Charset=Windows-125?
I'm not sure if this is a bug. I'm sorry if I am just messing up the mailing
list.
The variable bitrate modes of LAME encoder have some strange behavior.
I encode 32000 Hz stereo files at VBR3 using the best Huffman search (q0),
joint-stereo (mj), [k]eeping all freqs and 320�kBit/s frames disabled (B256).
However, almost everything applies to 44100�Hz waveforms as well. (The reason
to disable the largest frames is that they are not filled with data completely
at 32�kHz samplerate.)
(-) *VBR* encodes frames with high frequency content (sibilants, cymbals,
hihats) with low bitrate (80-128�kBit/s), while most of the remaining
frames are encoded with 160 or 192 kBit/s. The average bitrate does not
increase by a noticable amount when I increment the VBR quality. It
increased from 143�kBit/s to 148�kBit/s in one example; other cases
were similar. The bitrate decrease can be spotted using M�Pesch's
Mp3DirectCut application, which emphasizes these high freqs on its graph.
In contrast, frames with small amount of high frequencies, like old
recordings at high samplerate, or choral pieces, are encoded with more bits.
48�kHz files processed with 15.95�kHz lowpass filter are a lot bigger than
32�kHz files, both encoded with the same Lame version (3.94beta/3.95/3.96).
Lame 3.35, however, produced files of almost equal size. The average
bitrate on files with a lot of treble is low at 44100�Hz too.
Why then it is suggested to use lowpass filtering? (�The high frequency
coefficients can take up��) If CD content is lowpassed to around 16�kHz
then the samplerate should be decreased too (and no need to that damn
padding ":)")
(-) The other choice is to use *ABR* instead of VBR, which encodes the
problematic frames well (@ABR160). ABR could be used when there are no
_almost_ silent parts, such as pauses between speech phrases or fading music,
and therefore the need to save bits. The problem with ABR is that when
decoded the sound has lower amplitude. The same decrease in volume was
observed on 4 songs at 32�kHz samplerate. RMS power decreased by 0.44�dB in
these files (peaks occasionally were more or less). VBR files decode with
almost the same amplitude with difference not more than 0.02�dB using VBR_3.
I do not use the ReplayGain.
By comparing the VBR3 at 147�kBit/s and ABR160 at 165�kBit/s one can see that
ABR-encoded version is of higher quality at the problematic treble-filled frames
(ignoring the lower volume).
The songs that Lame encodes with relatively low bitrate (155�kBit/s VBR) Vorbis
encodes with higher bitrate (190�kBit/s) and vice-versa (220�kBit/s Lame VBR2
vs 160�kBit/s Vorbis). I compared an album with synthesized music (Warcraft 2)
and an album performed by a philharmonic orchestra. Both albums contained songs
with high (Orc Song 1 � 175�kBit/s) and low amounts of treble (200�kBit/s).
I still want to use Mp3 because of the high portability of this format (many
hardware players are able to decode it) and its 'fragmented' nature (no need to
manually write any _file_ headers).
Is there a workaround to the amplitude problem with ABR, or at least an
explanation of this phenomenon? Is there a way to encode high frequencies without
reduced bitrate in VBR? Increasing the minimum bitrate in VBR mode leads to
wasted bits on the _almost_ silent parts, and this is not acceptable. Saving
space on these parts is one of the main reasons why I use VBR.
Thanks.
J7N
mailto:[EMAIL PROTECTED]
_______________________________________________
mp3encoder mailing list
[EMAIL PROTECTED]
http://minnie.tuhs.org/mailman/listinfo/mp3encoder
