Let me try to explain how dithering works. If one needs to convert a
high-dynamic range recording into a low dynamic range digital format,
dithering can help to transmit even soft signals that would be
completely rejected by the coarse resolution of the final PCM words.
Without dithering, the soft passages of a recording would be
completely silent (the DA converter would output a unnatural costant
voltage of 0 V). If then appears a slightly louder sound event, the
DAC would suddenly procude a short rectangular voltage pulse. These
pulses will really sound "ugly". If one adds a soft white noise signal
(also referred to as dither) to the original recording before
executing the bith depth reduction (truncation), the soft signals
(that would be otherwise completely ignored) will be slightly pulled
above the critical threshold level of one LSB (least significant bit).
Because our ears (or our brain) inherently average the perceived
sounds in some way, we should also be able to detect even very soft
but more or less staionary signals (such as whistle-like bird calls)
that are burried in the (quantization) noise floor. Bye the way, this
phenomenon is also beeing exploited in many technical systems such as
GPS receivers or digital communication systems.
To illustrate the principle of dithering, I have prepared a quick
example by using the Black cap recording that I made a few weeks ago
with my new Fostex FR-2LE (thanks to Rombout de Wijs who notified me
that is was not a Garden warbler).
!!! REMINDER !!!
If you believe in the almighty advantages of the holy 24 bit format,
then the following information might change your religion and you
should stop reading here :)
.
.
.
.
.
.
.
The following example employs a 16 to 8 bit conversion in order to
make the differences more audible.
This is the original (but already digitally amplified) 16 bit recording:
http://www.avisoft.com/dither/mgr16bit.wav
I then prepared a synthetic noise signal (the "dither" signal) with a
level of about -40 dBFS (professional audio software probaly uses
other dither signal levels and different noise shapes):
http://www.avisoft.com/dither/dither.wav
This noise signal has then been added to the original 16 bit recording:
http://www.avisoft.com/dither/mgr16bit_dither_added.wav
Finally, the above 16 bit sound file has been truncated to 8 bits:
http://www.avisoft.com/dither/mgr8bit_dithered.wav
Without adding the dither noise prior to truncation, the final 8 bit
file would be like this:
http://www.avisoft.com/dither/mgr8bit.wav
(the ugly quantization noise is certainly audible at the beginning)
The effects can also be seen on the waveform representation of the
sound files:
Without dither, there would be completely silent sections:
http://www.avisoft.com/dither/mgr8bit.jpg
Dithering would produces a more natural waveform:
http://www.avisoft.com/dither/mgr8bit_dithered.jpg
So, now imagine this conversion had been done from 24 to 16 bit
(providing an increase of the final dynamic range of 48dB). I'm afraid
that no one could hear the difference=85
Regards,
Raimund
|
