CBR vs VBR?
Reply #2 – 2008-01-06 10:25:17
As benski wrote VBR is smart in the sense that it uses the bitrate necessary according to the psy model (and user quality demand), and as the psy model usually is fine, VBR usually is advantegeous. But for the encoding practice this isn't the whole story. Look at the 2007 64 kbps multiformat listening test (or the 48 kbps test). Readers were astonished that WMA Pro which was used in CBR mode came out on par with its competitors. What's often overseen is that CBR's constant bitrate addresses the bitrate of the transporting frames in the audio stream. AFAIK all of the usual codecs have the possibility to put part of the audio data that logically belong to a frame into neighboring frames thus effectively yielding a variable audio data bitrate - though in a restricted way. However the way audio data bitrate is chosen is different with CBR and VBR - the CBR mode is still less smart in this respect, but also more robust when it comes to psy model and other encoder weaknesses. For mp3 it's worth mentioning that bitrate is restricted to 320 kbps anyway so even with a good VBR implementation you get diminishing returns compared to CBR the higher you go with (average) bitrate. In the end it's all up to encoder implementation. Vorbis and MPC for example are totally based on VBR, and obviously they are doing good. With AAC and mp3 usually VBR and CBR is available, in the case of Lame also ABR which means audio data bitrate control the CBR way but without the CBR restrictions (and so variable bitrate transport frames). My personal impression is for Lame (at high to very high bitrate): with current version 3.98b6 VBR mode is to be preferred, whereas it was the other way around up to 3.97.