Is it possible for LAME to produce higher quality at -V 0 than -b 320?
Reply #6 – 2011-07-29 19:09:52
As for which -Vx level to use is a personal matter of taste. -V5 provides a quality which is perfect to most people for nearly all the tracks they listen to the usual way. -V2 or more so -V0 provides a safety margin for critical tracks which is appreciated by sceptical listeners. With respect to today's storage capacities and prices many people can easily afford going even higher in bitrate than is done when using -V0. V0's safety margin can be further improved by using the additional switches -b 320 -F. This provides maximum data space for encoding the audio data. Without it -V0 can demand for audio data space to encode for the expected quality which can't be satisfied by the actual data space available. In this thread I showed that I found a track of regular pop music where V0's accuracy demands had to be reduced due to lacking data space for roughly 10% of the track. This isn't necessarily audible, but it's certainly nothing we should like. Using -b 320 -F improves the situation essentially, though at the expense of producing a 320 kbps data stream. This 320 kbps data stream isn't totally filled with audio data, and you can squeeze the air out of it by using Omion's fast and lossless mp3packer tool. You can improve on -b 320 -F's effect by adding a lowpass like --lowpass 17.5 or similar, or by adding -Y. Moreover Lame 3.98.4 has some switches --ns-bass/alto/treble to improve encoding quality for the bass/alto/treble region separately. As there is a little error in the implementation of these switches I created my own variant 3.98.4n which is totally identically to 3.98.4 with the exception of these switches which can be used now with full (and a bit extended) functionality. Moreover you can use it for finding out the amount of accuracy reductions. The vital question of course is: are the improvements audible? Usually not, of course, simply because usually plain -V0 is perfect. In very critical situations however the improvements are audible. I care most about tonal problems, and for two of these problems I could ABX an improvement. /mnt is most sensitive to pre-echo problems, and he was able to prove the better quality for the one track he tested. It's all about improving the safety margin, just a little step further than when going plain -V0. As for the original question: I think it's nearly impossible to decide whether this solution is better than using CBR320 (as a funny side note: 'my' setting when not followed by mp3packer is a real CBR320 encoding created by the VBR mechanism, but it's of course not the CBR320 procedure the OP was talking about). I personally beleive that this defensive VBR approach provides a better quality than CBR320 because a) there has been evidence that CBR development has been neglected for quite a while. Lame development concentrated on VBR. b) CBR's and ABR's audio data production is less intelligent than VBR's. This can be an advantage, especially with high bitrate, when the more intelligent VBR mechanism has some serious weaknesses for certain kind of music. In contrary to prior versions Lame 3.98.4 has no such serious weaknesses IMO, that's why I think Lame 3.98.4's VBR mechanism used defensively with very high quality demands provides the better overall quality than CBR320 does.