32 bits operations

[steve]

Do you speak in the Audacity's developer team name ?
To know if I have the tiniest chance to be heard.

I speak for myself, but I am a developer and I have heard you.

I may have been a bit 'blunt' in my initial response, but the facts remain the same. Perhaps I could offer some technical insight.

Internally Audacity works in 32-bit float format, and many of the shared libraries that Audacity uses are also working in 32-bit float format. This format was chosen because it offers incredibly high audio quality (much higher quality than can be achieved by audio hardware), it avoids the possibility of clipping during processing, and it can be processed efficiently. 32-bit float and 32-bit integer cannot be converted directly or completely losslessly from one to the other, so "full" 32-bit integer support would require a huge rewrite of the Audacity code, with no tangible benefits (certainly no "audible" benefit).

With modern digital signal processing (DSP) algorithms, a sample rate of 44100 Hz is "just enough" to reproduce the full audio spectrum. There are some advantages to using slightly higher sample rates (48000 Hz is usually used for video), but above about 80kHz there are disadvantages that start to outweigh the advantages. There are a couple of excellent videos about the technical side of "high definition audio" here: https://xiph.org/video/
(Xiph are the people that developed FLAC and Ogg Vorbis)

Also, an article by the same guy: https://xiph.org/~xiphmont/demo/neil-young.html

[cyrano]

If you put up examples of DSD remastering on Youtube, you're faking it.

Either you don't know what you're doing (which I doubt in this case), or you're in it for the money.

DSD is already out in professional studios as there is no interest from the general public. That doesn't mean DSD doesn't work, or doesn't sound marginally better. It's just the reality of market adoption.

Next year's hype will probably be ambisonics. Be prepared to buy new, expensive headphones, or eight identical speakers. Of course, that might be a flop too, but it what the hardware manufacturers are betting on right now.

[Gale Andrews]

Can you clarify what the feature request is? I tried importing one of the offered 32-bit FLAC files (Elgar's Pomp and Circumstance March No.1) into Audacity. The file was read as clipped at +48 dB but could be amplified back down to 0 dB (a behaviour that should only be possible in a 32-bit float file). Is this the same behaviour as you mention "the mystake comes with Audirvana understanding full 32 bits as floating 32 bits"?

When I have deamplified that FLAC file to 0 dB I don't hear a lot of difference between that and the WAV file. Both have an unpleasant flutter distortion that is not on the LP of this recording. When played in Foobar2000 or VLC, the FLAC or WAV file doesn't have so much "flutter distortion" but it also sounds duller and more generalised. This could be because resampling is going on in Windows in those applications, I don't know.

As for exporting 32-bit FLAC, Audacity could probably do that if you export using (external program) and you have a binary EXE encoder you could point Audacity to. Do you have such an encoder? The fact that choosing "FLAC files" to export has no 32-bit option is because libFLAC does not implement 32-bit encoding (and doesn't support decoding 32-bit FLAC files either).


Gale

[Adriatika]

Hello, thanks for your interest in my question.

Well I'm not sound tech, but talking with you could be interesting for Guillaume himself, I think. May I ask him ?

[Gale Andrews]

Hello, thanks for your interest in my question.

Well I'm not sound tech, but talking with you could be interesting for Guillaume himself, I think. May I ask him ?

I don't know who you are addressing, but I am usually the person who records feature requests, and without a clear answer to http://forum.audacityteam.org/viewtopic ... 63#p306863 no feature request will even be recorded.

If Guillaume wants to and can speak good English, it may be a good idea if he registers on Audacity Forum and posts in this topic. Or you can ask him yourself if there is an executable encoder Audacity can point to.


Gale

[Gale Andrews]

I've just test another. This time "RM v CDCHET BAKER BRECKER SCOFIELD Un Poco Loc"

Where did you test it, Steve. In Audacity?

Gale

[steve]

Where did you test it, Steve. In Audacity?

Yes, and I had to de-amplify the FLAC file by minus forty-something dB, as you did. Strangely, MediaInfo reports the FLAC files as 32-bit integer, so there's something rather strange about why they are over 0 dB. I've tried making some 32-bit FLAC files myself (using "flake" encoder) and they came out with a normal 0 dB range.

[Gale Andrews]

I've tried making some 32-bit FLAC files myself (using "flake" encoder) and they came out with a normal 0 dB range.

Hmm, the Windows binary from http://flake-enc.sourceforge.net/download.html insists on downconverting to 16-bit:

Code: Select all

S:\flake-0.11-win32\flake-0.11>flake 44.wav -o flake32bit.flac

Flake: FLAC audio encoder
version 0.11
(c) 2006-2007 Justin Ruggles

block time: 105ms
variable block size: none
prediction type: levinson-durbin
prediction order: 1,8
partition order: 0,6
order method: estimate
stereo method: mid-side
header padding: 4096

input file:  "44.wav"
output file: "flake32bit.flac"
Signed 32-bit 44100 Hz stereo
WARNING! converting to 16-bit (not lossless)
samples: 12017211 (4m32.499s)
block size: 4608
progress: 100% | ratio: 0.282 | bitrate: 796.9 kbps | bytes: 27111832

It does the same if used with (external program) in Audacity.

Gale

[steve]

Hmm, the Windows binary from http://flake-enc.sourceforge.net/download.html insists on downconverting to 16-bit:

Yes, you're right. I was messing about with several different encoders (libflac, ffmpeg, avconv, sox Foobar 2000 ...), and lots of different settings, but on closer inspection the highest I'm getting is 24-bit FLAC (which is more than enough anyway).

[steve]

The MediaInfo reports are weird:

Code: Select all

Format                                   : FLAC
Format/Info                              : Free Lossless Audio Codec
File size                                : 83.5 MiB
Duration                                 : 4mn 24s
Overall bit rate mode                    : Variable
Overall bit rate                         : 2 649 Kbps
Comment                                  : RM Remaster
CODE                                     : June 2016   11 2 210 3 
DEVELOPER                                : G Chalaron
PHONE                                    : +33781380369
RM SOUND                                 : 32bit/176kHz Flac
WEB                                      : vinylizedcdsound.com

Audio
Format                                   : FLAC
Format/Info                              : Free Lossless Audio Codec
Duration                                 : 4mn 24s
Bit rate mode                            : Variable
Bit rate                                 : 2 643 Kbps
Channel(s)                               : 2 channels
Sampling rate                            : 176.4 KHz
Bit depth                                : 32 bits
Stream size                              : 83.3 MiB (100%)
Writing library                          : libFLAC 1.3.1 (UTC 2014-11-25)

but libFLAC does not support 32-bit