Hello,
I'm interested in learning the differences between the mp3 V0 and mp3 320kbps audio file types. I have a basic understanding of sampling and understand that the mp3 320kbps scheme samples at a fixed rate of 320 kbps while the V0 type samples at a varying rate, depending on the density of information at a given point in the song. I'm under the impression that the main method of compression that the V0 type uses is that it eliminates higher frequency information when the song doesn't contain high frequency content. Am I wrong about this? What else does the V0 type do that the 320kbps type doesn't?
Thanks in advance
I'm under the impression that the main method of compression that the V0 type uses is that it eliminates higher frequency information when the song doesn't contain high frequency content. Am I wrong about this?
Yes, that is not correct.
What else does the V0 type do that the 320kbps type doesn't?
It changes the bitrate based on what is present in the audio using less bits if there is less to encode. Otherwise, there is not much difference. CBR 320 files can often be losslessly repacked into roughly V0 bitrate files.
Thanks to bit reservoir native V0 files may be of higher quality than (repacked) CBR 320 files.
Thanks to bit reservoir native V0 files may be of higher quality than (repacked) CBR 320 files.
I know this is an older post.
It's my understanding that thanks to the bit reservoir, a 320k CBR file might be of higher quality than -V0 but that in practice, few would notice and that it's more correct to use -V0 and let the encoder decide because more often than the additional bits available to the reservoir being of use, the encoder will have no use for the rest of a frame and insert useless padding to keep the user (who requested 320k CBR) happy, or encode data that's not terribly important to perceived quality just because it has the room to store it.
However, this is not necessarily always the case. I found some 320k CBR files of some Adventure Time soundtracks and tried to use mp3packer on them, and most did not have any real reduction in size. The average bitrate remained at 306 kbps - 319 kbps.
I believe that there is no lowpass filter applied at 320k CBR (at least by LAME 3.98 and later) because it has plenty of data to work with and the global gain probably won't be adjusted too much just to accommodate it at that bitrate, which is probably why there's not much padding.
Seems LAME's older psy model (the one in 3.97 and older) produced 320k CBR files that used a lot of padding and were more likely to yield significant savings if you used mp3packer.
FWIW I typically use -V0 with the lowpass disabled and (to my untrained ears) for the reason described above, and can hear no difference between it and cbr320. That said, now that I'm aware of the existence of halb27's lame3995o variant, I'll have to play around with its -Q0 setting, which supposedly is superior to both -V0 and -cbr320 yet uses around 315kbps.
(Superior to -V0 because it better handles short frames followed by long frames, and better than -cbr320 because it can hold a larger bit reservoir when needed.)
I believe that there is no lowpass filter applied at 320k CBR (at least by LAME 3.98 and later) because it has plenty of data to work with
Not sure about LAME 3.98, but LAME 3.100 definitely sets a lowpass filter at 20.5 kHz for 320kbps CBR.
By comparison, for V0, that lowpass filter is set at 22.1 kHz.
It's difficult to tell which setting works best in the very high bitrate range for any kind of music.
One reason is that for many kind of music even V5 works very well. With this in mind using V0 yields identical quality as using CBR320 or my Lame3995o variant with high bitrate Q settings. You have to accept though that there are (few) exceptions.
As for the exceptions things can be in favor for the one or other variant. As for tonal problems (for instance a long-stretched voice in a song which dominates the music at that point) I can recommend my Lame3995o variant. For situations of sharp transients especially in electronic music CBR320 can be best as was shown here recently for an extreme killer sample. My variant is based on the VBR code, improves some things also for sharp transients, but basically inherits the VBR behavior. CBR works different when it comes to code details. So it's natural that it can be better or worse when it comes to specific situations.
Bitrate by the way can be maximal (more than 320 kbps) no matter VBR or CBR due to bit reservoir.
No matter these details: you can be pretty sure you can be totally satisfied when using very high bitrate what kind of ever when it comes to encoding 'normal' music.