Hi,

the question says it all. We know the (statistical) problem of high bitrate encodings and abx, where only a few people may abx tracks but being statistically irrelevant.

Would it be a valid method to reencode the compared files <n> times? This would probably point out specific problems of a codec, problems you cannot hear after a first encoding...

What do you think?

It would be a bad idea IMO. The best psymodel would be the most aggressive one - reducing bitrate and maintaining transparency. It objective quality would be very poor but subjective quality would be top notch. So with transcoding you might actually reveal an inferior codec to be superior. I could be wrong, but I think transcoding would complicate an already complex issue.

Trying to abx a sample at a lower bitrate makes more sense to me.

Didn't you just already point out the flaw in this method?

It would be a valid test for the transcoding capabilities of a codec... mixing (abx-like) the codecs for each reencode could give some more objective results, too... I guess. I mean not transcoding with the same codec and settings over and over again, but double-blindly choosing one of a pre-defined sets of codec+settings for the next transcoding round and then comparing it to all other transcodes of this round.

However...



You know... coming back to your initial question here... all of the lossy audio codecs I know are optimized for a lossless-to-lossy encode only, and since most of the music is available as lossless source material, there's really no need to test them for their transcoding capabilites, maybe in a near future when the music industry finally abolished publishing music in lossless quality, transcodec comparisions will be an option. So I think there's no need for testing the effects of transcoding and yes it is a bad idea and says nothing about the codecs glitches or disadvantages nor can it help improve the codec, since as I said before, the current audio codecs are and will be optimised for lossless-to-lossy encoding only.

But if you want to test their transcoding capabilities then it's the way to go, but you can't expect to get any useful clues from this for the usual lossless-to-lossy encoding performance, the only thing you can tell afterwards is how good or bad the codecs are performing when being used as in source material for a re-encode or used as an encoder for a re-encode.


Maybe doing such a public listening test would be interesting and a good idea anyway, actually I'm quite curious how bad transcoding really is. It could gives us some practical tips when we have to transcode some audio material.




Exactly!

Problems you cannot hear after the first encoding, are no problems because you cannot hear them after the first encoding. This is what lossy compression is about essentially.