[Flac-dev] Idea to possibly improve flac?

[Brian Waters]

> Regarding dithering, I am not aware of many programs that do that without
> very specific user selection.  Any user savvy enough to turn on dithering
> would hopefully be paying attention well enough to avoid promoting 16-bit to
> 24-bit without noticing their mistake.

I suspect that this is sometimes done dishonestly in order to sell
hi-fi enthusiasts DVD-A's made from lower-quality source material.
After all, they do sell $35 dollar "high definition" digital coaxial
S/PDIF cables at your local Radio Shack.

And you lost me on the last paragraph there, but that's okay. Thanks
for the clarification.

- BW



On Sat, Jan 8, 2011 at 1:13 AM, Brian Willoughby <brianw at sounds.wa.com> wrote:
>
> On Jan 7, 2011, at 21:38, Brian Waters wrote:
>>>
>>> My 16-bit detector does exactly that, except that it only looks for
>>> 0x00 in the lowest 8 bits of each sample.
>>
>> What if the program that did the 16-to-24 conversion also did some
>> dithering? If I'm not mistaken, that would probably be the case if
>> they did some sample rate conversion as well (maybe they were going
>> from CD quality up to 24/96). In that case, shouldn't your program
>> look for values in the LSB that are under a certain threshold, instead
>> of just zeros?
>
> That's a very good suggestion, and would be appropriate for a more advanced,
> but fuzzy detection.
>
> My program was designed to catch the absolute simplest case of setting your
> DAW to 24-bit files when recording from a 16-bit interface.  Or, if you have
> a 24-bit digital audio interface, but feed it from a DAT.
>
> In fact, the day I wrote the program was when I was transferring an
> 18-channel live recording that I made for a band that wanted to tour Europe
> based on a live demo, where 16-channels were 24-bit A/D and the extra 2
> channels came from my old DAT via AES/EBU.  Even though I knew that the DAT
> channels couldn't possibly be full 24-bit, I still wanted to confirm that I
> would not lose anything by converting the master files to 16-bit.  I archive
> all recordings using FLAC to save space, and a 16-bit FLAC would obviously
> take less room than a 24-bit FLAC.
>
>
> Regarding dithering, I am not aware of many programs that do that without
> very specific user selection.  Any user savvy enough to turn on dithering
> would hopefully be paying attention well enough to avoid promoting 16-bit to
> 24-bit without noticing their mistake.
>
> Regarding 16/44.1 up-conversion to 24/96, I would notice that in the FFT,
> because there would be nothing but very low-level quantization noise above
> 44.1 kHz.  But this sort of thing is a manual process for me now, not
> automatic.  I imagine that it would be especially difficult to detect
> aliasing in an up-conversion, at least with software, but it's almost
> completely obvious to a human looking at the linear frequency FFT display.
>  By the way, the 44.1 kHz to 96 kHz conversion would create valid 24-bit
> samples - there would be nothing "16-bit" about them any more.  Because of
> the change in sample rate, the 16-bit values would fall in between sample
> periods, and thus the exact sample would hopefully use the full 24 bits for
> maximum accuracy.  In my opinion, all sample rate conversions should
> preserve the 24-bit results.
>
> As for your suggestion to look for values in the LSB under a certain
> threshold, I think that you have a misunderstanding about dither.  The LSB
> would always be changing in dithered material, regardless of amplitude
> thresholds.  Dither has to be applied to all samples, loud or quiet, or else
> it doesn't do its job.  In other words, dithered audio is going to have
> constant motion in the LSB.  I do think that a dithered 16-bit file stored
> in 24 bits might have some patterns, such as perhaps all 1s or all 0s, but
> there's no guarantee.  Dithering starts by adding noise, and so the
> intermediate result has valid bits everywhere.  When the dither process is
> complete, the result should be masked so that the 8 LSBs are all 0, but if
> the masking isn't done then they really could be anything.  I have noticed
> interesting patterns with bit meters, but I'm not sure whether to trust a
> bit meter that I did not write myself, as the two I have used show
> completely different pictures of the same digital audio stream.
>
> Brian Willoughby
> Sound Consulting
>
>