I just finished my first listening test.
With this I wanted to use a moderate bitrate where mp3 should be fine except for problem samples.
I will do a very high bitrate test later.
As a high quality moderate bitrate setting -V4, -V3 or -V2 are interesting settings and I decided to use -V3.
My most important test samples for such a setting are those which aren't specifically problematic, but with which I wasn't totally satisfied on occasion even with high bitrate mp3.
The samples I used:
- Sandrine Kiberlain: La godiche
- Camille: Là où je suis née
- Smokie: Oh Carol
To make it short: I couldn't find any problem using -V3.
I continued with samples which I know they are a bit problematic for mp3, but of which I would expect of an mp3 encoder at such a quality setting to yield only a minor issue.
- Wombat's sample Birds
- herding_calls
- trumpet
- trumpet_myPrince
I abxed Birds 9/10, herding_calls 10/10, trumpet 10/10, trumpet_myPrince 8/10 - more or less as expected.
trumpet_myPrince is a minor issue, absolutely negligible. Not so clear to say so however for the other samples though the issues aren't very obvious.
I finished with samples where major issues have to be expected (and accepted) with a setting of -V3 due to mp3-intrinsic restricted temporal resolution:
I abxed castanets 7/10, no need to abx harp and eig.
I'm personally totally satisfied with castanets, and to me eig too is encoded fine with this setting keeping in mind the extreme nature of this sample.
harp40_1 is the worst encoding among this group to me, the only sample among all I've tested I'd call really annoying. A non-subtle issue is to be expected though, harpsichord music can't be encoded with mp3, moderate bitrate, and high quality expectations.
For a comparison I also tested --abr 170 with part of the samples (Birds, castanets, eig, harp40_1, herding_calls, trumpet).
The result for Birds is clearly inferior to -V3.
I abxed castanets 10/10 though to me the quality is very good.
The eig result also is more obviously non-transparent than when using -V3 though it's still remarkable quality with respect to this sample.
The harp40_1 problem is very obvious though it's clearly better than -V3 to me.
Very bad is the ABR result of herding_calls: at second ~3.8 there's a gross error in the encoding which makes me wonder whether there is a real encoder bug when using ABR. With this result I didn't want to continue testing ABR.
I just tried trumpet for finishing up as this is a sample of special meaning to me. I abxed it 9/10, it's the same quality level to me as the -V3 result.
So for a moderate bitrate setting like -V3 it's better to use VBR than ABR, but I really wonder whether there's a real bug with ABR.
-V3 however works as expected: very good results for usual tracks, with more or less slight restrictions for tracks which are problematic to Lame (and other mp3 encoders).
