On 6 September 2012 14:28, Steve the Fiddle <stevethefiddle@...> wrote:
> On 5 September 2012 22:34, Gale Andrews <gale@...> wrote:
>>
>> | From Steve the Fiddle <stevethefiddle@...>
>> | Wed, 5 Sep 2012 21:51:22 +0100
>> | Subject: [Audacity-quality] shaped dither too "loud"?
>>> I think that the main problems are:
>>>
>>> 1) The main dip in dither noise is a narrow notch at 4 kHz (@44.1 kHz
>>> sample rate) rather than a broader dip at 3 kHz.
>>>
>>> 2) There is a substantial increase in noise from 5 to 9.5 kHz (I
>>> believe that it is this peak that accounts for the noticeable hiss
>>> when listening through headphones)
>>>
>>> 3) The noise is generated by sample count with no consideration of
>>> sample rate, so the noise profile moves proportionally with sample
>>> rate (though hearing is ignorant of sample rate).
>>
>> OK, so at 192000 Hz (which is increasingly popular from what I see),
>> the first shaped dither notch is at 17500 Hz and the second
>> "wasted" in ultrasonic territory (54000 Hz).
>
> Yes, exactly.
> What happens if the audio is then converted to 44.1 kHz? Anything
> above 22.05 kHz is lopped off.
> Which leads to the more fundamental question, why are we dithering
> prior to export at all?
>
> As Martyn Shaw wrote back in 2006:
>
> "you should keep the best
> information you have about a sample until the last moment, and then
> apply it - this way
> you don't throw anything out until you really have to."
> (to which Richard Ash replied "Exactly.")
>
> Other sources on the Internet agree that dither should be applied only
> once, after the final mix down, so why do we dither at every
> opportunity?
I think I've answered that one myself - it's because we are working
with multiple bit formats in the same project.
Converting a high bit format to a lower one creates harmonic
distortion due to quantise errors (+ non harmonic distortion due to
aliasing of the upper harmonics). Once the distortion has been
created, dithering the result does not remove the distortion, it just
adds noise to the distortion.
To avoid problems of applying dither multiple times, it seems that if
a lower bit rate is converted to 32 bit float for processing, then it
should remain at 32 bit float until the final export, but that then
raises the question of what does the bit format shown in the track
control panel mean.
If anyone knows about this stuff and I'm talking rubbish, please tell
me to shut up :=)
Steve
> It seems to be agreed by all that dither should only be
> applied after the final mix down and only then if it involves a
> lowering of the bit format. Is there any case for applying dither
> other than on export?
>
> Steve
>
>
>
>>
>>
>>
>> Gale
>>
>>
>>> The other applications that I have tested have tended to have
>>> significantly less noise in the range 5 to 10 kHz, and in some cases
>>> greater noise below 100 Hz (which is inaudible at low volume).
>>>
>>> The trade-off with dither noise is how effectively it reduces harmonic
>>> distortion, so getting it "right" is probably not an easy task. Do we
>>> have any experts in this area?
>>>
>>> Steve
>>>
>>> On 5 September 2012 20:04, Gale Andrews <gale@...> wrote:
>>> >
>>> > | From Steve the Fiddle <stevethefiddle@...>
>>> > | Wed, 5 Sep 2012 17:49:48 +0100
>>> > | Subject: [Audacity-quality] shaped dither too "loud"?
>>> >> We periodically get complaints on the forum about Audacity adding
>>> >> "hiss". This invariably turns out to be dither noise.
>>> >> The default is "shaped" dither, which should be the least obtrusive at
>>> >> normal listening levels.
>>> >> Indeed it is pretty quiet at normal listening levels, but comparing
>>> >> with shaped dither from other programs, if applied to a 44.1 kHz
>>> >> track, it does seem to be noticeably louder than other
>>> >> implementations.
>>> >>
>>> >> Looking at the spectrum of shaped dither at 44.1 kHz, the dither noise
>>> >> is noticeably higher in the 4 to 10 kHz range than shaped dither from
>>> >> other programs.
>>> >> Examples of shaped dither, not from Audacity:
>>> >> http://shibatch.sourceforge.net/ssrc/gspi_dither.png
>>> >> http://www.tnt-audio.com/sorgenti/dither_e.html
>>> >>
>>> >> Looking at equal loudness contours
>>> >> (http://en.wikipedia.org/wiki/Equal-loudness_contour) hearing at low
>>> >> level is most sensitive at around 3 kHz, so it appears that our
>>> >> "shaped" noise at 44.1 kHz is not shaped to the right shape.
>>> >>
>>> >> The reason I keep stressing "at 44.1 kHz" is because the shape of the
>>> >> dither noise changes dramatically according to sample rate. Should it
>>> >> do that?
>>> >
>>> > There are complaints about every month or so on feedback@ also.
>>> > Many (but not all) are by people reporting bug 22 in regards to
>>> > 16-bit exports from 16-bit projects:
>>> > http://bugzilla.audacityteam.org/show_bug.cgi?id=22 .
>>> >
>>> > So is Audacity's dither noise at 3 kHz loudest at 44100 Hz sample
>>> > rate, or some other relationship?
>>> >
>>> > If I'm reading Plot Spectrum (at 44100 Hz sample rate) correctly,
>>> > our shaped dither is quietest at 4000 Hz but there is a second
>>> > notch at 12500 Hz; noise at 8000 to 12500 Hz is above the level
>>> > of that notch.
>>> >
>>> > "Triangle" dither is indeed louder at 44100 Hz than shaped dither
>>> > in the 3kHz - 4 kHz range, but I don't recall complaints about
>>> > dither noise in 1.2.6 where the default dither was triangle. I don't
>>> > think this is just that 1.2.6 did not have bug 22, because most
>>> > users will be using default 32-bit project quality.
>>> >
>>> > The reason we changed to shaped dither for High-quality conversion
>>> > and efforts made to "fix" shaped dither which was said to be "broken"
>>> > are described here:
>>> > http://audacity.238276.n2.nabble.com/Mixer-was-Re-FW-Dither-td251450.html#a251465 .
>>> >
>>> >
>>> >
>>> >
>>> > Gale
>>
>>
>> ------------------------------------------------------------------------------
>> Live Security Virtual Conference
>> Exclusive live event will cover all the ways today's security and
>> threat landscape has changed and how IT managers can respond. Discussions
>> will include endpoint security, mobile security and the latest in malware
>> threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/
>> _______________________________________________
>> Audacity-quality mailing list
>> Audacity-quality@...
>> https://lists.sourceforge.net/lists/listinfo/audacity-quality
|
