[LAU] Pro Audio? OT rant.

Fons Adriaensen fons at linuxaudio.org
Sun Dec 23 21:35:17 UTC 2012

On Sun, Dec 23, 2012 at 12:30:37PM -0500, Thomas Vecchione wrote:

> So I refer you right back to exactly what I said, though, which adds more
> noise?  Recording at 44.1 and not doing a SRC (Assuming that the audio
> interface of course handles it as well as 48k) or recording at 48k and
> downconverting to 44.1?  This is an honest question as I am assuming you
> know more about this than I do.  How does this carry through when dealing
> with 40-80 tracks of material, all undergoing said SRC before being mixed,
> normalized, and dithered down from there?

Why would you convert all of them ? Just do the entire production 
in 48 kHz, and convert the final result, e.g. in Ardour's export.

Not that converting all tracks separately would make any difference,
it's just a waste of time :-)

You seem to think that resampling is some very complex operation
that involves compromises and deliberate loss of quality, just as
e.g. mp3 or ogg encoding. That is not the case. As I wrote, it's
just a filter. It's a bit more difficult to write than e.g simple
EQ because the coefficients are changing for each sample. But in
the end the DSP operation that is done is exactly the same as for
a simple FIR filter, and any loss of quality will be similar as

Which in practice means it's *irrelevant*. Your 24 bit DACs may
provide 145 dB S/N in theory, in practice even the most expensive
ones don't even have 120 dB. The limit is the analog electronics,
and it is not imposed by technology but by physics. Even with the
best 24-bit converters you can buy at least the lower 4 bits will
be pure noise. And for a 32-bit converter (snake oil if there ever
was any), the lower 12 bits will be noise. That's assuming you have
a high level line input. For mic signals the actual S/N ratio will
even be lower, much lower in most cases.

Any noise added by a resampling filter or any other DSP operation
done in floating point is completely irrelevant in that context.
As are considerations about distortion etc. The processing done
by e.g. the FFT based filter in Jamin is orders of magnitude more
complex than what a resampler is doing. Even the best dynamic
processors will generate distortion at -100 dB or so, in most cases
a lot more. Any delay-based effects using cubic interpolation will
add a thousand times more degradation than resampling ever will.

In short, quality loss due to resampling is a non-issue, unless it
is done badly. Libsamplerate and zita-resampler will do it right.



A world of exhaustive, reliable metadata would be an utopia.
It's also a pipe-dream, founded on self-delusion, nerd hubris
and hysterically inflated market opportunities. (Cory Doctorow)

More information about the Linux-audio-user mailing list