Audio (playback speed) messed up with noise reduction plus large room. Export ok! [edited]

edited August 1 in Pro Support

Hello guys,

I have a composite shot, where I use the audio effects noise reduction and large room on the same audio track. When both are activated, the whole sound of this shot begins to stutter like a heavy fast tremolo and also the video playback is suddenly slow and stuttering. So I can only use one effect at the time to get a proper preview. In the exported file is everything fine. Then both audio effects work nice together.

Is this a general issue with the noise reduction?

At this point, I work around by switching noise reduction on, when I want to export the file.

Best regards

AlexB

Comments

  • Does the payback stutter in a ram preview with the audio effects applied?

    It is always possible to add enough video and/or audio effects to the point where playback can no longer occur in real time. This why we have ram preview. The full export will be fine because it is not constrained by the needs of real time playback. The export takes however long it takes. Same for ram preview.

  • Aladdin4dAladdin4d Moderator

    Noise reduction is usually processor intensive no matter what you use so it's no surprise you're having playback problems. You can do what you're already doing, turning off Noise Reduction until time to export, NormanPCN's suggestion for RAM preview or when you know you're going to need noise reduction, do it separately and import the finished result as a new audio track.

    I always transcode clips and extract the audio of to wav to files before starting a project just to make situations like this easier. Many would see that as a major hassle but I've found it pays off in the long run.

  • edited August 1

    Hello @Aladdin4d and @NormanPCN,

    Thank you guys for your answers.

    The issue occurs also in the ram preview, even if I use quarter resolution.

    When looking at the timecode, it seems to me, that the playback speed is reduced to a third. So that is why the audio stutters.

    It is not an audio problem. It is more an processing problem that results in reduced playback speed, even in ram preview.

    And I added the audio effects to a audio only track, because I exported the audio first, to enhance the sound with an external audio editor (audacity). Audacitys noise reduction is not the best, so I decided to try out hitfilms noise reduction, which is good, if you don't do to much. Even my already "denoised" audio track (voice only) sounds better with the added internal noise reduction. (Threshold 70%, Noise reduction 8db). At 100% and 24db there are to much glitches in the sound an there is this metallic coloration appearing.

    So I think my workaround is at the moment the only solution for me. Switch on noise reduction only for exporting.

    BTW: I'm editing on a Alienware 17R3 with i7-6820HK with 3.8GHz 16GB RAM Nvidia GTX980m 8GB VRAM and Windows 10 64bit from 2016. I don't know how fast Hitfilm 2017 should be with this specifications, or is there a recommendation for full resolution editing without proxies and realtime scrub through multiple composite shot layers with nested CS? :-)

    Best regards

    AlexB

  • Triem23Triem23 Moderator

    Some random observations on the topic for AlexBoris @Aladdin4d and @NormanPCN

    As Aladdin and Norman have already stated, Noise Reduction is a processor intensive task. I note that in my dedicated audio software my various NR plugins are either not considered "real time," or offer latency adjustments to account for processor lag. This is for software that's primarily audio-only, and the roughly 11MB/min of uncompressed 48k, 16-bit audio is a data flow orders of magnitude under the roughly 63MB/sec data rate that a computer has to deal with for uncompressed 1280x720 24 fps video (which means any higher resolution or framerate increases this disparity).

    Let's not forget that I'm citing the uncompressed data rates--now, digital video formats are highly compressed (as is mp3 audio) which just means the PC has additional work to decompress the data before streaming the uncompressed data.

    I'm going to guess that a RAM preview in Hitfilm is only worried about the video data. This would mean that audio in RAM preview playback still requires pulling the audio from a drive--if the audio is embedded in a video file this means the drive is still accessing the original file and still doing all the work of decompressing the audio then processing the audio while trying to maintain a real-time playback. This is where Aladdin's workflow of separating audio into it's own file makes sense. This means his computer can deal with audio and video streams totally separately, and, as we've discussed, audio data is a lot smaller and esier to process than video.

    Audio has always been the weakest aspect of Hitfilm, and I'll just assume that Hitfilm's audio code isn't the most efficient or optimized at this point.

    In short, I actually DON'T as a rule expect real-time playback of noise-reduced audio for the most part.

    Alex in terms of audio glitches and a "metallic" sound in heavily noise reduced audio--this is typical, and something you'll find with almost any Noise Reduction plug-in. To oversimplify the process, one can almost think of a Noise Reduction filter as a graphic EQ on steroids--rather than 5, 6 or the mighty 31 bands of EQ one might find in a normal graphic EQ an NR filter is going to have (literally) tens of thousands of EQ bands. A noise print is identifying specific frequencies to eliminate, but it's still just notching down those identified frequencies by the set amount. Make no mistake, this is degrading the audio file. Specific frequencies for high-midrange noise (for example) are also in the frequency range for "S" sounds. An "S" sound is a very hissy tone similar to hissy noise. Reducing noise almost always reduces "S" sounds in human speech. In fact, that's the same frequency range where most consonants sit, including "F's", "TH's," and the resonance of "P," and "T." Because of the nature of how sound waves work, there's only so far you can push a NR filter before the degredation of audio inherent in the filter stops degrading what one wants to eliminate more than what one wants to keep. This is physics and there's nothing to be done about it.

    So, all that brings us back to "shut off the NR until rendering."

    As far as the next question about a computer setup that will let you scrub timelines with multiple nested CS's without proxies and everything being silky smooth all the time? Sorry, man, you're dreaming. That sort of system isn't going to exist for Hitfilm at all, anywhere. Period.

    Other than certain high-end simulation work (the kind of thing that would be done on a supercomputer), digital video work is considered just about the most intensive thing done on computers. Above I brought up the data rate of uncompressed 1280x720 24 fps video of 63 MB/sec... Guess what? That was for RGB channels only. Add in transparency, and now we're up to 85MB/sec Internally the computer absolutely is processing this uncompressed data rate. Now--that's per layer. Every layer of media you add adds another 85 MB/sec to the processed data (because each layer has to be processed, before it can be summed to the final render), and we haven't even discussed the processing cost of scaling, keying or any actual  effects.

    If we're talking about 4k (3840x2160) video at 24 fps the data rate is now about 760MB/sec, and if it's 4k at 60fps your computer is now pushing almost 1900MB/sec. PER LAYER!

    And now you see why 4k editing in Hitfilm requires a GPU with 4GB of RAM. I think this should also explain to you why text and plane objects are limited to being 8k pixels wide. These aren't limits that are in there for no reason, these are limits designed to keep you from creating a media object that will immediately make your GPU weep and pout.

    You could upgrade to a 12-core or 18-core i9 and grab a maxed out Nvidia 1080Ti (and you'll pay over $2000 just for those two chips--now add in your mobo and the high-end RAM needed to take advantage of this and you're looking at over $3000 for this rig before adding in things like storage drives a monitor the PSU, etc, etc, etc) and you're still going to need proxies and RAM previews.

    Back around 2000 A.D.  or so SD video was still the standard. CPU speed since then has gotten about 24 times faster. Drive speed has gotten about 10 times faster. GPU's  Have gotten about a 100 times faster when comparing a Nvidia 226 to a 1080Ti. On the other hand there is 144 times the pixel data in 8k video than SD. (or 36 time the data if we're talking 4k). Now--if you're trying to deal with 36 times the pixel data (we'll stick with 4k), but your CPU has gotten 24 times faster and your drive has gotten 10 times faster, well... that means as far as those components go, video editing today is SLOWER than it was then. Fortunately the GPUs are picking up the slack, but, in general, there's not going to be a system that's going to handle complex video editing in real-time without proxies, etc. At least under video standards stop adding data faster than our ability to process it.

    Hitfilm users on this forum often say silly things about wanting to edit multiple layers of 4K video in real time on an i3 with no dedicated GPU. In the meantime I'm doing a little bit of side work with an indie studio (but one with a distribution contract with a major) on a film project, and it should be noted that we're not editing the 6K footage from the RED. All editing is being done at 1080p. So is the color grading. Only VFX shots will be worked on at full res and that will be the VFX guy who will still send a 1080p version back to editoral to go into the cut. That film will basically never have the 6K footage used until final renders. STUDIOS don't work at super-high resolutions when editing, and that's often on a really badass custom system with multiple octo-core processors and multiple GPUs stuffed into the same machine. This is the entire reason proxies and RAM previews exist.

    Last bit of trivia. Disney's "Frozen," has a single frame in the song "Let it Go," that took 178 HOURS to render, and that was on a 48 machine render farm where each machine in the farm was an octo-core with a Quadro on a custom renderer. I did say that digital imaging was the most complicated thing people do with computers, right?

  • edited August 1

    Hello @Triem23,

    Thank you for your long explanation. The mentioning of my working horse and asking for smooth working specifications was more with a ;-) than a :-) and I didn't really wanted an answer, because I know there are limitations. I only did not know where they are in detail. Now I know. :-)

    I am totaly understand, that even a high end workstation is not enough for every situation, not enough for most situations with many vfx and that smooth real time editing on full resolution 4k files without proxies is even impossible for professional systems. And I know, that my laptop is more a gaming computer, than a video editing machine. :-)

    But perhaps it is worth a hint in the manual, that audio effects are still processed in realtime, even if you use RAM-preview. And that some effects (I don't have tested other audio effects on their processing speed) lead to some strange behaviours, if you combine them. The noise reduction alone is working in real time  very well.

    That I talked about my noise reduction settings an how the standard settings sounds, was my passive aggressive approach of suggesting (;-D) perhaps lower settings like 50% Threshhold and 12db reduction as standard or starting point. 24db noise reduction leads in the cases I know often to this "metallic glitching" sound of heavy compressed mp3s. And you mentioned this too. Thats a vicious circle. The more noise there is, the more reduction you want, the more glitches you get. ;-) I think, that the noise reduction in hitflim works really good, but the standard settings are, in my opinion,  much to high.

    So I appreciate your answer and it would be nice, if Hitfilm would use it in their FAQ or perhaps in their manual (espacially the part with the NR and the real time audio rendering).

    Best regards

    AlexB

     

     

  • Unfortunately it appears that audio effects are not cached/buffered with ram preview. Only video effects.

    Large room and noise reduction both require a lot of CPU in Hitfilm. Large room seems to have a higher load than noise reduction. Individually each is pretty harsh on CPU utilization on my 4Ghz CPU. Both together is a no go.

Sign in to comment

Leave a Comment