SoundBite: Put an end to clipping with the 32-bit float WAV file format
In this video I intentionally bounce some music at too high a level into a 24-bit WAV file. Distortion! But if I bounce at the same level to a 32-bit float WAV file, the audio is clean. Hear for yourself in this video.
Transcript
Here's some music time traveled all the way from the 1980s...
As you'll notice it peaks at 0 dBFS, which is what we used to do in the 1980s. We know better now but this is good for my example. I'm going to raise the level by 12 dB. You will hear some distortion...
But all is not lost because my digital audio workstation software, and yours, has incredible internal headroom. I can lower the master fader by 12 dB or just a little more and it sounds like this...
Perfectly fine. But what if I bounce it to a WAV file? A 24-bit WAV file doesn't have any headroom above zero dBFS. Let's try...
Now that I've bounced and re-imported the file we can audition it...
Yes, it is distorted and there's no way to fix that. The distortion is baked into the file. But let's do something different. I'll go back to the original, raise the level by 12 dB again, and bounce it. But this time to a 32-bit float file. It is still a WAV file but in 32-bit float format.
I've re-imported the file and it sounds like this...
We're still hearing that distortion but I can do something magical... I'll lower the fader by 12 dB, and it sounds like...
No distortion. This is because the 32-bit float format has similar massive headroom to your DAW so no clipping. I wouldn't say that 32-bit float is the right format for finished work. The world isn't quite ready for that yet. But as an intermediary format or sharing material with your collaborators it's great. Clipping really can be a thing of the past. Thank you for listening.
Comments on this video
You can comment on this video at YouTube
You can comment on this video at YouTube
David Mellor
David Mellor is CEO and Course Director of Audio Masterclass. David has designed courses in audio education and training since 1986 and is the publisher and principal writer of Adventures In Audio.
@Terry-nx8kg: I'm living in the dark ages and using the old redbook standard 16/44.1. I do this because I don't record any acoustic instruments with microphones. Most of my music is done with VST emulations of synths from yesteryear. Now, how much headroom does 16 bit have? BTW, I like dynamics and I don't use any compression or limiting whatsoever when I mix.
@Catinthehackmatrix: Even with PA powered speakers playing audio line in sound, I still have the need to use the preamp for louder volumes. I find that Vlc software has the best preamp, paired with EAQ. The problem sometimes is the sound cuts out and gets lower, so I may damage speakers this way.
@erwintimmerman6466: I never knew why it made any sense to use 32 bit float over 24 bit. What could possibly make any audible diference even with hundreds of stacked effects on 96 channels when the quantization errors are somewhere at the -130 dB level or so. But it simply prevents mixing errors because we want to be as hot as possible, even when at 24 bits that really doesn't make any sense whatsoever anymore. At the same time I wondered how so many audio plugins could come back from input clipping without any deterioration of the signal. Now I know! Thank you!
@j7ndominica051: Floating-point also has a massive "foot-room" down there. You have a floating SNR giving 25 bits at any level. You can repair clipping fairly well with a declipper such as Stereo Tool.
@focuspixvideoaudioservices68: Great video, thanks...
@hwccheng: Cool!!!
@cmdess: This applys to how PT uses faders and their mixer. This does not work like this in logic with the right settings in the mixer. You can bring down and entire mix to reduce gain creep. The output file bounced to 24bit from 32 doesn't distort. I did notice this issue when using cubase, that the file distorted when bounced, this video explains why, it just doesn't seem to apply to logic they are doing something else. It's really hard to get things to clip audibly, I still manage though. ha If the sound audibly clips on the fader there can also be gain issues within the channel inserts. I also tested on 16bit file, again, no clipping. If the file doesn't clip audibly in logic the bounce will not clip audibly (assuming 16/24bit), part of the magic sauce of why people got spoiled on the prog.
@ProfesorDocenteSISE-bj7be: Thanks a lot, it was so easy to understand the benefits from it.
@Zickcermacity: I won't even use float to get around clipping beyond dBfs. I'll just mix and master it to a NORMAL, SANE level, one that requires the listener to actually TURN IT UP at their end, to rock out. In other words: loudness where it should come from!
Signed,
- Lover of 'raggedy' waveforms!
https://cdn.head-fi.org/a/10130154.jpg
@AudioMasterclass replies to @Zickcermacity: Just to be clear, I didn't recommend 32-bit float for mastering. Only for recording if you have something like the Zoom F6, and for intermediate stages before mastering. DM
@Zickcermacity replies to @Zickcermacity: @@AudioMasterclass Oh ok, thanks for clarifying.
Still, as I illustrated with the image I shared, I prefer not to create 'sausage' or 'picketfence' masters. raggedy on top is all from me. Users just need to turn up their gains or volume settings.
@nosucherror: Why would you do this? Surely it's just as easy to correctly gain stage your signal flow to avoid clipping?
@AudioMasterclass replies to @nosucherror: It's another level of flexibility to use or not use as you please. You could say why should DAWs have massive internal headroom when they could peak out at 0 dBFS and you could just, as you say, correctly gain stage. Headroom in audio has always been an issue, but in field recorders we now see 32-bit float and I believe it will become the norm in future. For final output however, I'd stick with 24-bit or the loudness war could easily go nuclear.
@drkastenbrot replies to @nosucherror: @@AudioMasterclass when uploading to streaming platforms it will be truncated to 24 bit
@nosucherror replies to @nosucherror: @@AudioMasterclass Fair point, if the loudness war gets any worse I'm going back to vinyl!
@EgoShredder: Nicely explained. My recording devices are 24-bit and my best listening device handles everything including 32-bit and DSD 1-bit and even DXD.
@drkastenbrot replies to @EgoShredder: there is no such thing as a 32 bit listening device because floating point numbers make no sense in a dac. they will internally convert it to 24 bit and even 24 bit goes far into the noisefloor on the electrical level. if mastered correctly, even 16 bit is well above the human hearing range in dynamic range.
32 bit pcm is not meant to be the final export format, its more of a side result of how the daw works internally
@AudioMasterclass replies to @EgoShredder: As I said in the video at https://youtu.be/NYYTMxKRBvs?t=109, I don't recommend 32-bit for final output.
@sheltersonic5499 replies to @EgoShredder: @@AudioMasterclass why you don't recommend it?
@AudioMasterclass replies to @EgoShredder: @@sheltersonic5499 The practical reason is that as far as Google seems to know, there are no 32-bit float digital-to-analogue convertors either in pro audio or hi-fi. If there were, and they became commonplace, then the loudness war would probably go thermonuclear. 24-bit is fine for final output; even 16-bit handled properly is fully adequate for most people's needs. DM
@Illkacirma: awesome! thanks!