Hi everybody

Until recently i have been using 3.97a10 for encoding my mp3s to -V2 for the library on my computer and hava had no problems, but scince i got a deck for my car that plays mp3 wanted to increase the albums per mp3-CD ratio so i wouldn't have to switch CDs a lot in my car.

With -V2 i can cram about 6 to 8 albums per CD and wanted a little more, so i tried -V3 and -V4 (--vbr-new on both), and noticed that the bitrates jump pretty dramaticaly from -V2 to -V3, most songs i have in -V2 are around ~230kbps, while those same songs in -V3 are ~163kbps and in -V4 are ~151kbps.

-V3 and -V4 are pretty close on most of my stuff with bitrates like the ones above, but the jump form -V2 to -V3 is pretty big dont you think, i expected something like ~190 on -V3.

I know not a lot of you use 3.97a10, or -V3 and -V4 for that matter, but thanks for your opinion!

skelly831

Edit: I listen to metal moslty, pretty noisy and constant levels through the songs, if that helps.

That's it: -V3 and -V4 use the -Y switch, which cuts most stuff beyond 16kHz that would normally cause a big bitrate bloat. Metal usually has lots of high frequency noise, so -Y can reduce the bitrate dramatically.
BTW, 230 kbps for -V2 are not normal, target is ~192. If you use -V2 -Y you will for sure also see a dramatic bitrate drop, the result will probably be not much larger than -V3.

230 isn't normal!?, most of my -V2 stuff is like that, for example, the last two Children of Bodom albums are around ~230, and Lamb of God's As the palaces burn is around ~240, am i doing something wrong?, for those rips i used alpha10 with "-V2 %s %d".

thanks again.

Metal has sfb21 issues with VBR mp3. Due to the high amount of high frequency noise which --preset standard tries to encode faithfully it bloats the bitrate a lot (search for sfb21 for more information on this). -Y disables noise shaping I believe in the 16kHz and up subband drastically reducing the bitrate but also pretty much lowpassing at 16kHz provided that the algorithm can't encode what's above 16kHz without bloat.

I believe Dibrom once said that with -Y 80% or so of 16kHz and higher material wouldn't be encoded.

For use on my computer i really dont mind large files, but if metal really has all those high frequencies, wouldnt -Y compromise quality or cause artifacts?

I for one have trouble discerning much above 16kHz in acutal music. Especially louder music like metal so for me not really. Listen and ABX some files and see if this is the case for you.

While we've been discussing this, i've been doing some thest encodes with alpha10 -V2 -Y, and yes, samples that were ~230kbps before, are now ~170kbps!

this is interesting, but do -V3 and -V4 use -Y only in --vbr-new mode or always?
and for my original purpose wich sould be better, -V2 -Y or -V3 --vbr-new?, couse now they're actually very close, like dreamtactix291 said they'd be.

I did not mean that something is wrong with your high bitrates, just that with other musical genres -V2 is usually around 192 kbps, but 230 is definitely normal for metal.
Everything from -V3 on uses -Y, also in --vbr-new, but my (limited) tests (see here) indicate that --vbr-new encodes a lot less above 16kHz than vbr-old. Whether this is relevant is dependent on your individual HF sensitivity.

-V2 -Y seems like it'll work pretty well for you in terms of filesize. At any rate cars are a noisy environment so I think you wouldn't notice the less HF content anyhow even if you could ABX it at your PC

Thanks a lot.

If your test showed that with a10, vbr-new killed everything above 16kHz, i think my best bet is to use plain -V3, but before i do that, some listening tests have to be done by me.

Edit: by me.

Youre right, but most of my future encodes are going to end up in my PC in whatever preset i use, couse im lazy when it comes to reripping and reencoding, so loosing HF is something i would rather avoid.

speaking of bitrates for LAME 3.97a10 V2: my average is 199kbps. highest bitrate: 242kbps, lowest 175kbps.

Anyway, the 242kbps file seems pretty interessting:

Lame 3.96.1 -V2: 263kbps
Lame 3.96.1 -V2 -Y: 263kbps
Lame 3.96.1 -V0: 298kbps
Lame 3.90.3 -V2: 224kbps
Lame 3.97 a10 -V2: 242kbps

I analysed the file with mp3Gain and the song has a volume of 92dB and is not clipping...that's suprising because usally songs need at least a 1.5dB adjustment to prevent them from clipping. That made the bitrate bloat even more mysterious and therefore I did a spectrum analysis and guess what...there doesn't seem to be anything above 10khz! Most red and orange colors are below 5khz...that is really strange...from looking at the regular graph the file does not look anything like being overcompressed either...

So there's no sfb21 bloat, no 'expensive' frequencies, no overcompression and no loudness race...

Gabriel didn't really know the answer either.

Edit: I've just uploaded a sample of the track: http://www.hydrogenaudio.org/forums/index....showtopic=35279

@Jojo
i am not 'in' MP3 at any sense, but just look on spectrum of uncompressed file. it is all under 10khz too, how compressed one can be not ?

yes you are right, but that was not my point. The reason why I noted that was, because high frequencies need a lot of bandwidth (that's why it is wise to use a lowpass filter since it would create artifacts otherwise). Anyway, now taken the fact that there's not much above 10khz makes me wonder even more why LAME chose such a high bitrate for that file. Do you get me know

After doing some tests yesterday and reading the last couple of posts, i decided to go with -V3 --vbr-new on 3.97a10.

I was able to ABX (to some degree of success) the original WAV and the -V4 encodes.

I was ablo to ABX (to some degree of success) the original WAV and the -V4 -vbr-new encodes.

I noticed some artifacting in the HF on both -V4 and -V4 --vbr-new, mainly cymbals, and some guitar parts that started to sound 'shhshh'y.

I was NOT able to ABX between the original WAV and the -V3 encode.

I was NOT able to ABX between the original WAV and the -V3 --vbr-new encode.

And it was a complete failure to try to ABX the -V3 and -V3 vbr-new encodes.

So, -V3 --vbr-new it is, just becouse its faster and most samples end up in the ~163kbps range, and i figure thats as low as i can go while keeping the samples transparent to me.

thanks for everyone's help!

Out of interest (maybe?) I transcoded my FLAC collection over the weekend to mp3 using LAME 3.97a10 '-V4 --vbr-new':

~144kbps looks like the average bitrate i get with -V4 --vbr-new, i wouldnt know about the files with really low kbps couse i dont have any spoken word or stuff like that, but i guess it works for that stuff.

I just put up the details for info really; the results are nothing too unexpected. It's 21kbps off the target bit range of 3.96.1 (165 --> see here), but I haven't seen an updated table for 3.97a10 (or 3.97) yet (I guess cos it ain't stable yet?).

What I have found odd is the bitrates reported in Foobar for a few of my files. I know this isn't the forum for that discussion; has anyone had a similar problem or should I raise it as a Foobar issue? When I look at the properties of the 'offending' file, as mentioned above, it shows the 'bitrate' value set to 211kbps, but in every other program I use (Mp3Tag, AudioShell, Mr QuestionMan / Audio Identifier) the bitrate is 195kbps.

I recently bought mp3-player(1GB)
I tried several ways to make mp3 from CD and then decided to use 3.97a10 with -V4 --vbr-new.
The points are :file size ,sound quality and encoding speed.
(I could use -V2 --vbr-new,but the file size is a little bit too large for my mp3)

Number of sample files : 82.
Lowest bitrate : 92(4 files)
Highest bitrate : 176
Files over 170 bitrate : 1
Files 140-170 bitrate : 31
Files 100-140 bitrate : 41(most of them are in 120-130 bitrate)
Files under 100 bitrate: 9

Most of under 140 bitrate files play relatively simple music which are such like
:acoutic guitar solo,vocal with acoutic guitar,piano solo,or so
(some are loud ,some are quiet)

The target range of V4 is around 160bps though -Y option is set by default,isn't it?
It seems target range of V5.
In another words,
one should choose V3(not V2) for those under 130 bitrate files and can get better quality?
Does that make no sense,generally?

Umm...difficult to express.
Anyway ,it is not so big problem for me.
But I just wonder why?
The answer is maybe No14 and No 15 post,now I found that .
right?
Sorry ,I am not good at English.

thx

The V4 mode can also use 170k + at times, depending on the sample. I'd say anything between 130-170 is normal. I encoded several playlists using 3.97a10 V4 vbr new, AVG bitrate is 150k.

You could go up to V3 if you don't mind 10% increase in filesize. For a portable with limited space 10% maybe too much.

Thanks.
Yes,I can understand what you mean.

------
Lowest bitrate : 92(4 files)
Highest bitrate : 176
------

The both sourses are in the same album.
In that album,the sources which go into higher bitrate contain the bass play and the lower do not.
Other conoditions don't seem so different between them.
(vocal with acostic guitar or piano)

I just bagan to study mp3-endoding,and so I wonder why the encoder don't satisfy enough the capacity of the target bitrate.
(ex.in case V4 is selected, at least upper 130kbps in stead of 92bps)

thanks again.

Because vbr is not targetting any bitrate but a quality level. The goal is to have constant quality, whatever the bitrate.

Oh,I see.

thanks

I've been encoding some files at --alt-preset medium (NOT fast medium), with the modified 3.90.3, and found that it produces results that are very similar to -V4 --vbr-new in both 3.96.1 and 3.97a10.

In fact 3.96.1 and 3.97a10 produce almost identical filesizes and histograms with -V4 --vbr-new, while 3.90.3 produces very similar histograms but file sizes are a little larger due to a slight use of 224 and 320kbps (i saw this in EncSpot's histograms of the test files).

Does 3.90.3 use the -Y switch also as default for preset medium?

I find 3.90.3's preset medium to be a nice middle ground between 3.96.1's and 3.97a10's -V4 --vbr-new and -V3 --vbr-new modes (not using --vbr-new in 3.96.1 and a10 produces different bit rate spreads that are not similar to 3.90.3's preset medium, i find this weird, that using the new vbr mode in 3.96.1 and a10 produces bit rate spreads very similar to 3.90.3's alt preset medium, wich uses the old vbr mode).

I decided not to use fast medium because its well known that the new vbr method produces slighlty better results only in newer releases. although 3.90.3 is painfully slow i stuck with the normal preset medium setting.

The produced bitrates are around ~157kbps, again i'm using mostly metal type music for the tests. I've not done any listening tests, i will once i encode some more files.

EDIT: After doing some more encodes i found that 3.90.3 gives a little bit more breathing room in the bit rate spread, for example: encoding OSI's Memory Daydreams Lapses, 3.96.1 and a10 gave nearly identical filesizes of 6.97 and 6.98Mb respectiveley and bit rate spreads at an average bit rate of 164kbps on both files, while 3.90.3 gave a filesize of 7.47Mb and an average bit rate of 176kbps.

@skelly831: Do you find your 3.90.3 preset medium encodes have any HF artifacts (eg smeary cymbals and some guitar parts) like the ones you noticed in 3.96.1 and 3.97a V4? Earlier you said you were going to use -V3 --vbr-new. What made you go back to V4? Just curious. I've been using 3.97 aps, but I've been thinking of stepping back to V3 or V4 for space considerations. I've noticed some HF smearing in V4 vbr-new too, so I have doubts about using it. Havent ABXed anything yet though.

I went back to 3.90.3's preset medium because i remember doing some test encodes to check file sizes some time ago and finding that 3.90.3 preset medium encodes had larger size over preset medium in 3.96.1, but i didn't test quality.

Now that i am looking for quality i went back to 3.90.3 to check it out. I've only done a couple of informal tests and found that 3.90.3 preset medium encodes does have less HF smearing in the files in wich i found smearing with 3.97 -V4, im not sure but i think its because -V3 and -V4 in 3.97a10 and 3.96.1 dont encode anything over 16kH in vbr-new mode, wich is the only mode i tested. Read sTisTi's thread about the -Y switch, then it will make more sense, scince i didn't use fast mode on 3.90.3.

i think the safest options to compare are -V4 on both 3.96.1 and 3.97 without vbr-new so the -Y switch performs as it should, -V3 on both versions too with vbr-new, and 3.90.3 preset medium.

After some 25 encodes of the same file (Opeth - The Leper Affinity), i found some watery effects on 3.90.3 preset medium and on 3.97a11 and 3.96.1 -V4 (i decided to not use --vbr-new because of the -Y bug), although a bit less noticeable on 3.96.1 and 3.97a11. -V3 also showed the same effect but to a much smaller degree on both 3.96.1 and a11.

This effect was reduced nicely while testing the file at --alt-preset 160 on all 3 encoders, but still the effect was most present on the 3.97a11 encode (this time i did ABX tests on all encodes to be shure), but after searching about ABR here on HA i found a thread about a bug in low/mid bitrate 3.96.1 ABR and got suspicious, but couldnt differentiate between the 3.90.3 and 3.96.1 160kbps abr files.

Finally, i also tried 160kbps CBR, testing all three lame encoders mentioned above and iTunes mp3 encoder. i wanted to try WMP10 but it doesnt have a 160kbps preset .

3.90.3 and 3.96.1 were the same to my ears, and couldnt differentiate between these and their respective 160 ABR counterparts. 3.97a11 still had the most trouble with the watery effect. The iTunes encode worked nicely, the watery effect was less noticeable on the iTunes encode than on the 3.97a11 CBR encode and is comparable (to my ears) to both 3.90.3 and 3.96.1 CBR 160kbps.

skelly831: what would be good is if you could provide some ABX results and upload some samples (no more than 30secs [10secs is probably ok] and lossless is needed) of the audio tracks that sound 'watery' etc so that others can test too.

Will do. I'll put the test results here later and is a .WAV clip enough?, the problem section of the sample is about 10 secs long.

one problem, how do I upload a sample?

thanks!

EDIT, i know how to upload now.

check the uploads section for the sample.

thanks again.

You mean that you did 25 successive encode/decode on the same file? In this case it is not that surprising to hear artifacts, and this is hard to conclude based on this that one version or the other is better.

No. I encoded the same .WAV file with each encoder separately, using the original .WAV file for every encode.

BTW, the sample of the original .WAV is now posted.

Which settings does --alt-preset extreme use? And is it affected/improved over 3.96?

Here are the listening tests.

WAV and 3.97a11 -V4:



WAV and 3.97a11 abr 160kbps:



WAV and 3.97a11 cbr 160kbps:

Now 3.96.1.

WAV and 3.96.1 -V4:



WAV and 3.96.1 abr 160kbps:



WAV and 3.96.1 cbr 160kbps:

Now 3.90.3.

WAV and 3.90.3 --alt-preset medium:



WAV and 3.90.3 abr 160kbps:



WAV and 3.90.3 cbr 160kbps:



And finanally, WAV and iTunes cbr 160kbps:

I find that the one with the most noticeable problems are 3.90.3 preset medium, 3.97a11 -V4 and all three Lame 160kbps encodes, the three Lame 160kbps abr encodes are more or less the same with slightly better quality than the Lame cbr encodes, and the ones with the less annoying effects are 3.96.1 -V4 and iTunes cbr 160kbps.

Please do some testing yourselves and post what you think.

This isn't interesting, is it?

hi skelly. afaik, V4 (aka preset medium), even for the latest 3.97 alpha is not supposed to be indistinguishable from the original, so it should be easily ABXable. V2 is supposed to give you transparency.

have you tried ABXing the V4 encodes against each other (i.e. 3.90.3 V4, 3.96.1 V4 and 3.97 V4 in various pairs)?

also, have you tried ABXing V4 against V4 vbr-new? i've found the HF to be a bit less annoying with V4 compared to V4 vbr-new. (i have no abx tests, sorry).

i know -V4 is not supposed to be indistinguishable from the original, but have you heard the sample?, the whole song is metal (guitars, drums, etc.), but that one little piece with acoustic guitar and vocal is the only one i hear as having really audible differences from the original.

And yes, I have ABX'ed various of my encodes with each other, thats how i determined that all of the 3.90.3 encodes i did (160 abr, 160 cbr, apm and apfm) have the poorest sound.

Maybe next i'll encode the sample again using --vbr-new on the settings i used before just to try it out.

EDIT:

That's it, im not going to lose my head fussing over every little artifact i hear , i have all my music in MPC -q6 for use on my computer, no wonder i heard all those artifacts after getting used to that kind of quality. So for all my mp3 purposes i'm going to use 3.90.3 --alt-preset medium, just because i see it gives a little bit more breathing room in the bit rate spread than -V4 on the later versions, and that gives me some kind of peace of mind i guess.

Sorry for being an obsessive dumbass .

Don't be so hard on yourself. Remember, you are the best judge of what sounds good to you, so just do what you feel works best for you. Like someone here said, this is supposed to be fun, so enjoy your music collection.

regards,
kindofblue

@skelly831:

I think it's good to see people other than Guruboolez doing some ABXing and posting the results.

Also, AFAIK, 3.90.3's ap medium wasn't originally in there. The medium preset was merged later from another version. I always used --alt-preset standard -Y with 3.90.3. Try ABXing that one! The file size should be comprable with the medium preset.

edit: added the file size thing

I tried ABXing the Opeth sample with -V4 and LAME 3.97a11 and 3.96.1.

I first ABX'd WAV Vs 3.97a11 and acheived 10/10.


Then I ABX'd WAV Vs 3.96.1. I could not reliably tell the difference and gave up after 4 trials.

For me, at -V4 and on this sample, 3.96.1 was superior.

@evereux: did you use V4 --vbr-new or just plain -V4?

Just plain -V4.

Interesting.Thanks.

Hey thanks for being interested, i'll try -V2 -Y and see what happens, also i noticed that in 3.90.3, when encoding in --preset medium, the encoder reads VBR q=3 instead of 4, isn't that supposed to be like -V3?

Anyways i noticed that 3.90.3 preset medium gives on average larger filesizes than 3.97a11, and is more comparable in filesize with the results from -V4 on 3.96.1, would you guys agree with this?

EDIT:

I've encoded some files with aps -Y (3.90.3) and -V2 -Y now, i'll have to do some more ABXing to see if the quality increase over preset medium and -V4 are worth the ~2Mb file size increase.

Lame versions pre-3.94 use a different preset system, it's not directly comparable to later versions with regard to the -V value.


If you care for low filesize, give Lame 3.93.1 a try. I found it to be consistently lower bitrate than either 3.90.3 and 3.96.1 with --alt-preset medium. Don't know about quality, though.


Let us know your results Also if you prefer 3.90.3 over 3.96.1...

OK, i'll also use 3.93.1, this version uses "--preset" and "--alt-prest" instead of "-V" right?

thanks

EDIT:

where can I find 3.93.1 (recommended download source)?

Yes, you have to use either --alt-preset or --preset.


You can find it at rjamorim's reallyrarewares (link)

Hmmm... 3.93.1 is not one of the recommended versions. Either 3.96.1 or 3.90.3 (or maybe the latest alpha). I haven't tried it myself, but I think you will find 3.93.1 inferior (in most cases) to the recommended versions, qualitywise.