variable bit encoding in mp3, why do people do this???

[glen]

<< CBR 320 kbit is too large of a file, and I can tell quite easily the difference between 256 kbit and CD; I can even differentiate between 320 kbit CBR and CD if I listen closely.
>>



The you have super human hearing.
Most can't.

 

[silverpig]

It's not super human, but it is pretty good.

A blind test was conducted in this thread, and I managed to get the right file, and said that the 320 Kb mp3 was second.

 

[glen]

let's do another, you post you results before we give the answers

 

[mithrandir2001]

Actually, MP3 VBR is not the perfect solution and has problems of its own. Why? Because VBR mode relies solely on the encoder's psychoacoustic model. When you encode in CBR mode, the encoder asks "how can I best preserve fidelity using this fixed allocation of bits?" It's one-dimensional. With VBR, not only do you have to allocate the bits properly, you have to determine what is the appropriate framesize for a given timeslice. It's more complex and much slower to encode. Psychoacoustic models are not perfect and sometimes VBR mode picks a framesize that is too small to maintain a given quality level. This is why some people add a "-b 112" or "-b 128" to their LAME command line: to prevent the VBR encoder from choosing too low a bitrate. I'm not saying VBR is bad; rather, I'm saying it's just not as perfect as it might seem to be.

The best mode is ABR, which is a hybrid of CBR and VBR. You specify the average bitrate you want the file to have, so filesize is more or less predictable like CBR (it's rarely dead-on exact, but very close usually). However, ABR mode uses dynamic framesizes like VBR. I've seen analyses that say that 192kbps ABR is transparent to at least 95% of the population...at least as good as 256kbps CBR at 3/4 the size.

 

[silverpig]

The answers were posted after I guessed... Note the time and date of the edit on the first post, and the time and date of my post...

 

[mithrandir2001]

From a scientific standpoint, that 5 file test didn't hold much water. To be accurate and conclusive, we'd have to run that test about 16 times - randomizing the encoding types and filenames each time - in order to satisfy statistical significance. It is entirely possible that somebody picked the right file by accident - 1 in 5, 20% chance. If somebody can pick the right file over and over again, however, then we have a strong case that the choice is real and not guessed.

That's why people run ABX tests. I've run a few files through the PC comparator and was surprised to find that differences I thought were real could not be identified double-blind. I can tolerate a lower bitrate that I thought I could because your mind can play tricks on you. If you download an 128kbps MP3 and play it, you may hear it as sounding worse than it really is because you may have an inherent grudge against 128kbps files. On the flipside, you may play back a 256kbps MP3 file and assume it's perfect when there may be artifacts you subconciously overlook. Perhaps if you played an MP3 file without knowing its bitrate you could strip away subjectivism from your aural analysis.

 

[silverpig]

I still think I did pretty good

I know for sure that one of the files was bad (red), and I was about 90% sure that the real file wasn't blue or green, but I was only about 60% sure that it was black over yellow.

But, as I've admitted, I don't notice the difference unless I'm specifically looking for one. 256 CBR is great, but I still prefer the VBR I use.

 

[yakko]

<< Please people, whining about space when you have 40+ gigs. >>

I also own 700+ CDs as well as crates of records so if I encode them to .mp3 I will need to worry about space.