On Wed, 3 Nov 1999, Robert Hegemann wrote:
> As many of you may have noticed, if a track fades out
> the bitrate decreases downto min_bitrate (ie. 32kbits).
> But, if that track has a few more frames with digital
> silence (zero energy), the bitrate switches up to
> max_bitrate (ie. 224kbits). For a quick fix look at
> quantice.c around line 662, there you see the following:
The correct solution is not to remove the -2. :) We want that because we
know that the psy model is often too optimistic, as it really was designed
for CBR apps, and is thus more useful for choosing how to allocate bits
among subbands then for determining how good something actually sounds.
The problem is that the psy models effectiveness for VBR selection use is
very veriable. Sometimes, it thinks there is a ton of masking when there
really isn't. Sometimes it's right on. The linear -2dB or the like really
isn't all that effective.
Perhaps we should try doing bitrate selection purely with tuned ATH only
model, then actually encode using the real psy model. This would have a
number of advantages. ATH will produce more predictiable results, and I
suspect that we're less likely to get the unpleasent 'why the hell did you
choose 32K for this *$*@$ frame?!?' situation. This also could help bring
VBR's speed upto par with CBR, no?
I figure we can probably tune a dummy ath_model not to over estimate
bitrate better then we can get the l3 model not to under estimate.
--
MP3 ENCODER mailing list ( http://geek.rcc.se/mp3encoder/ )
