Hi all,
LAME 3.55beta is now on the web site. The downsampling/lowpass
filtering is worth trying if you are interested in low bitrates. For
example, take a stereo 44.1kHz input file:
lame --resample 16 -m m -b 24 input.wav
this will downsample to 16kHz mono, using linear resampling and a
lowpass filter. Linear resampling is not the greatest, and the
lowpass filter is implemented via the polyphase/MDCT coefficients, so
none of this is optimal. But I would think most of the noise is
aliasing, which will be mostly in the higher frequencies, and the hope
is that this noise is insignificant compared to the MP3 quantization
noise :) It sounds much better than it used to.
If you leave off the --resample, but specify a bitrate which results
in a lot of compression (more than a factor of 13), LAME will
automatically downsample for you. (like Xing and FhG). The formulas
for when to downsample and what type of lowpass filter to use surely
need a lot of work.
I also finally changed all the doubles to FLOAT8 (and all the float's
to FLOAT). So you can specify the precision in machine.h. The
default is FLOAT8=double for now, but I think the consensus is we
should move to all floats, at least on intel machines.
lame 3.55beta:
Albert Faber: updated BladeEnc.dll
Simple lowpass filter added to linear downsampling routine.
Nils Faerber: updated man page.
All floating point variables are delcared FLOAT or FLOAT8. Change the
defination of FLOAT8 in machine.h to
run at 32bit preceision.
Bug (introduced in 3.54beta) in stereo->mono downsampling fixed.
--
MP3 ENCODER mailing list ( http://geek.rcc.se/mp3encoder/ )
