Feedback Compressor II - Gentleman's Edition (discussion)

[FabienTDR]

[Zombie Queen]

Gentleman's edition, pink lady's edition, sounds cool, although it would be hard for me to identify with either, how about "bum's edition" with even less controls than the regular one and some "less shiny" gui?

Seriously I really like the thing, but I still haven't learned how to handle it properly, it's developing faster than I am.

[FabienTDR]

Gentleman's edition, pink lady's edition, sounds cool, although it would be hard for me to identify with either, how about "bum's edition" with even less controls than the regular one and some "less shiny" gui?

Seriously I really like the thing, but I still haven't learned how to handle it properly, it's developing faster than I am.

Oh, don't worry. The standard edition will remain as it was (except bug fixes of course). I am aware that the compressor isn't particularly easy to understand. Concerns about over-boarding complexity already appeared early in this thread.

The GE edition is still a work in progress, and I'm trying to keep the options as simple and relevant as possible. This thread documents it well, it began with countless ideas, but only very few pass my tests!

With regard to your specific demand for an easy to use, "set and forget" type of dynamics processor, note than I have a 3-4 knob tracking/mix oriented compressor on my todo-list. It's based on the "big" compressor, but will hide most parameters and offer the huge amounts "grab", "style" and "color" required for these tasks.

Also, the GE edition itself will feature two "easy" Peak and RMS modes which reduce the work-flow to a single release stage.


Tell me which parts of the manual troubled you, I would be happy to improve them or explain a few concepts via PMs.

Always wanted to make a video tutorial explaining the elementary principles of the device, but I rarely find the time for such things (I even bought a fraps licence!). But I realize it's a must for the GE edition at least.

Thanks for your input Zombie-Queen!

[Zombie Queen]

The manual is quite clear, I had no trouble understanding the principles. It's practical use, where I need to learn how to handle things. Having RMS and peak detector switching the steering wheel is a bit confusing at first... How about mono? The current (regular) version is supposed to work with mono input?

[FabienTDR]

The manual is quite clear, I had no trouble understanding the principles. It's practical use, where I need to learn how to handle things. Having RMS and peak detector switching the steering wheel is a bit confusing at first... How about mono? The current (regular) version is supposed to work with mono input?

The standard edition package comes with a dedicated mono version.

[Zombie Queen]

The standard edition package come with a dedicated mono version.

Of course it is... I will explain my confusion here. I was trying it out on a vocal track, which was mono, (I tend to use stereo versions in such situation, in case I decide to make a panned double-take, while editing), I added some effect in-track and switched the track to stereo, which caused the compressor to produce some heavy distortion, it spooked me out then. I checked it out just now and both versions (mono/stereo) work that way, that is they don't like stereo to mono or the other way switching while playback - after playback restart is sound normal. I guess I've got caught by surprise here.

[FabienTDR]

Definitely keep an eye on the GE edition. It will feature simple* peak and rms modes, too.

*Both paths are specialized for their own tasks, the RMS detector is program dependent by its own, and the peak path is very fast and precise with regard to the true waveform.

[sdbmastering]

Looking forward to the GE!

For my current uses there isn't much that I feel that needs to be added to the regular TDR Feedback Compressor. Though having more features is never a bad thing for me...

Sorry for going off-topic, but can we expect other products from TDR, maybe a limiter or an EQ?

[FabienTDR]

Thanks for your feedback.

I have countless prototypes and ideas on the desk, but making a stable and reliable product out of them is very time consuming. I have to chose my projects wisely.

I will definitely focus on dynamics processing first. No need for another EQ. But I'm experimenting with unusual, perception based interaction approaches to EQ, but nothing particularly exciting yet.

[xtrax]

GS version Release time

Can we expect it this month?

[Hermetech Mastering]

Looking forward to this!

[FabienTDR]

xtrax, as soon your time machine's ready!

[Blipper]

I propose that the sidechain highpass filter, together with the tilt filter offered by Klanghelm and elysia, in compressors is an attempt to provide a subjective loudness weighting curve. The ITU BS.1770 committee has done a wonderful job of making a subjective loudness weighting curve with two second-order IIR filters. See pages 3-5: http://www.itu.int/rec/R-REC-BS.1770/en

I suggest that the SIDE-CHAIN section of GE edition includes a BS.1770 button, which would engage the head-modeled pre-filter on page 3 above, and would use the same HP filter slope (12dB/oct) and Q as the second-stage filter on page 5 above. I also suggest a BS.1770 preset, where the HPF corner frequency setting matches the response on page 5 above.

Alternately, if the 12dB/oct HP filter always has the same Q as the BS.1770 filter on page 5, then the BS.1770 button would simply engage the head-modeled pre-filter on page 3. This would allow combinations with the other HP filter slopes for more flexibility. Perhaps this button could be labeled "HEAD MODEL".

I am very happy that the GE edition will have an external sidechain input. Am I correct in my assumption that external sidechain input is only used when RMS SOURCE=feedforward? If so, implementing feedback compression with an external sidechain filter will cause at least an extra DAW buffer size of delay in the feedback path, and it is tricky to set up. Therefore, it would be of great benefit to users if the GE edition included the BS.1770 head model in the internal sidechain filter (which can be used in both feedforward and feedback modes).

[FabienTDR]

Interesting suggestions! Find my comments below...

I propose that the sidechain highpass filter, together with the tilt filter offered by Klanghelm and elysia, in compressors is an attempt to provide a subjective loudness weighting curve. The ITU BS.1770 committee has done a wonderful job of making a subjective loudness weighting curve with two second-order IIR filters. See pages 3-5: http://www.itu.int/rec/R-REC-BS.1770/en

I suggest that the SIDE-CHAIN section of GE edition includes a BS.1770 button, which would engage the head-modeled pre-filter on page 3 above, and would use the same HP filter slope (12dB/oct) and Q as the second-stage filter on page 5 above. I also suggest a BS.1770 preset, where the HPF corner frequency setting matches the response on page 5 above.

I will read the paper later today, thanks for the link. Page 3 seems to relate to stereophonic and surround effects?

IMHO, the idea that a compressor (or other dynamics devices) should react to the equal loudness curve is flawed.

It's maybe just a philosophical thing, but I expect a compressor to compress the audio waveform rather than the subjective loudness. From my experience, it doesn't work as expected. Or in other words, it sounds bad. A reason is that the equal loudness curve is extremely level dependent, with extreme variance even for a few dBs of difference, and even more biological variance from individual to individual. It only depends on the end-users playback level! Such a SC filter is great if you're looking for intentional pumping and weird effects, but is incapable of doing "good" compression (at least from my experience).

From the theoretical point of view, the inverted pink filter (3dB/oct high pass) already represents the ideal filter for wideband audio dynamics compression with highest transparency (i.e. low pumping artefacts). The typical frequency distribution of well produced music does not follow the loudness curve, it approximates pink noise distribution. The sum of all sound event in nature approaches this distribution.


I already added a wider 3dB/Oct mode (which extends over ~90% of the audio band). It has an even more pronounced positive effect than the "simple" 3dB/oct filter.

Also, a new "HF Tame" mode does more or less the same thing as the curve in the paper. It wasn't planned to make it switchable in addition to a high pass, but you made be reconsider the subject!

I am very happy that the GE edition will have an external sidechain input. Am I correct in my assumption that external sidechain input is only used when RMS SOURCE=feedforward? If so, implementing feedback compression with an external sidechain filter will cause at least an extra DAW buffer size of delay in the feedback path, and it is tricky to set up. Therefore, it would be of great benefit to users if the GE edition included the BS.1770 head model in the internal sidechain filter (which can be used in both feedforward and feedback modes).

The RMS Source parameter seen in previous screenshots has been removed.

Instead, the GE edition will feature 4 general modes: Parallel (default), Serial, Peak and RMS. It will also provide much higher ratios. The latter has the side-effect that low ratios act like classic FB compression, but more and more resembles typical FF compression behaviour with increasing ratios.

Also, the old FB comp already uses an advanced approach to act like a FB compressor, but without all the technical drawbacks of the "naive" implementation. It can handle external SC without extra delay, no matter how it is tuned. No need to worry about this.

No matter all the points above, the external SC input will still pass through the SC processing block (filter and stereo difference network). The external SC input will most of all ask for a substantial amount of additional CPU power (about 25% more), but no extra delay.

[Blipper]

I will read the paper later today, thanks for the link. Page 3 seems to relate to stereophonic and surround effects?

To clarify, I am referring to the page numbers displayed in the English/PDF version of the latest (BS.1770-3) document. These page numbers are different than the logical page numbers displayed in my PDF viewer. Also, yes, BS.1770 applies to stereo (2-channel) and multi-channel audio.

IMHO, the idea that a compressor (or other dynamics devices) should react to the equal loudness curve is flawed.

It's maybe just a philosophical thing, but I expect a compressor to compress the audio waveform rather than the subjective loudness. From my experience, it doesn't work as expected. Or in other words, it sounds bad. A reason is that the equal loudness curve is extremely level dependent, with extreme variance even for a few dBs of difference, and even more biological variance from individual to individual. It only depends on the end-users playback level! Such a SC filter is great if you're looking for intentional pumping and weird effects, but is incapable of doing "good" compression (at least from my experience).

From the theoretical point of view, the inverted pink filter (3dB/oct high pass) already represents the ideal filter for wideband audio dynamics compression with highest transparency (i.e. low pumping artefacts). The typical frequency distribution of well produced music does not follow the loudness curve, it approximates pink noise distribution. The sum of all sound event in nature approaches this distribution.

I already added a wider 3dB/Oct mode (which extends over ~90% of the audio band). It has an even more pronounced positive effect than the "simple" 3dB/oct filter.

Klanghelm and elysia have a tilt filter. Try elysia's filter here:
http://www.plugin-alliance.com/en/plugi ... ilter.html
What do you think about implementing a tilt filter (over ~90% of the audio band) with a range of say 0 to 12dB/Oct (for more flexibility)? Maybe have a preset for 3dB/Oct tilt?

Also, a new "HF Tame" mode does more or less the same thing as the curve in the paper. It wasn't planned to make it switchable in addition to a high pass, but you made be reconsider the subject!

Glad that I am not too late with suggestions for this development.

The RMS Source parameter seen in previous screenshots has been removed.

Instead, the GE edition will feature 4 general modes: Parallel (default), Serial, Peak and RMS. It will also provide much higher ratios. The latter has the side-effect that low ratios act like classic FB compression, but more and more resembles typical FF compression behaviour with increasing ratios.

Also, the old FB comp already uses an advanced approach to act like a FB compressor, but without all the technical drawbacks of the "naive" implementation. It can handle external SC without extra delay, no matter how it is tuned. No need to worry about this.

No matter all the points above, the external SC input will still pass through the SC processing block (filter and stereo difference network). The external SC input will most of all ask for a substantial amount of additional CPU power (about 25% more), but no extra delay.

The extra delay I am concerned about is in the DAW software when routing the compressor output back to the external sidechain filter. Let's say this is 1024 samples for 192kHz sample rate. Are you saying the GE edition won't be affected by this feedback latency?