schematic that could be tested. It is also one of the reasons I follow all the postings on this forum. Some of my ideas tend to surpass my hobby skills with FS, yet I find it a strange type of fun ... intrigue.

Thanks again to everyone. I welcome the participation.

Wow, it's simply great martin

Tried to play with the ruby code a little, but couldn't apply the colors correctly (again, coding limitations).

Any chance that someone would show us how to add to it an outsider coloring painting where you can simply choose your colors to the graph and the background?

COOL STUFF