Skip to main content
Topic: Multi-format comparison at "transparent" bitrates (Read 4675 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Multi-format comparison at "transparent" bitrates

I did some searching through the past listening test pages and I couldn't find anything on this topic.

 

Basically, I'm curious if anyone has done a listening test of several of the major lossy formats to compare them at the "transparent" level? What I mean by that is e.g. LAME APS is considered to be "transparent" at a nominal bitrate for mp3. But is LAME+APS as good or better as OGG, MPC, WMA etc at the same level of encoding?

This came to mind when I saw so many 128cbr tests being done on the boards. Maybe I'm missing something but for people looking for "archive" quality encoding in these formats I don't think they would use cbr 128 so to answer this question I don't think these results will help much.  I dunno it just seems from looking through the tests alot of work (and great work I might add) has been down to compare at the lower level bitrates but  I couldn't find anything for the high level stuff.  Maybe there is a reason though so please excuse my ignorance.

I think LAME 3.90.3+APS would be the proper for MP3 but what about other formats?
What can compete with it?

If it just requires somebody simply "do it". Well I'm open to ideas or how to get such a project off the ground? 

Multi-format comparison at "transparent" bitrates

Reply #1
the problem is that at such bitrates, it becomes VERY difficult, to tell the difference - because - as you said - its transparent-bitrates. Transparent means indistinguishable to the original - which of course means "unable to abx".

So, testing would only be focussed on "problem-samples". However, whats a problematic sample for one encoder, may be easier for another one - and the other way around. So, it becomes very difficult to select a pool of samples which is fair for all encoders.

Many people probably even didn't sent in their results for the latest multiformat 128k test, because they were unable to tell the difference at 128k!!! Now imagine how difficult it would be to get enough people with golden ears together for a near-200k-test.

So, i think it makes much more sense to tune each encoder seperately for such a "transparency-setting". And then compare the average-bitrates for each encoder at that setting. The one which then archieved the "transparency-setting" at the lowest average bitrate, would be best.

- Lyx
I am arrogant and I can afford it because I deliver.

Multi-format comparison at "transparent" bitrates

Reply #2
There is a lack of formal listening tests at high, supposedly transparent bitrates but you might have a look at the High Bitrate Tests thread.

Best regards,
~ Florian

Multi-format comparison at "transparent" bitrates

Reply #3
BTW:
1. It seems that it is much harder to perform high bitrate test, becasue you became tired faster - artifacts you can focus on are much harder to find...
2. It is harder to score "practically exelent" and "a bit less than exelent" 
This is IMHO, of course...
Just from my own expirience.

Multi-format comparison at "transparent" bitrates

Reply #4
Another problem is that the conducer would need hundreds of results to try to bring the error margins down, else all codecs will be tied at the 4.5-5.0 range and the results won't be significative.

Multi-format comparison at "transparent" bitrates

Reply #5
 

ok you guys bring up some great points. So as Lyx brought up, how about at their given level of transparency which encoder tends to give the lowest bitrate? I guess the question Id ask now is is there a way to find out just "generally" which encoder can say do what LAME+APS for mp3 does BUT come out with a lower bitrate?

From what people are saying it seems that at this level of encoding quality almost all the formats are the same so even choosing one for your own personal use solely depends on outside factors beyond quality. I'm thinking of terms of support/compatibility given that is the case MP3/WMA probably wins and maybe OGG/AAC following. While formats like MPC fall behind because of lack of broad range support (i.e. hardware).

So anyways what are the general bitrates for other formats than mp3 at the transparent level given say generally speaking:

MP3 using LAME+APS gives about 170-210 (~190ish)
OGG ???
MPC ???
WMA ???
any other formats?

Another question I'd like to ask then at these levels you guys say each format has its own issues and problems unique to it. I know that mp3 has a really hard time with high freq sounds and keeping a low bitrate so genres maybe like rock/metal aren't as ideal.

Is it fair to say that maybe mpc or ogg etc because they don't have these limitations is a better format for that kinda music? What I'm getting at is can one say which genres or kinds of music encode efficiently in 1 format over others?

Multi-format comparison at "transparent" bitrates

Reply #6
Transparency is in the eye of the beholder.  As far as I'm concerned, a mass testing to see what the lowest possible transparency bitrate for each codec would be an utter waste of time considering everybody's level of transparency for each codec will vary.    I would test the transparency threshold for myself and use that for myself because what works best for all may not be for you in the end.

Multi-format comparison at "transparent" bitrates

Reply #7
A way that IMHO could be feasible to test transparency for different codecs, is to make a series of listening tests for each relevant codec at escalating bitrates . At some point, a given setting will stop being statistically different from transparency and that point could be taken as the "statistical transparency threshold" of the codec for a given population. Such a scheme would not be without methodological complications, though, since it would be quite sensitive to listener "laziness" or deliberate sabotage, which would bias the threshold toward lower bitrates unless a proper design prevents or limits this to some extent. The test could could depend on ABX only or somehow incorporate ABC/HR to make it more quantitative, and would probably require a different statistical analysis.
Just an idea, what do you think about it?

Multi-format comparison at "transparent" bitrates

Reply #8
Quote
I think LAME 3.90.3+APS would be the proper for MP3 but what about other formats?
What can compete with it?

OK, I'll bite....

Musepack
iTunes
flac>fb2k>kernel streaming>audiophile 2496>magni>dt990 pro


 
SimplePortal 1.0.0 RC1 © 2008-2019