What AAC VBR preset do most people on this forum consider to be more or less equivalent to lame --aps quality-wise? Is normal quite as good? Does AAC have problem samles which lame copes with fine? Is Pystel 2.15 the best AAC encoder to use (I am completely new to AAC and am asking this because I am considering buying a portable "mp3 CD player" which plays AAC).

Yeah, I believe -normal can compete fairly with --aps.



Maybe. I never heard of any, though.



It's the best one available (FhG AACdemo 2.2 is theoretically better overall). And it seems it's the best of all regarding VBR, but that's just a supposition.

I've just been playing around with AAC myself, and I have to say it really doesn't sound as good as lame aps to me. I used the latest version of pystel from rarewares, EAC (for both the aps, and the aac), and Dibrom's 3.90.2 recommended comile for lame. Using -normal for AAC, and --aps for lame. I'm using the standard winamp decoder for mp3, and the one in the pystel package for AAC.

I did a blind test in winamp (ie one aac file, one mp3 in a randomised playlist) with no EQ, and playing through my hifi. The two files sounded different and the one I liked least was always the AAC. Since I haven't got much time, I only tested it on tracks from one metal album (Breaking Me Down, Halo, and Unreal from Soil - Scars). The difference I could hear was in the high frequencies, especially the cymbals which sounded kind of "smeared". I don't know how to explain this.

I know what I should have done is to compare them both to the wav and see which is closest, because maybe I am just used to the sound of aps, and therefore consider it better, but all the same this seems surprising.

I guess maybe different people mind different artifacts more or less. Maybe It's to do with the lower frequency cutoff of aps (19 kHz vs 22.05 kHz). Maybe I'll wait for a vorbis portable player rather than buy an AAC one. I seem to have trouble noticing vorbis artifacts even at fairly low q settings.

Sorry but I can properly reply with this board... Wierd quoting... And why can't you delete posts? This board software needs a few improvements...

Anyway, make sure that you have a 100% working decoder for AAC (there was a version with lots of problems), also make sure that you use WaveGain on the wav files before converting to avoid clipping... Last but not least, use an ABX application...

I am using in_faac.dll dated 20th September for decoding in winamp. Is this a buggy decoder (this would be surprising since it's the version hosted on Rarewares and Roberto is usually veru good about what's on his site). I wasn't using WaveGain, but doesn't the encoder warn you of any clipping errors during encoding? How do I use WaveGain with EAC?

I guess I should be using ABX, but I just haven't had the time (it's my final year of uni, and I really ought to be working...), but surely there's nothing wrong with a blind test in winamp (since the differences can't be psychological)? And using the original files ratehr than decoding to wav shouldn't make any difference since the decoders should be the same anyway?

I was really glad I was getting a player that supported AAC since it meant I could use a superior format to mp3 without worrying about compatibility, but now I'm not that sure I'll be using AAC rather than aps (I know aps will be overkill for the portable player, but I want to have the same files for playback on my PC as on my portable player, no re-encoding).

I saw your quote problem. The problem was that one of your quotes was spelled 'qoute', and so it screwed up the parsing for the rest of the message. Quoting works the same as vb otherwise.

"in_faad.dll" from September, 20th is the newest version, so there should be no obvious problems with this plugin, I think. I'm using in_mp4.dll from the same package for my low bitrate tests, so I can't contribute much to your problem.



They should, yes - but to be absolutely sure it would be nice if you did another test also with reconverted WAV files (I'll have to do that anyhow because of my slow PC).

What I'm missing in your comments is the resulting size of both samples, -aps and -normal ones. I would assume that the LAME file is much bigger than the AAC file, or am I wrong? So you might experience the usual problem with heavy metal music producing a lot of output in the high frequency range that is hard to encode for most of the lossy encoders.

If I'm right, you should also try the -extreme preset with AAC, so that it can use more bits for your special samples. Of course this is not the way how it should be, because AAC is meant to be more efficient than MP3, but perhaps -extreme files will still be smaller than -aps on heavy metal.



Yes, I read your comments in the forums at Doom9 and Audiocoding, and I also would like to know more about this apparent "AAC low-volume bug" of the Philips Expanium 303. But at least it should play your -aps files fine, I think.

I bought the Philips Expanium 401 (AAC-compatible mini-CD MP3 player) last year after reading this discussion about portable AAC players. While waiting for the product to arrive, I did some AAC tests and immediately ran into a trouble sample, Bullet With Butterfly Wings (by Smashing Pumpkins).

The trouble sample still hasn't been resolved, and several other songs on the Pumpkins album have nasty audible artifacts with the -normal switch. Even --alt-preset 160 doesn't glitch like Psytel AAC does on these samples.

I haven't come across a song in a collection of ~20 CD's that produced a noticeable, irritating artifact when encoded with --alt-preset standard. Contrast that with Psytel AAC, which added very distracting artifacts to several songs on the only album I've bothered to test it with.

Even though my portable player is AAC-capable, I've always used --aps MP3's with it, and I've been very happy with the results.

Edit: some of the Pumpkins problem samples are available from this thread.

Hmm, can you post more samples than only that one? It would help in tweaking the specific problem if Ivan has more examples. I'm sure HA can host it if you just can provide more samples for a short period of time. Or come to the irc channel and dcc the files for somebody.

Or does anybody else have Pumpkins albums?

From my experience, Lame --aps -Y doesn't make a file too much bigger than AACenc -normal. Then again, I haven't tried AAC with my noisy rock CD's, although Lame --aps gives me ~210kbps and --aps -Y ~175kbps for Tool and A Perfect Circle albums.

Smashing Pumpkins - Mellon Collie (Disc 1):
AACEnc v2.15 -normal: 66MB
Lame v3.90.2 --aps -Y: 75MB
Lame v3.90.2--aps: 78MB
MPC v1.1 --standard: 68MB

Cardigans - Life:
AACEnc v2.15 -normal: 58MB
Lame v3.90.2--aps -Y: 63MB
Lame v3.90.2--aps: 73MB
MPC v1.1 --standard: 62MB


With my Philips 401, I estimate that AAC files play about as quiet as MP3's that have been lowered in volume 9dB by MP3Gain. The player is still loud enough to blast my ears, but if I WAVGain my AAC sources before encoding, the player is fairly quiet even at maximum volume... maybe a bit louder than a Sony CD player with the AVLS (automatic volume limiting system?) activated, which limits the peak volume to ~85dB.

@ hans-jürgen

Yes the aps files were much bigger (one of them was 264 kbit, which is the biggest I've ever seen aps produce) so this music is "hard to encode". One thing I noticed about AAC -normal is that the bitrate changes much less from song to song, which seems to mean it doesn't produce constant quality output over different genres of music. I will try ABX once I have some time.

The fact that I managed to pick it out at these bitrates surprises me since I consider my hearing to be pretty crap (I have trouble distinguishing vorbis -q 2 from the original, even on music like Korn - Here To Stay (which is metal)). I expected -normal to be transparent to me, and -streaming to at least come close. This leads me to belive there is an actual problem with either the encoding or the decoding (clipping maybe?). Could someone post a link to somewhere which shows how to use WaveGain with EAC?

Thanks in advance. It's just that it seems a little silly that I'd be using mp3 when my player supports a "better" standard. There must be something wrong. I'll see if using the Wa3 plugin in Wa3 makes any difference.

OK, but how small was the AAC file with the same song then?



No, it means that at least the new methods of MPEG-2 AAC do work in PsyTEL compared to those from MPEG-1 (better short/long block switching etc.).



OK, so you probably like the sound of Ogg Vorbis better than that of AAC, which seems reasonable to me considering your preferred music and my impression of the "Ogg sound" (no flaming intended here).



Maybe, but I don't think it's clipping, because SometimesWarrior seems to have made similar experiences with PsyTEL AAC. I can also say from my ongoing low bitrate tests that I'm hearing those problems with crash cymbals, too (compared to the best codecs like FhG AAC, WMA9 or mp3PRO). But of course this doesn't mean much, because I'm "diggin' through the dirt" of 55-80 kbps at the moment. My solution is to resample to 32 kHz (which would make no sense in your case, of course) and/or to lower the cut-off frequency with the -c switch that also works with the standard presets (but not with -qvbr). This might be an option for you, because you mentioned 22 kHz as the actual cut-off with -normal, which doesn't need to be that high in my opinion and could contribute to a "warmer" sound, especially with your music. So maybe using -c 18000 or -c 19000 would come closer to LAME's aps...



Don't forget to compare the reconverted WAVs from the MP3 and the AAC file to the original WAV - just to make sure with there's no problem with the plugin...

And concerning the player: if you decide to wait for an Ogg Vorbis support in IRiver's SlimX, you should take into account that it will probably cost much more than a Philips EXP (around 100 EUR, I guess).

Edit: Why does this board software not like my Euro sign? I can see it when I enter my post, but not when I've submitted it, only some garbled characters. I'm using MSIE 5.5 on Win95B, and have no problems with that sign in other applications.

OK, but he did not use the -Y switch, otherwise he wouldn't probably get a 264 kbps file with aps. As far as I know, with -Y LAME does not try to encode the very high frequencies at all, so this would perhaps sound equal to a lower cut-off with PsyTEL AAC and might come out with comparable file sizes, too. But I'm only guessing, because I never tested at these bitrates with those codecs.



Well, 9 MB bigger seems like something to me, more than 10%...



Aha, so there's an Expanium user who knows, at last... Thank you for your clarification.

The AAC -normal version of the file has a bitrate of 190, which I will freely admit is MUCH smaller than 264. Part of my point is that if the song is really that hard to encode, AAC should have given it a higher bitrate than usual (since 190 isn't enough) like aps did (though it did give it higher than the 175-180 which -normal seems to average out at on what i've tested).

I will try re-encoding these files using wavegain on the wavs (should re-encode the mp3s too, to be comparing like with like?). Again, how do I use wavegain? Do I have to rip the wav files, then use wavegain, then use something like PystelDrop or LameDropXP, or can i set it up to work with EAC? What settings do I use in WaveGain?

Incidentally, I've just bought an Expaniuim 303 (£70, second hand but unused) since the vorbis players from iRiver will be MUCH more expensive than this. The Expanium seems really nice except for the AAC volume bug and the fact that you can't upgrade the firmware (I assume, since there isn't any way of connecting it to a computer). It says in the manual that it only plays AAC files with bitrate at most 160, but it seemed to play my -normal files ok (even though the average bitrate is higher than 160, so the max bitrate certainly is...?). Odd.

Actually, I'll put one of the AAC files up here:

[Edited] No full songs. Max 30 sec clip.

I hope this doesn't break the board's copyright rules, this is just for testing though (and most people probably won't like this song anyway ).

Could people tell me if this file is just fucked (due to clipping or something) or is this how AAC sounds?

Did you try out -extreme with PsyTEL AAC? And/or lowering the cut-off frequency as suggested?



Sorry, don't know, never used it.



For sure, as far as I know the listed price is somewhere around 300 EUR. At least a test report in Stereoplay, a german HiFi magazine, stated this in their September issue.



As far as I know, those CD players can be flashed with a CD that either comes directly from the manufacturer or can be burnt with a downloaded Flash ROM update. But I don't know if this is also true for your model.



That's what was also stated in the thread from last December here at HA, I think it can even handle 320 kbps files.



Sorry, no time, no mo' HDD space, maybe later...

Hmmmm.... I've encoded Korn and the like with AAC and really like the sound. Sadly, I couldn't ABX an 80kbs copy of Blind from the original!




I've tried it, and I just can't get it to work. I'm using 2.15, with the Ivan & Menno front-end. Being able to lower the high-pass on preset's is something I've wanted to do but can't, it just doesn't work!! Although I was told, there's not really much need for it, as although -normal low-passes at 20500hz, it only goes above 16khz occasionally for snares etc..

Sorry about posting the full song; I didn't know it wasn't allowed on these boards. I'ev replaced it with a short clip from the song (less than 30 seconds). I used SoundEngine (from RareWares) to split the wav before encoding, then used PsytelDrop to encode to normal. I didn't change any of the settings in SoundEngine so I assume this hasn't changed anything. The link is the same as before:

http://users.skynet.be/bk231049/sp/media/halo.aac

I don't quite understand all these sad guys that can't ABX their favorite originals from low bitrate encodings... From what I've found during my tests, ~80 kbps (-radio preset for normal tracks) is already enough to sound very good with PsyTEL and comes closest to FhG AAC or WMA9 at 64 kbps. This also corresponds to the comments from Karlheinz Brandenburg in "MP3 and AAC explained", where he states 96 kbps as the "sweet spot" for AAC.

But I'm still not finished, because I haven't had the time to test -pns or maybe even intensity stereo with Psytel (if this switch isn't disabled in the code anyhow).



You're right, I've just made a short test with -normal and my usual c't reference.wav, and -c doesn't work with any presets, only with a constant bitrate - sorry... So one option less for The Belgian, and he could e.g. try -br 192 -c 18000 for his problem sample. This is not as bad as it may sound, because AAC uses a bit reservoir even with a constant bitrate, just like the other formats, and so will vary the used bits in an intelligent manner. Ivan also seems to have spent some effort with that module, judging by his AAC_Implementations.pdf.

The problem I have is that I would have to reconvert your file to WAV in order to be able to listen to it on my old PC, and then probably would have to make some own tests at that bitrate in order to say something reliable about the problem. But my HDD is already filled up, so there's no chance for me to help you at the moment.



The last thing I can think of (see my reply to Mac) is to use a constant bitrate with your problem sample, but I don't know if PsytelDrop can provide that command line to AACEnc. It would have to be something like this: -br 192 -c 18000 -if test.wav

You could also change the settings to other values, if the problem with the "smeared cymbals" is still there. To lower the cut-off frequency is one method to enable the codec to use more bits for problematic sounds in the audible frequency range, that's why I suggested it.

Another update on cut-off frequency and variable bitrate/presets with PsyTEL: You can lower (or raise) the cut-off in combination with -vbrhi (VBR High Quality) or -vr (VBR Medium Quality). Furthermore you can roughly direct the outcome of the file size with an additional -br (called "base rate" in that context), but have to avoid invalid bitrates for that setting, e.g. -br 32 would be the lowest base rate allowed together with -vbrhi, because -br 16 (or any other value below 32) will crash AACEnc. The same is true for -vr together with lower values than -br 16.

So a possible command line would be: aacenc -vbrhi -br 32 -c 15000 -resample 32000 -if test.wav in order to force AACEnc to use the High Quality VBR mode with a cut-off at 15 kHz and resampling to 32 kHz. Using this line on c't reference.wav resulted in a nominal bitrate of 96 kbps, so the file was too big for my purposes (even with -vr it approximated 86 kbps). But maybe someone else is interested in this... The Belgain could e.g. try -vbrhi -br 192 (or more) -c 18000 for his problem sample.

[Edit:] There seem to be problems with -vbrhi together with very low "base rates", because sometimes AACEnc crashes at the end of the encoding, sometimes not. In both cases Winamp will freeze trying to play these files, maybe the header can't be read by the plugin (in_mp4.dll). Combining -vr with -br 16 and -c <x> works, at least with my setup, and also -vbrhi with high base rates and -c <x> (as suggested to The Belgain).