Here's the clip:

http://www.hydrogenaudio.org/forums/index....ndpost&p=378421

Obviously, this has nothing to do with CBR 128 not being transparent, as no one is claiming it should be. Rather, there is a very strange artifact produced on this clip with LAME 3.97b2 (as well as 3.98a3) at -b128 that is far less noticeable with 3.96.1. With 3.97b2 at -b160, the artifact is noticeably reduced, and at -b192 it is (to my ears) completely eliminated. I am unable to reliably detect the artifact at -V5 with any of the versions.

It is somewhat similar to the artifact produced on herding_calls - not the usual warbling/swishing/fluttering sound, but that odd "scratchy" distortion. This time, it is lower in frequency, and the sample is musically quite different.

Edit: Added a second sample to the above Uploads thread, in_a_state.flac. Same type of "scratchy" distortion with 3.97b2 and again greatly reduced in 3.96.1, although 3.96.1 adds an odd little flutter at a couple points - 3.90.3 --alt-preset cbr 128 is indeed the best. Note that iTunes AAC 128 has no problems with this one, at least to my ears.

Just tried encoding this with iTunes AAC at 128 (CBR and VBR)...very interesting. Doesn't produce the "scratchy" artifact, but definitely does something weird to the stereo image. In the original clip, her voice is just to the right of center. With iTunes AAC 128, it is to the *left* of center by almost the exact same amount (at least with my headphones), and there's a bit of a "stereo widening" effect going on, sort of like SRS, with her voice sounding like it's being expanded into both channels.

ive been noticing pretty nasty quality using the following setting.

--preset cbr 128 -q0 --lowpass 15.8

Note that -q0 often causes *more* artifacts, not less, especially with 3.96.1 (it should really be viewed more as an experimental thing...I don't think any LAME dev has ever recommended that it be used in actual encoding (?)).

no, this is using 3.97b2

audioflex, does the poor quality go away when you stop using custom commandlines?

no, actually.

hm... I realized such artifact with an other song (Nas - Nastradamus). This artifact isn't there when you encode in VBR or ABR mode. CBR mode could be buggy....

Note: In my case this problem doesn't exist in CBR until Lame 3.95.1... Since 3.96 I realized more artifacts in CBR mode, so I'm only using VBR modes in these versions, for CBR encoding I use Lame 3.90.3. (just my opinion. You could test it on your sample.)

the 3.97 alpha builds sounded better IMHO.

Maybe try to use Stereo mode instead Joint Stereo.

I had also an annoying artifact with 3.97b2 AND 3.96.1. It happes in all CBR modes all the way up to 192kbps. It is in the beggining of one techno song (Scooter - Soul Train if you care) and is a strange "bubbling" (or warbling?) sound added to the bass. I must note that 3.90.3 does much better in all CBR modes, not only on this sample, and the bubbling is completely gone. I can post the sample if you want.

edit- 97b2 and 96.1 does the artifact also in ABR modes

edit2- I had a little unwanted swap in files. The Lame 3.96.1 mentioned above is actually 3.98a3, sorry for confusoin. 3.96.1 is fine.

also in ABR files? Oh I didn't know that... but in my case also 3.96.1 produces the artifact in CBR mode...

a question to the lame developers: what did you change in CBR since 3.96? And what was the sense of it when the CBR mode produces artifacts which you can easily realize? Are there maybe other advantages?

Improving quality in one thing might cause new problem samples to pop up, even though overall the quality is better. At the same time, other problem samples may no longer be problem samples at all.

The artifact is gone when using default quality (not using the -h switch). Can anyone prove this?

can -q0 really break CBR?

I used -m f to force joint stereo and than artifacts gone. -m f force joint stero to all frames, when you encode -m s you get more that artifacts, but when encode -m j you get LR 20% which is that artefacts and 80% is good clip. I recommend you when encode at <128kbs always use -m f if use 128><192 use -m j >192 always use -m s, sometimes Joint stereo at high bitrates can produce some similar artifacts.

umm....

-m f destroys stereo imaging.

audioflex: yeah, -q0 really can introduce a lot more artifacts, apparently.
-m f does not destroy stereo imaging (it forces all frames to be M/S), but it can reduce quality or inflate the bitrate because M/S can be less efficient than L/R for some frames.

Is not taht bad, better than -m j, when i encode second clip with -m j i get LR 100% same as -m s, when i forced to -m f artifacts gone, btw i encode using 3.96 and his better use -m j i get 95% JS and 5%LR, and whith 3.97 i get 80% JS , 20% LR