Audio clipping

When I export a video, the sound is always clipping.
Everything's fine in Fairlight. No clipping.
Can someone help me with that?

My version
DaVinci Resolve v15.0.0b.05 (Windows/MSVC)

My systhem
Win 7
2x GTX 1080
Realtek Audio

Timeline

resolve.png (70.51 KiB) Viewed 20167 times

[Peter Chamberlain]

Please review this thread as we could not guess the circumstances of your issue.
viewtopic.php?f=21&t=69837

Having the same issue. Fairlight is fine, Edit Tab and Export are Clipping.

Having the same issue. Fairlight is fine, Edit Tab and Export are Clipping.

Hello,

Surely (?) a solution but in doubt (?)
Like Peter : viewtopic.php?f=21&t=69837

Thanks for the reminder!
I LOVE being able to do everything in one program, learning everything from scratch, leaving After Effects, Audition and Premiere behind, but this "bug" really forces me to go to Audition every time:

I think the "bug" is:
In the Fairlight Tab, there's ALWAYS an invisible Limiter added after the Master.


Studio version: 15.1.2.008
Windows 10 1803
GTX670
i7-3770K
16GB RAM

Detailed description:

1. Importing wav file in Resolve
2. Opening new timeline
3. Putting the wav file in the timeline
4. Right-click, normalize audio levels to 0 dbfs (results in 14.70 db boost in the inspector)
5. Clicking on the Fairlight Tab
6. Increase the audio level in any way (for example pushing the Audio 1 fader to +10)
7. Audio sounds louder, levels are higher, but no clipping (neither a clipping indicator nor any audible distortion)
8. Switch back to Edit Tab
9. Play
10. Extreme clipping (as should be expected after normalisation to 0 dbfs and +10 boost)

Now maybe this is a "feature", I don't know much about audio, but for me it's a big problem, because I need to know when the audio is clipping without always going back to the edit tab. Also I need to use the Limiter in Fairlight to make sure, the audio is not clipping while tweaking the compressor to make the LUFS louder. This doesn't seem to work either:

1.-5. the same
6. Open dynamics of A1
7. Enable Limiter with Threshold at -1
8. Increase Make Up to +20
9. Play (Limiter works fine)
10. Switch back to Edit Tab
11. Play
12. Extreme clipping (as if the limiter isn't applied)

Thanks for the reminder!
I LOVE being able to do everything in one program, learning everything from scratch, leaving After Effects, Audition and Premiere behind, but this "bug" really forces me to go to Audition every time:

I think the "bug" is:
In the Fairlight Tab, there's ALWAYS an invisible Limiter added after the Master.


Studio version: 15.1.2.008
Windows 10 1803
GTX670
i7-3770K
16GB RAM

Detailed description:

1. Importing wav file in Resolve
2. Opening new timeline
3. Putting the wav file in the timeline
4. Right-click, normalize audio levels to 0 dbfs (results in 14.70 db boost in the inspector)
5. Clicking on the Fairlight Tab
6. Increase the audio level in any way (for example pushing the Audio 1 fader to +10)
7. Audio sounds louder, levels are higher, but no clipping (neither a clipping indicator nor any audible distortion)
8. Switch back to Edit Tab
9. Play
10. Extreme clipping (as should be expected after normalisation to 0 dbfs and +10 boost)

Now maybe this is a "feature", I don't know much about audio, but for me it's a big problem, because I need to know when the audio is clipping without always going back to the edit tab. Also I need to use the Limiter in Fairlight to make sure, the audio is not clipping while tweaking the compressor to make the LUFS louder. This doesn't seem to work either:

1.-5. the same
6. Open dynamics of A1
7. Enable Limiter with Threshold at -1
8. Increase Make Up to +20
9. Play (Limiter works fine)
10. Switch back to Edit Tab
11. Play
12. Extreme clipping (as if the limiter isn't applied)

+1 This!

For info, I'm sure I read somewhere else on this forum that the Fairlight limiter (step 7 on your list) is a "soft" limiter so effectively applies some really high-level, but not infinite compression and will allow the audio level to peak above "the limit" when really hit hard. In other words, what you want is a hard limiter to do a -1 dB level, with no possibility of ever going over this peak.

I'm off to read the manual to see if there is some "hidden" post master fader limiter in the Fairlight... options

Not a bug as far as I'm concerned as I do not see this and never have. No DAW has a limiter that is always in the signal path. I believe Fairlight uses 32bit floating point sampling so it would not be needed anyway. Also the last thing you should do is normalise to 0 dbfs. That leaves no headroom at all. 0dbfs is not 0vu. 0vu is usually -18dbfs in DAWs, 0dbfs is therefore +18db above 0vu. What you are doing is overloading the digital to analogue converters of your soundcard hence the distortion.

Not a bug as far as I'm concerned as I do not see this and never have. No DAW has a limiter that is always in the signal path. I believe Fairlight uses 32bit floating point sampling so it would not be needed anyway. Also the last thing you should do is normalise to 0 dbfs. That leaves no headroom at all. 0dbfs is not 0vu. 0vu is usually -18dbfs in DAWs, 0dbfs is therefore +18db above 0vu. What you are doing is overloading the digital to analogue converters of your soundcard hence the distortion.

I'll double-check my D/As on my Focusrite 6i6, but I'm fairly certain I'm not overloading them in the Fairlight page as it wasn't pushing anywhere near the audible volume though my active studio monitors that be indicative of a hot enough signal to clip digitally in the Focusrite. When it comes to audio I'm not a noob, but I'll admit that I didn't check the meters in Mixcontrol to confirm the final output level for the Focusrite (which is a proper rookie error )

Just out of interest, why do you think that 32-bit FP would not need a limiter - 32-bit FP just means more signal headroom overall, but it is still perfectly possible and often seen as desirable in "Pop" to drive to near 0dBfs in the DAW, "slam the limiter", drive the flattened signal even hotter to just below true 0dBfs and "win the loudness wars" - I should add that a) I really hate the loudness wars as it leads to a loss of the thing that makes great music (i.e. dynamic range), and b) I'm glad that most of that movement has past now ).

OK, now I can't remember what my point was... Oh, yeah, I agree that there is absolutely no reason to have a limiter in the signal path, but the way Fairlight is acting suggests there may be?! It's either that, or Fairlight is using an entirely different audio engine/reference output than the rest of Resolve, which would be a strange decision as it would be (and clearly is) very difficult to reconcile what is "heard" in Fairlight and what is "heard" in the Edit and Deliver pages, but perhaps that is what you are meaning by having a DAW (Fairlight) -18dBfs indicated as 0dBfs vs. a true (Edit/Deliver) 0dBfs.

If that is the case, would you know of any way to either switch all parts of Resolve to -18dBfs monitoring (so it is at least consistent!), or switch Fairlight to 0dBfs monitoring (so it would hard-clip when the rest of the program does) as I couldn't find anything obvious in the manual?

I guess until I can work out how to do that, I'm going to have to rely a whole lot more on visual metering to -18dBfs on the master fader level on the Fairlight page!

As I said, I've had no problems with audio through all the versions from 14 up to 15.1.2. As to having the signal at or near 0dbfs, Vu metering ballistics cannot show transients that go over. You can still be clipping even if the clip indicators do no register. 32 bit floating point should provide plenty of headroom but the weak link in the chain are the analogue sections of your audio interface will have trouble coping with higher than designed levels by distorting.
As to the loudness wars, that tended to apply to music. Video soundtracks for broadcast comply with the R128 standard, which like film is mixed on calibrated monitoring. No there are no separate settings for each page. I do not mean that -18dbfs is 0dbfs. -18 on the dbfs scale is usually chosen as a reference point and is eqivalent to 0vu on a VU meter. 0dbfs is therefore +18vu. I think you are confusing Volume Units with Decibels Full Scale.
When it comes to audio I'm not a noob either. Audio professional since 1970, Pro Tools user since 1994.

Thanks Charles,

I only normalized to 0dbfs for the experiment.
(Usually I normalize to -4, which in this case would not solve the problem)

The point is the audio doesn't behave the same way in the Fairlight Tab as in the Edit/Deliver Tab.
If there's actually a limiter in the chain, I don't know of course, but that's how it sounds.

Can you reproduce this test with your system?

I'll try 15.2. but I haven't seen this addressed in the changelog.

Same here. These are examples of what I'm experiencing, with a very simple clip with compression in A1 and some limiter in Master:

No clipping (Fairlight tab): https://drive.google.com/file/d/19YgWxP ... sp=sharing Notice the last word, "ahora" ("now").

Clipping (Edit tab): https://drive.google.com/file/d/1hK3N5z ... sp=sharing

Audio peaks in Fairlight tab (taken from Audacity):


Audio peaks in Edit tab:


As we can see, audio definitely doesn't behave the same way in Fairlight as in Edit tab or export and this is obviously not a desirable behavior.

But there's more... When you enable "dim", audio plays smoothly in Edit page, but the clipping persists in export.

To me, honestly, this is the hugest BUG I've seen in DaVinci so far.

Same here. These are examples of what I'm experiencing, with a very simple clip with compression in A1 and some limiter in Master:

No clipping (Fairlight tab): https://drive.google.com/file/d/19YgWxP ... sp=sharing Notice the last word, "ahora" ("now").

Clipping (Edit tab): https://drive.google.com/file/d/1hK3N5z ... sp=sharing

Audio peaks in Fairlight tab (taken from Audacity):


Audio peaks in Edit tab:


As we can see, audio definitely doesn't behave the same way in Fairlight as in Edit tab or export and this is obviously not a desirable behavior.

But there's more... When you enable "dim", audio plays smoothly in Edit page, but the clipping persists in export.

To me, honestly, this is the hugest BUG I've seen in DaVinci so far.

+1

Here is the same music track in both the Edit page and Fairlight page. It plays at exactly the same level in either page without clipping. No audio processing applied. This has been the case with all versions of Resolve that I have used from all the betas of 14 and 15 plus the release versions up to 15.2.2.

Testing on my systems confirms, that playout via the Edit and Fairlight pages to a Stem recorder (in this case, a Tascam HS-2 unit), produces identical broadcast wave files to the stereo 24bit/48kHz source broadcast wave file on the Resolve Timeline, confirmed via null tests.

An Audio-Only Export via the Deliver page produces a deliverable identical to the source, again, confirmed via a null test.

Comparison.jpg (393.4 KiB) Viewed 19883 times

For clarity, both broadcast wave files produced via playout from the Edit and Fairlight pages to the Stem recorder, contain a single frame lead-in of digital black, not present in the source or deliver page files, hence the minimum RMS level readings reflect as negative infinity on these two files.

Testing on my systems confirms, that playout via the Edit and Fairlight pages to a Stem recorder (in this case, a Tascam HS-2 unit), produces identical broadcast wave files to the stereo 24bit/48kHz source broadcast wave file on the Resolve Timeline, confirmed via null tests.

An Audio-Only Export via the Deliver page produces a deliverable identical to the source, again, confirmed via a null test.

Comparison.jpg

For clarity, both broadcast wave files produced via playout from the Edit and Fairlight pages to the Stem recorder, contain a single frame lead-in of digital black, not present in the source or deliver page files, hence the minimum RMS level readings reflect as negative infinity on these two files.

But you didn't boost the audio in the fairlight page, did you? Try adding a limiter with make-up or just increase the volume slider and then compare the files. (Of course the source audio will be different in this case)

But you didn't boost the audio in the fairlight page, did you? Try adding a limiter with make-up or just increase the volume slider and then compare the files. (Of course the source audio will be different in this case)

With the Fairlight Channel Dynamics Processor enabled and with significant Makeup applied (set to a value of 16.1), M1 Buss metering is pegged at 0dBFS and hovering around +4dBTP on dedicated RTW metering. No Peak Limiter inserted on the M1 Buss.

Playback is identical on both the Edit and Fairlight pages, with no audible clipping.

Increasing Clip Gain significantly (12dB), again, M1 Buss metering is pegged at 0dBFS and hovering around +5dBTP on dedicated RTW metering. No Peak Limiter inserted on the M1 Buss.

Playback is identical on both the Edit and Fairlight pages, with no audible clipping.

In both examples, playout via the Edit page and the Fairlight pages to Stem recorders, and creating a deliverable via the Deliver page produces identical 24bit/48kHz broadcast wave files from the same 32bit/48kHz Resolve Timeline, except for a slight variance in DC Offset, and differences in minimum RMS values between files.

Comparison P2.jpg (296.26 KiB) Viewed 19818 times

Obviously, in post workflows, peak levels and integrated loudness values that high will never be reached, and these examples would immediately fail QC.

Stems produced via the Edit or Fairlight pages to Stem recorders, always null with deliverables created via the Deliver page, and Resolve produced deliverables always pass QC when conforming to delivery specs.

_

Thank you! Interesting.

... with no audible clipping.

Any idea why there's no audible clipping? I mean there should be when the audio is indeed clipping.

Any idea why there's no audible clipping? I mean there should be when the audio is indeed clipping.

Resolve Timelines are 32bit float (1-bit sign, 8-bit exponent, and 23-bit fraction, with an implied MSB of 1), with the 8-bit exponent providing a potential dynamic range in excess of 1500dB, assuming one remains within the 32bit float environment (and when exceeding Full Scale Digital, compensating attenuation is applied before converting to 16bit or 24bit fixed point).

Interpolating DACs are able to compensate (to some extent) for inter-sample peaks, providing a small safety margin, by way of the reconstruction filter when converting periodic sample levels back into a discrete signal.

Any idea why there's no audible clipping? I mean there should be when the audio is indeed clipping.

Resolve Timelines are 32bit float (1-bit sign, 8-bit exponent, and 23-bit fraction, with an implied MSB of 1), with the 8-bit exponent providing a potential dynamic range in excess of 1500dB, assuming one remains within the 32bit float environment (and when exceeding Full Scale Digital, compensating attenuation is applied before converting to 16bit or 24bit fixed point).

Interpolating DACs are able to compensate (to some extent) for inter-sample peaks, providing a small safety margin, by way of the reconstruction filter when converting periodic sample levels back into a discrete signal.

Ok so it seems like this bug is only present on some systems. I just tried it again, the audio is clipping in the edit page, but not in fairlight.

Any idea why this annoying bug still persists? Here are two more extreme examples, with compressor and limiter applied:

Audio in Fairlight page: https://drive.google.com/file/d/1z3moxK ... sp=sharing

Audio in Edit page or the final video: https://drive.google.com/file/d/1sbgxqm ... sp=sharing

Notice the difference. Audio outside Fairlgiht is just unusable.

Audio was recorded with a Zoom H1. Here are the details (taken from Foobar2000):

Sample rate: 48000 Hz
Channels: 2
Bits per sample: 24
Bitrate: 2304 kbps
Codec: PCM
Encoding: lossless

What's tricky with this is that it seems to be a Windows-only problem and isolated only to some Windows systems at that.

[Peter Chamberlain]

Álvaro, I can’t see this.. please save and export a drp of the project that has the issue you see.
Go to help menu, create a diagnostic log.
Also in Windows create a NFO.
Detail in a note the external audio hardware you are using.

Zip them all and place on shareable storage, like drops and post a link here, or PM me.
Thanks

Same thing. At Fairlight Page Master is not clipping, whatever I do to make it clipping.

I increase all audio track levels and master level at max to make everything clip. But no clipping. Just clearly audible limiter with long release time.

When I switch to Edit Page, there are tons of clipping, as it should.


Windows 10 1809
Resolve 16b2

I have just started to use Davinci (on my Macbook Pro 15' 2017), and I've already found 2 critical bugs: this audio one, and the other with caching Fusion clips which reports they have been cached fine (blue line on top), but still they cannot playback smoothly. What are those Quality control guys at Davinci doing?!?!?

Álvaro, I can’t see this.. please save and export a drp of the project that has the issue you see.
Go to help menu, create a diagnostic log.
Also in Windows create a NFO.
Detail in a note the external audio hardware you are using.

Zip them all and place on shareable storage, like drops and post a link here, or PM me.
Thanks

OMG, I didn't read this message, sorry! Find attached another project with this bug, made in DR 16 b5 with a different audio recorder this time (Tascam DR-60D MKII):

https://drive.google.com/drive/folders/ ... sp=sharing

As you can hear in the video (sorry for the loud volume) audio is extremely clipped in Edit page and perfectly compressed in Fairlight page. For an audiophile like me, this is by far the most annoying bug in DR.

Tried opening your .drp file but need the video or audio it relates too.

Álvaro, I can’t see this.. please save and export a drp of the project that has the issue you see.
Go to help menu, create a diagnostic log.
Also in Windows create a NFO.
Detail in a note the external audio hardware you are using.

Zip them all and place on shareable storage, like drops and post a link here, or PM me.
Thanks

OMG, I didn't read this message, sorry! Find attached another project with this bug, made in DR 16 b5 with a different audio recorder this time (Tascam DR-60D MKII):

https://drive.google.com/drive/folders/ ... sp=sharing

As you can hear in the video (sorry for the loud volume) audio is extremely clipped in Edit page and perfectly compressed in Fairlight page. For an audiophile like me, this is by far the most annoying bug in DR.

Link opens my Google drive only.

Kalpox, I have uploaded a .drp and its' associated media to Dropbox. It has the finished audio track on it. On my system audio is the same in any page with no distortion. Give it a try. Find the file folder here.https://www.dropbox.com/sh/dg2bolm52q8n9ll/AABlS-TzLYkQLNUcPwE2e_w0a?dl=0

Álvaro, I can’t see this.. please save and export a drp of the project that has the issue you see.
Go to help menu, create a diagnostic log.
Also in Windows create a NFO.
Detail in a note the external audio hardware you are using.

Zip them all and place on shareable storage, like drops and post a link here, or PM me.
Thanks

OMG, I didn't read this message, sorry! Find attached another project with this bug, made in DR 16 b5 with a different audio recorder this time (Tascam DR-60D MKII):

https://drive.google.com/drive/folders/ ... sp=sharing

As you can hear in the video (sorry for the loud volume) audio is extremely clipped in Edit page and perfectly compressed in Fairlight page. For an audiophile like me, this is by far the most annoying bug in DR.

Link opens my Google drive only.

You can download the folder and all the files related to the project (as DR exported).

Kalpox, I have uploaded a .drp and its' associated media to Dropbox. It has the finished audio track on it. On my system audio is the same in any page with no distortion. Give it a try. Find the file folder here.https://www.dropbox.com/sh/dg2bolm52q8n9ll/AABlS-TzLYkQLNUcPwE2e_w0a?dl=0

Try applying a compressor and then lifting the volume of the audio clip to an absurd level (like I did in my example). In my case, in Fairlight page the volume is auto compressed as it should be, but in Edit page and in the final video it isn't.

Edit: latest DR 16 beta 6 update doesn't solve the problem.

Now why would I want to do that? The audio is at the correct level already. For what reason did you decide to turn up the level to "an absurd level" to induce clipping?

Not sure of the details in Fairlight but most DAWs will process the audio using 32bit floating point, so inside the program it will have lots of headroom to deal with a badly setup signal chain of excessive levels and should be able to deal with it quite well under normal mixing practices.

As soon as this 32bit floating point is rendered or converted into a fixed bit rate something has to give and that is the audio quality especially if it's been push to the max. I imagine the edit pages is showing this result and I would expect the audio is no longer at a 32bit resolution as the edit page is built to use more resources to give good visuals and not really to fine-tune the audio.

If you are mixing to one of the LUFS stands and still getting this strange results I would be concerned. I haven't had a chance to do any testing but will do when I get a chance. My approach would be to insert a loudness metre on the master output in Fairlight mix down to 23 LUFS. Export and then import the audio in to a DAW read the loudness levels using the same loudness plug-in.

Now why would I want to do that? The audio is at the correct level already. For what reason did you decide to turn up the level to "an absurd level" to induce clipping?

Just to see how it bypasses compression! You can see what it does in the video I attached in the folder (the one that starts with 2019...). Audio compressor works fine just in Fairlight reducing the volume, no matter how much you increase levels (this is what a compressor should do).

I tried in my laptop in Windows 10 too with the latest DR15 stable version and the same error occurs, so I'm totally sure it's a bug in DR!

Not sure of the details in Fairlight but most DAWs will process the audio using 32bit floating point, so inside the program it will have lots of headroom to deal with a badly setup signal chain of excessive levels and should be able to deal with it quite well under normal mixing practices.

As soon as this 32bit floating point is rendered or converted into a fixed bit rate something has to give and that is the audio quality especially if it's been push to the max. I imagine the edit pages is showing this result and I would expect the audio is no longer at a 32bit resolution as the edit page is built to use more resources to give good visuals and not really to fine-tune the audio.

If you are mixing to one of the LUFS stands and still getting this strange results I would be concerned. I haven't had a chance to do any testing but will do when I get a chance. My approach would be to insert a loudness metre on the master output in Fairlight mix down to 23 LUFS. Export and then import the audio in to a DAW read the loudness levels using the same loudness plug-in.

It totally makes sense, but just in case the final video applies levels like in the Fairlight page, not in the Edit page (this is what it currently does). Otherwise, what's the point of Fairlight if I don't have an accurate tool to do something simple like measuring levels? Loudness metre is set to -23 LUFS by default, is that what you mean?

I swear I tried everything in settings, properties and Fairlight with no success. There's just one workaround, which is reducing audio compression levels blindly in Fairlight page until it sounds good in the rest of the program, but because the compression doesn't work (I bet my savings that IT DOESN'T work, it's disabled outside Fairlight) it might clip somewhere unexpectedly.

Kalpox, I have uploaded a .drp and its' associated media to Dropbox. It has the finished audio track on it. On my system audio is the same in any page with no distortion. Give it a try. Find the file folder here.https://www.dropbox.com/sh/dg2bolm52q8n9ll/AABlS-TzLYkQLNUcPwE2e_w0a?dl=0

Try applying a compressor and then lifting the volume of the audio clip to an absurd level (like I did in my example). In my case, in Fairlight page the volume is auto compressed as it should be, but in Edit page and in the final video it isn't.

Edit: latest DR 16 beta 6 update doesn't solve the problem.

Just done my own testing on this and yes Fairlight and the edit page do play out different audio levels. The Edit page will clip the audio above 0 dBFS and the fairlight page will not.

This is not a fault with the software, this is by design. Like i said above its to do with Fairlight using Floating Point Bit Depth and the edit page using fixed point.

This will only ever be noticeable if you were to incorrectly mix your audio above 0 dBFS and why would you do that, its totally wrong to do so.

But this can be a real problem for people who haven't set up a reasonable calibrated reference monitoring level in their audio monitoring system/room/headphones. By having the sound level setup in your room so hitting 0 dBFS would sound very load, you should have no reason to be pushing your mix up to this level at all. For more reference on this you might want to Google "Reference Monitoring Levels".

Panic Over

Kalpox, I have uploaded a .drp and its' associated media to Dropbox. It has the finished audio track on it. On my system audio is the same in any page with no distortion. Give it a try. Find the file folder here.https://www.dropbox.com/sh/dg2bolm52q8n9ll/AABlS-TzLYkQLNUcPwE2e_w0a?dl=0

Try applying a compressor and then lifting the volume of the audio clip to an absurd level (like I did in my example). In my case, in Fairlight page the volume is auto compressed as it should be, but in Edit page and in the final video it isn't.

Edit: latest DR 16 beta 6 update doesn't solve the problem.

Just done my own testing on this and yes Fairlight and the edit page do play out different audio levels. The Edit page will clip the audio above 0 dBFS and the fairlight page will not.

This is not a fault with the software, this is by design. Like i said above its to do with Fairlight using Floating Point Bit Depth and the edit page using fixed point.

This will only ever be noticeable if you were to incorrectly mix your audio above 0 dBFS and why would you do that, its totally wrong to do so.

But this can be a real problem for people who haven't set up a reasonable calibrated reference monitoring level in their audio monitoring system/room/headphones. By having the sound level setup in your room so hitting 0 dBFS would sound very load, you should have no reason to be pushing your mix up to this level at all. For more reference on this you might want to Google "Reference Monitoring Levels".

Panic Over

I really appreciate your comment, but honestly, do you really think DR must show different audio levels according to the page you are? Because I don't, as this is really confusing. I have many reasons to push the levels or to make sure they are under the threshold as long as the program compresses them correctly: because I like the effect of a heavily compressed sound, because I'm reckless, because I have just "half a second" of a "hit" with a very high volume and I want to make sure that won't clip (happened to me billions of times)... They should either disable Fairlight output levels or make them to be exactly the same across each and every page and the final video. It's audio, no 3D rendering!

Ever heard of gain staging?

Technically it is showing the same audio but Fairlight is able to play back a higher level but if you look at the meters they are there to warn you that you need to turn your level down below zero.

You are looking at this all wrong. As soon as you hit 0 dBFS digital Zero there is no information anymore so the waveform gets cliped. When you render it out it will sound clipped because there is no data there anymore. Rendered files all have a fixed bit depth and all digital to analogue converters have fixed bit depth too so you cant work the way you are currently trying to working unless you use the audio meters and keep away from 0 dBFS. You will still get a very loud mix working this way if that's what you require.

Resolve could introduce 32bit floating point into the edit page but it will not fix your problems as it will sound ok untill you export. The only way to fix it is to change your approach to how you mix audio. It sound a little harsh but that's how digital audio recording and mixing works.

I have taught a lot of sound engineers in the past and levels are always difficult at first as a lot of people tend to turn their studio speaker volume up and down when they are mixing, instead of setting it to one position and leaving it there and then using the faders to control the volume and the mix instead.

Actually, if the edit page and the mix page use different engines when playing back then that's not a particularly good design. There's more than one reason for why all other DAWs of value switched over from fixed point to floating point processing - not just in the summing path but for any audio playing back (save for the final output).

It is confusing and if this is by design it is a poor one. I would hope that this is simply the result of merging two programs together and finding the fastest way to do so, resulting in an 'older' fixed point audio engine now coexisting with a newer floating point one.

As for dos and donts…; Well, yeah, we shouldn't go over zero, that's true. But on the other hand signal flow can be such that we manage levels down the line rather than up front, and the occasional over in a floating point path is of no consequence whereas in fixed point path you now have to hunt it down and eliminate it. We can argue all day about gain staging and what is "correct" or "best practice", but in many scenarios we now don't worry about it when using DAWs simply because we don't have to. We manage levels any way we choose at any point we choose.

So if indeed Davinci Resolve is essentially split in such a way that I'm at one point using a fixed point engine and at another a floating point one then I absolutely think that's counterintuitive and should be addressed. In 2019 the whole path really should be float.

By the way; I haven't tested this myself and it makes little sense that this should happen. After all, if the Fairlight path is float we still need to get that back to fixed point for audio playback and render. So if signals are present that are above 0dBFS (because we're in float) then when playing back on the Fairlight page

a) they are being brought back below clipping, or
b) we still hear distortion (because all our >0dBFS samples now clip)

Since we don't hear distortion "a" must apply. This leaves two options:

1) Fairlight does it without our knowledge/invisibly - I would characterize that as bad.
2) Something else is going on, like an instantiated plugin that only plays back on the Fairlight page.

The reason #1 is bad is because we should have complete control over our audio. Having an invisible limiter or dynamics control is a bad thing. If by accident we mess up our levels we now don't hear the obvious error (clipping) but the audio is still affected by a process we're unaware of and unable to change. I don't see how that is good.

Hopefully I'm just misunderstanding something, in which case, 'sorry'.

I have just made a video to demonstrate what is happening.

The thought of being able to export something that is wrong yet not hearing it on playback before export is just... not a great thought...

The thought of being able to export something that is wrong yet not hearing it on playback before export is just... not a great thought...

You might benefit from a LM2n plug-in or something similar.

I do almost all mixing in Pro Tools with Insight 1 or 2 or Dolby Media Meter, so I have that covered.

The point still stands: We really shouldn't be able to listen to an output which then renders differently from what we hear. I don't recall this being possible in Pro Tools or Nuendo for example. It's like audio 101 that the application doesn't do stuff in the background that you're neither aware of nor in control of.

I really don't think is good design, regardless of what best practice is.

Still not able to reproduce this behaviour on macOS with either version 15 or version 16.

Playout from the Edit and Fairlight pages is identical, as are deliverables produced within Fairlight and the Deliver page. Results still match those illustrated within my previous posts in this thread.

I do almost all mixing in Pro Tools with Insight 1 or 2 or Dolby Media Meter, so I have that covered.

The point still stands: We really shouldn't be able to listen to an output which then renders differently from what we hear. I don't recall this being possible in Pro Tools or Nuendo for example. It's like audio 101 that the application doesn't do stuff in the background that you're neither aware of nor in control of.

I really don't think is good design, regardless of what best practice is.

Let's see what reply comes back from Blackmagic, hopefully they will get the same results when they try.

[Peter Chamberlain]

FYI... I’m looking into this...

I really don't think is good design, regardless of what best practice is.

It may be unavoidable. I certainly don't want Fairlight working in anything less than a 32 bit floating-point space. Yet that is rarely what we end up delivering to a client. The phenomenon users are experiencing in this thread is the result of unavoidable differences between 32 bit floating point and lesser bit depths.

To my thinking, the only possible 'solution' here is for users to know what they're doing and always use the meters, not just their ears. What else is possible?

I think as default, there should be a 1 to 1 correlation of levels between the internal audio engine and the export result. The highest floating point level should correlate to the highest integer level. But there should be options to manually set up different levels in the floating point engine that represent clipping in export, with the meters telling you what’s happening. The whole purpose of floating point audio engines is the same as working with video in your NLE/VFX app with floating point/16bit/32bit on the GPU, even though you deliver an 8bit file.

I think as default, there should be a 1 to 1 correlation of levels between the internal audio engine and the export result.

I don't think that's possible. 32 bit floating point is not the same as 24 bit fixed point.

You're asking for the audio equivalent of "no banding" in an 8 bit export coming from 12 bit RAW originals.