AAC VBR vs AAC CBR vs AAC ABR
Reply #21 – 2010-08-13 19:20:01
It happens again and again: as soon as one of the major beliefs on HA (such as 'VBR is best') is addressed here as questionable reactions are like those here. shadowking didn't but say that there is no proof that VBR is superior at high bitrates. That's simply true because it can't be proven. There's not even even a strong evidence for it. Few years ago in a low bitrate listening test WMA professional CBR came out second best, outperforming several VBR encoders in terms of quality (which in contrary is no evidence that CBR is best). Bit reservoir helps a lot making even CBR a variable audio data procedure. Whether VBR, ABR, or CBR is best will also be a question of the encoder used. As for the OP's question concerning AAC I think nobody can give a real answer. We can only have a feeling towards it, and for most of the HA members this is probably in favor of VBR. I personally would use VBR too with Nero or Apple AAC. For the OP's question about the difference of VBR/ABR/CBR: mp3 and aac encoders separate the music stream into single 'frames' of music. Each frame corresponds to a short interval of the music of very roughly 1 msec (exact details vary with the format and the musical content). Each frame is encoded separately. The first thing to know is when talking about CBR, ABR or VBR that this addresses frame bitrate. With CBR encoding bitrate is identical for every frame. With VBR it's not necessarily so. ABR in this respect is a VBR encoding mode as frame bitrate can vary. frame bitrate is not identical with audio data bitrate because of the concept of bit reservoir. If the data space provided by the current frame bitrate is not totally needed (according to the encoder's accuracy estimation) the unused space can be used for the audio data of the next frame(s) - within certain restrictions. That's why even CBR is a variable audio data procedure. Another important aspect is the usual way audio data accuracy is controlled. This is identical for ABR and CBR (ABR just releases the restrictions imposed by the bit reservoir as well as the limited discrete steps in choosing target bitrate), but VBR is different. With VBR the psy model is used for the accuracy control, for CBR/ABR this is so to a minor degree. As a consequence VBR quality relies more on the underlying psy model having no flaws. With CBR/ABR this is so to a minor degree. Roughly speaking when the psy model used is great then VBR should be the more intelligent approach leading to more efficient encodings. High bitrate CBR/ABR on the other hand can overcome to a certain degree flaws in the psy model. Another point is experience and testing. If most experience for an encoder is with say VBR, and experience is good, this is an important point for using VBR. So I guess in case of Apple AAC the most promising way is to do it the way Apple does it with its 256 kbps iTunes store music.