Subject:Sound levels when using floating point data

Posted by: cyberbeat

Date:2/22/2015 2:37:38 PM

In prior versions of Sound Forge Pro there used to be an option called something like "use floating-point temporary files". In V11 there is no such option, so is it safe to assume that it uses floating point processing on opened files by default now?

If this is the case, let's say I load a 16-bit 44.1 file in Forge and I raise the volume slightly over 0 DB such as +1 or even +2 and then save it to a file type that uses floating point data like a high quality MP3 file, will I not need to worry about the audio clipping?

Subject:RE: Sound levels when using floating point data

Reply by: rraud

Date:2/22/2015 5:47:25 PM

That's not how it works. If the master section is clipped and rendered to a non-floating point file type, clipped is clipped and it is what it is. It's best to avoid clipping in the other sub-master and plug-in stages as well.
I assume VP13 is like previous SCS products.

Subject:RE: Sound levels when using floating point data

Reply by: cyberbeat

Date:2/23/2015 4:00:39 PM

I'm sorry if I wasn't clear enough in original post, but let me try to clarify.
What I'm really asking is if the original sound material is loaded in SF and not clipped in any way and the sound levels are below 0DB.
Then I use Process/Volume and increase the volume by 1 DB which would normally cause it to clip.
Now I go ahead and save that file from with-in SF to a file which supports float data. Would that not do any harm to the audio even though it goes over 0 DB because it's floating point thus it would give me much more head room before it would ever clip?

Subject:RE: Sound levels when using floating point data

Reply by: Chienworks

Date:2/23/2015 5:21:35 PM

No. Floating point gives you more resolution, not more headroom. It will still be clipped.

Subject:RE: Sound levels when using floating point data

Reply by: cyberbeat

Date:2/24/2015 1:43:46 PM

This is for files that are 100% in the digital domain, in other words they are in the final mastering stage. I'm just trying to understand why if the SF audio engine operates in a 32-bit or higher IEEE float bit depth like it used to. There used to be a section under file properties where it would indicate the actual file's properties such as 16-bit, and then processing was being done at 32-bit or would ever.

Another thing I don't get is why if you open an MP3 (or most other compressed file formats) it automatically converts it to 16-bit 44.1Khz and gives you no indication that it was an MP3 file to begin with. I guess you can manually choose a different file type to open it as, but it still makes no sense.

Subject:RE: Sound levels when using floating point data

Reply by: Chienworks

Date:2/24/2015 2:26:01 PM

MP3 files and other compressed formats are not audio.

To let that sink in, i'll say it again ... MP3 files and other compressed formats are not audio.

They are a mathematical description of a waveform analysis that can be encoded in a smaller data size than the original data. If you used RAW mode to open an MP3 file in Sound Forge without converting it the result would be a pretty good approximation of random static, with no trace whatsoever of the original sound. The data has to be decompressed into a format that represents the sound, such as .WAV, in order to be editable. And, once this has happened, there isn't much advantage in knowing what the original format was. Even if you save it back into MP3 it's still a newly created encoding and is not the original MP3 anymore. That being said, it would be nice if Sound Forge prefilled the ID3 tags when saving with the original file's information.

It's not always 44.1KHz 16 bit. The opened result will use whatever sample rate and bit depth the MP3 file was encoded with. If you open a 16KHz 8 bit MP3 file then it will be 16KHz 8 bit in Sound Forge.

Frankly, the same thing happens when you play your MP3 file. The data in the file can't be sent to the D/A converters directly as that would result in static. The data has to be decompressed into the equivalent of WAV in order to be played.

Subject:RE: Sound levels when using floating point data

Reply by: cyberbeat

Date:2/24/2015 4:48:27 PM

Chienworks, Thank you for the explanation, I kind of get what you are saying about compressed formats, but I'm still a little confused. What about the following situation.
I download a high quality MP3 from a music service, load it into Forge and the levels will be at some crazy over-the-top DB and I won't notice any or hear distortion or clipping.
Then I'll take that same file in SF and normalize it to a more reasonable level such as -0.1 DB and I still don't see or notice any clipping or anything. Can you explain why this might be the case?

Subject:RE: Sound levels when using floating point data

Reply by: Chienworks

Date:2/26/2015 10:19:09 AM

"Crazy over-the-top DB" isn't really any kind of quantifiable measurement. Is it going over 0dB? How are you determining this? If you're looking at the VU meters then it is quite likely that you'll often hit values high above the peek level meters.

"See or notice" isn't really quantifiable either. Clipping effectively causes harmonic distortion and your ear is quite good at filtering that out up to a few percent, depending on how critically you've trained it. A few clips here and there may not be noticeable, but they are still clips.

It's up to you. If you are comfortable with the clipping then have at it. If you want to avoid clipping then make sure you never go over 0dB peak. If you want it louder, turn up your amplifier's volume control. I think though that you're misunderstanding what floating point does. 0dB peak is still 0dB peak whether it's fixed point or floating point.

Subject:RE: Sound levels when using floating point data

Reply by: cyberbeat

Date:2/26/2015 7:42:03 PM

I'm talking about commercial recordings that go over 0 DB by 2, 3 or more DB. I have no control over these types of files. These are from MP3's (256-320Kbps). and are mostly dance/rap, but I've also heard other types of recordings like rock and other stuff which do the same thing and I'm not saying the clipping isn't there, but I guess it's not that noticeable.

What if I were to take these files, clean them up for my own personal use and lower the volume, and then save them back to an HQ MP3 file. I wonder how much quality (if any would be lost), or if anything would be gained by doing this?

Subject:RE: Sound levels when using floating point data

Reply by: Chienworks

Date:2/27/2015 1:01:23 PM

You must be looking at the VU meters, not the peak meters. Peaks simply cannot go over 0dB. There aren't any numbers available in the data format to represent such a thing. On the other hand, volume units can often go many decibels above the peak value, even when the peak value is at 0dB.

Simple test: create a new file in Sound Forge and insert -> synthesis, create a sine wave at 0dB. Right-mouse-button click on the meters and make sure both Peak and VU/PPM are displayed. Play the file. You'll see Peak sitting at 0dB while VU is about +15dB. Now try the same thing with a square wave at 0dB. This represents pretty much infinite clipping. Peaks are still at 0dB while VU is now around +18dB.

As far as improvement, if there really are clipped peaks then the clipped peak restoration plugin can help, some. Other than that the fact that you've decoded an MP3 file and reencoded back into MP3 may do more damage than you've undone.