Hi,

Today I had made a test: wave to mp3 conversion. First I've ripped 12 min 20 sec track from cd to wave file and then encoded it under linux using lame and under windows using xing.
And I,m a bit confused, because under linux it took 1 min 50 sec to do this, and under windows only 32 sec. The bitrate was 256 kbps. I have Athlon 1 Ghz with 640 mb ram. Why there's such a big difference?? Is there any possibility to improve encoding under linux ??

Is that time just for the encoding, or for the ripping and encoding?

Only for encoding

What options did you use with lame?

There is a quality option. The default level is 5 (on a scale of 0 = best quality/slowest, 9 = worst quality/fastest).

I did some tests encoding a 4.2 MiB wav file with different quality settings. With the default quality level this took about 18 seconds. With the quality set to 9 (worst quality, fastest encoding), the file encoded in about 5 seconds. Here's how to pass the option:
To my ears with pretty naff headphones, I really can't tell a difference, but I'm no audiophile.

Of course, it could also be that lame is badly optimised, or written in a way which doesn't lend itself to performance. Also, lame is typically compiled with gcc, which isn't known for producing the most optimised programs.

It would be instructive to do some profiling of lame, and try building it with the intel compiler (which is said to produce much more optimised programs).

Having apparently criticised gcc, I'd like to make it clear that I think it's a superb project, and I feel all dirty if I use another compiler. I think the benefits of gcc's cross-platform features, and general flexibility more than outweigh the supposed lack of performance. Not to mention the most important part - that it's truly Free.

Quality setting seems to be a good solution I've encoded 120mb file with in 29 sec now, and I hear no difference either.
Tomorrow I will try to test everything with self-compiled lame

I'm doing it now... Didn't notice any difference, but I'm probably not encoding enough data to tell really.

There's no --enable-gprof, so I've been patching up the Makefiles to see if I can't speed it up a little. I probably can't.

There are some crazy function names in lame! It spends most of it's time (about 20%) in a function called L3psycho_anal_ns. Far out. It's beyond me anyhow.

But is there any possibility to disable this function ??

I don't think so - it looks like it's doing some of the heavy lifting. I think without spending some months really researching how lame is working (and the underlying ideas of how mp3 is encoded), trying to optimise the code is not going to be easy. Using the intel compiler might squeeze a few % more performance out of the old chips though, although I don't know how easy it is to drop it in as an alternative to gcc.