Hello, I just received my Philips Expanium portable player, and so I've been experimenting with Psytel AAC.

I found a very problematic clip: Smashing Pumpkins' Bullet with Butterfly Wings. After the bridge, it's just Corgan singing along with a quiet distorted guitar.

The problem is, with every 's' sound in the vocals, the sound pans sharply to the left. This is very apparent with -streaming and -normal, and with -vbrhi -ihsc. I couldn't hear it with -ultra, and I don't think I could hear it with -archive or -extreme. I also couldn't hear it with Lame --alt-preset 130.

I tried it with psyaac v2.0 (both the 1/1 and the 1/3 builds), and psyaac v1.5, and both exhibited problems. The panning problem was apparent when played through winamp, when decoded using the 1/1/2002 build of faad, and when played with the Expanium.

I have a clip available online, at http://briefcase.yahoo.com/SumWarrior/ (just go into the Bullet Clip folder, there are two LPAC'ed waves, the original and the -normal aac from psyaac v2.0, Jan. 3 2002 build)

This problem is enough to make me consider only using MP3 for the Expanium, despite the public opinion that AAC is far superior to MP3.

If it's something I'm doing wrong, I would really appreciate some help! Otherwise, maybe it can be remedied in a future version of AACenc.

-SometimesWarrior

This is a problem with M/S - -disable_ms switch removes the artifact, I will see what I can do

Ok, AACEnc 2.01 will soon be up with following new switches:

-safems (Safe M/S switching, might increase bitrate), and this eliminates problems like those in your clip

OR

-mst <n> Maximum SFB difference in dB under which M/S could be used, meaningful values would be between 0.0 (disabled MS) and 10.0 dB


I am not sure what value will be used in which preset, but right now I recommend leaving everything on default, except for high quality VBR presets, like -normal, where -safems could be used safely.

More M/S switching research is on the way...

AACenc V. 2.01 (dated 2002-01-04) is already at my page.

But the new switches don't appear when you type aacenc -h

Regards;

Roberto.

No, advanced switches aren't shown on the -h page

Is there a way we can see the advanced switches? (Like --longhelp?)

Hmm, not yet, but I'll see if I can add this to the options list

Thanks for the quick fix!

It does increase the bitrate quite a bit, though (over 25%), so I'm gonna sit tight and see what magic you can work over the near future to fix the artifact.

One thing I should say is that the sample clip was dropped -5.85dB; I used ReplayGain on the album in MPC format, and it gave me an album gain of -5.85dB. I didn't see a -scale switch in the list of commands for aacenc, so I used a normalize.exe to adjust the original waves to get the same effect. I re-encoded the clip from the original wavefile, and the artifact was definitely less pronounced, but still clearly audible. The normalize.exe program I used may not have been audiophile-grade...

-SometimesWarrior

Are these issues still being worked on?

The specific sample I gave (Bullet with Butterfly Wings) has been improved significantly as of v2.12 (feb 10), but there are still problems. There are also other samples on the album with the same problem that I can easily detect during casual listening and ABX perfectly up through the -extreme preset. Just for reference, I cannot ABX the samples with Lame3.91 --alt-preset 130 when listening casually.

The problem can be described as a combination of panning and a rough "garbling" of the vocals.

I can provide more clips, but I'm assuming that if I've run into multiple examples of this problem with my limited artifact detection ability, then this problem is pretty common. Or maybe AACEnc just doesn't like Billy Corgan

No response yet...

I will make several specific, easily ABX-able flac clips available if anyone wants them. These are several different songs from the Smashing Pumpkins' Mellon Collie album that have serious glitches (I only listened casually through the album once and picked up these samples, so there are bound to be more).

The clips have very bad distortions up through the -extreme preset, distortions which are not present in lower bitrate mp3, ogg, or mpc encodes.

Maybe I'm missing something here... is AACenc not considered a "ready for prime-time" encoder, so Mr. Dimkovic already has plenty of problem samples to work with? I ask this because, with my limited experience with AACenc, I would never choose it instead of MP3 for large-scale encoding over 128kbps, where alt-preset MP3 sounds decent to me on my portable player.

I picked it over Lame in FF123's listening test from a 5 second audio clip, but if it has these serious distortions several times in each album, I don't see how anyone can take it seriously as a competitor to MP3 yet.

My question is: am I experiencing an abnormal case with this Mellon Collie album, or is it common knowledge that AAC still garbles one out of ten songs?

Yeah, post the clips please. Tweaking codecs is quite difficult task because there's always a chance that when you fix one thing, you break 3 other things...

I have the mpc-insane files handy, but not the CD. I will ask some friends if they have copies I can borrow, so the clips will be a bit more "valid" (so one can rule out transcoding artifacts as the source of distortion, which aren't IMHO something as important to be looked into).

I'm home from college, and I'm back with my CD collection. A lot has changed, but the artifacts in the test clips are still there.

I know Ivan isn't really working on aacenc anymore, but I'm discussing these problem clips because AAC is becoming more popular, and users should be aware that aacenc is (in my experience) not as robust as the other popular audio encoders, Lame and mppenc.

I made a list of tracks and track times for the Smashing Pumpkins - Mellon Collie album where annoying artifacts could be heard, and when I find that list I'll post it . In the meantime, I have a pathetic free filesharing account hosting my sample clip. Go to www.junglemate.com and login with login/password: sometimeswarrior/swarrior. The clip is in the "files" section.

The artifacts are blatantly obvious with the -normal switch. With the -extreme switch, the artifact is most apparent 15 seconds in, when the "s" sound pans to the left.

Within the next week or two, I will try to make all of my problem samples (at least 4 or 5 from the album) available.

In the meantime, I still want to know if it's strange for aacenc to have this many irritating artifacts even at -extreme, or if it's something that AAC users just live with.

I'd like to get away from the MPC format, but it seems like it's still head-and-shoulders above all the alternatives.

**EDIT**
I tried this sample with Vorbis -q5 and -q6 (using the latest OggDropXPd, dated June 7), set up an 8-way ABC/HR test with four copies of each sample encode, and correctly identified all the encoded samples and their encoded quality level. I'll test Lame and mppenc tomorrow and see if this sample is a "codec killer", like the one Garf found a month ago.