SoundExpert testing engine is now fully operational and I am going to add hi-bitrate codecs. As the testing is “participation consuming” and the same time it has to be interesting in the results sense I suppose to add:

1. SBC (A2DP, Bluetooth) 375 kbit/s (joint stereo)
2. ATRAC3plus 256 kbit/s or ATRAC4.5 292 kbit/s
3. MP3 256 kbit/s or WMA 256 kbit/s or AAC 256 kbit/s

Also I am not sure about the particular codecs for the third point. Any suggestions and opinions would be helpful.

I would be interested in AAC. The question is, which aac-lc encoder? QuickTime 7.02 have a 'vbr' lc-aac encoder. I think you should consider that one.

You mean aac-lc 256 from iTunes 5.0?

Use Atrac3plus because there is no software encoder for standard Atrac and thus no possibility to use it with any Sony hardware except Minidisc players.

Yeah. They are exactly the same encoder.

It seems to me that mp3_256 would be preferable because most people know its quality level and so it could be a kind of an ancor for all other codecs. Even it might be added to the system first for the purpose. Then other ones could be added and AAC as well. What mp3 codec to choose is still a question.

I would like to see Ogg Vorbis in testings but not the 1.0 version (as I see on the official site) .
aoTuV b4 should be the right one.

Yes, de Mon, I am too. Finally all major codecs will be presented. But now I have to choose just a few one. The old codecs (like Ogg 1.0 or QDesign music 2) still be in system for comparison .... and memories.

I added first high bit-rate codec to SoundExpert:

mp3 CBR@256.1 (Audition 1.5) - MPEG-1 Layer 3 CBR, 256.1 kbit/s FBR
CODER: MP3/mp3PRO® Encoder for Adobe Audition 1.5
- Codec: Current - Best Quality
- Allow Mid-Side Joint Stereo: No
- Allow Intensity Joint Stereo: No
- Allow Narrowing of Stereo Image: No
- 44100 Hz Joint Stereo
DECODER: MAD 0.15.2b

It will play the role of "anchor" as it is stable and well known. All its test files were graded once by myself and the rating displayed is a kind of starting point. In other words the rating value be defined more precicely as volonteers will add their grades.

As I understood from the above posts the next candidates for inclusion are:

1. AAC 256 (CBR or VBR ?)
2. OggVorbis ~256
3. ATRAC3plus 256

and my personal desire - SBC (A2DP, Bluetooth) 375.

Any other proposals?

Honestly dude, you won't be taken even slightly seriously here if you don't use LAME - and at VBR!

Edit: BTW, SBC is known to be really godawful, maybe even worse than MP1 (yes, I mean it! MPEG audio layer one). So, if you really go with it, you might as well make it the low anchor.

Why not to include LAME. Don't see obstacles (and I'm not afraid to be taken here not seriously )




Good chance to prove this.

Well, I suppose he does have somewhat of a point there. Some FhG encoders mess up joint stereo, so when you go with high bitrates, it's better to use the headroom and use standard stereo instead.

Dunno if the FhG implementation in Audition has that issue though - specially because there are three of them!

Well, you can read the specifications to get an idea.

Here they are:
https://www.bluetooth.org/foundry/adopters/...P_Spec_V1_0/en/

Yes, I've got it. I also have Bluetooth headphones and software implementation of the codec. It sounds not bad. It seems to me that the comparison would be interesting.

Is it just me or did you specify turning on joint stereo once, and then turning it off twice ? Also, can I ask why you want to use such an outdated encoder (ignoreing the questionable settings)?

ummh, why not use the one bundled with windows media player 10? (It's also FhG, and dated 2004)


you can set everything to yes and yet it still use stereo
If I'm not wrong, WMP10 (FhG ACM) will encode everything in stereo from 160kbps and above.

I think you should also include LAME --preset extreme, many use that, especially this forum

The coder used:


These are recomended settings for 256 CBR. To be true I'm not shure the output is "joint stereo" because only MAD decoder stated this while decoding.

The reasons why:

1. Audition is claimed (by itself though) to be PRO.
2. Its mp3/mp3PRO filter has many settings for both CBR/VBR and mp3/PRO (apart from eg. WMP 10)
3. It became "classic" already (as MM JukeBox also uses it) and hardly be tuned further in the future. "Anchor" indeed.

LAME --preset extreme. Are there any other setting proposals for the coder?

Hrm... maybe I should add "LAME Pro" to the new LAME homepage design. It won't make the slightest difference other than a warm fuzzy feeling in the users, but oh well...



LAME has many more settings for CBR/VBR.



Tweaking is bad... mmmmmkay?

I tested this summer the "professional" version of Fraunhofer at 96 kbps. At this bitrate, it was well-known (understand: by tests older than Mathusalem) that LAME is uncompetitive against Fraunhofer encoders, especially the "Pro" ones.

http://www.hydrogenaudio.org/forums/index....pe=post&id=1654

more details: http://www.hydrogenaudio.org/forums/index....ndpost&p=314613

I wouldn't use Fraunhofer engine for a blind test competition.

Ok. Let's say Fgh mp3 coder will be the "low anchor" in 256 coders group at SoundExpert. Realy better? No problem. And its popularity is a sort of popularity of Latin among doctors.

Sorry?

I seriously doubt that any 256 kbps encoding could play the "low anchor" role. But if you insist, why not Blade instead?
But if your purpose is to directely compare MP3 to other (real) challengers, I strongly believe that LAME would be much more interesting to test than any Fraunhofer encoder.

LAME will be added for sure. I suppose: 3.97beta -V 0 --vbr-new ? (for 256 group).

It looks as the most consensual setting (I suppose) for this bitrate range. I'm not sure that it corresponds to 256 kbps encodings on average, but it's probably not far from this target.

For a chain of 9 SoundExpert test files 3.97beta -V 0 --vbr-new showed 217.3 kbps. I think it's ok for 256 group.

Should I add Lame first, choosing from:

1. AAC
2. ATRAC3plus
3. Lame
4. Ogg

its pretty unfair, being left behind 39kbps
a bitrate jump from 96kbps to 128kbps has great improvements in quality, especially mp3

you can try to use
-V 0 --vbr-new -b128
-V 0 --vbr-new -b160
-V 0 --vbr-new -b192

make it as close as possible to 256kbps

Oh that old contradiction between science and practice… I think that comparison of “real usage” settings has more sense. In most cases the choice between different coders means the choice between their “recommended” settings. Nobody wants to use less tuned settings, so I’m afraid that something like “-V 0 --vbr-new -b192” would be interesting only for developers, not users. Please, correct me if I’m wrong.

I agree with you. Either you accept the bitrate of -V0 or you choose another setting (like abr 256).
By the way, I don't think that the additionnal -b192 would significantly increase the bitrate. Could you try with your samples?

Moreover, people recently tried to find the average bitrate corresponding to -V0. AlexB has tested various musical genre and get ~240 kbps as average. I personaly get something lower, but my table is based on classical music only (very favorable for MP3: no or few sb21 bloat).
It's likely that -V0 is on average slightly inferior to 256 kbps for most musical genre, but the difference shouldn't be that great, as long as you test 'various' samples. I'm just surprised by the low value you get with your samples. Are some of these samples mono recordings?

It seems that the samples used don't make LAME to use as high bitrates as e.g. modern pop/rock/jazz etc music usually does. I got these results with real music files: http://www.hydrogenaudio.org/forums/index....ndpost&p=328558

In the same thread, Guruboolez's results with classical music were closer to yours.

Would you mind sharing the original lossless samples with HA community? You might get some interesting comments and perhaps some standard HA listening test reports too. I suppose it would be nice to have another doctor's opinion as a reference.


Edit: I noticed Guruboolez's post only after posting mine (If someone wonders why I wrote about the same things.)

No problem. I would be glad to get some comments.

Bach.J.S, "Oster-Oratorium, BWV 24" 925 Kb
Bass (SQAM) 1.11 Mb
Castanets (SQAM) 689 Kb
French Male Speech (SQAM) 0.99 Mb
Glockenspiel (SQAM) 1.18 Mb
Harpsichord (SQAM) 826 Kb
Lo-Fi Analog Tape Recording 781 Kb
Mike Oldfield, "Music From The Balcony" 1.12 Mb
Quartet (SQAM) 1.06 Mb

Most of these excerpts are single instrument/voice samples, that's why the overall bitrate is so low.

I can't download the samples yet (56K), but if I remember correctly, SQUAM samples contains large part of silence (I'm pretty sure that's the case for harpsichord). If I'm not wrong, it would explain the low bitrate you get on average.
Could someone confirm it (through encspot pro maybe)?

-V 0 --vbr-old => 216.8 kbps
-V 0 --vbr-new => 217.3 kbps
-V 0 --vbr-new -b192 =>224.2 kbps

Right you are, SQAM samples contain silence but the ones above don't.

I downloaded JSBACH sample, and encoded it with -V0new:



There are more than 3 seconds of digital silence. It has a big impact on bitrate (179 kbps). I add the additionnal command "-b192 -F" and bitrate was increased by 40 kbps.

Could you try -V0 --vbr-new -b192 -F ?
-F forces the usage of minimal bitrate (no 32 kbps frame for digital silence). Thanks

Actually the process of testing and target bitrate determination is a bit simpler at SoundExpert. For both purposes one single sample is used which consists of all 9 excerpts chained one after another without any silences. You could download it from here (8.71 Mb, sorry). After the whole sample have coded the target bitrate (or File Based Bitrate - FBR at SoundExpert) can be find by deviding the coded file size in bytes by 11641.573. The result is in kbit/s.

OK, thanks for clarification

N.B. I haven't downloaded your file, sorry.

In the end I'm still sure that “-V 0 --vbr-new” should be used for testing. All coders at SoundExpert are divided into bitrate groups for convenience only and could be regrouped according to some other criteria if such appear.

It takes me one day to add LAME 3.97beta to the system with first preliminary rating. What to choose next: AAC, ATRAC3plus or Ogg?

Sorry for some delay. The Lame has been finally added. Please note that the rating displayed is highly preliminary and needs listener’s grades.

To my first impression perceived quality of Lame –V0 is pretty similar to Fgh 256 except Glockenspiel which has less perception margin with Lame.

I'd like to see the latest Ogg Vorbis aoTuV beta4.5 tested. From the last 180 Kbps guruboolez test aoTuV beta4 was the better codec.

Binary here.

aoTuV beta4.5 produces the following bitrates with SoundExpert test files:

q6 => 183.7 Kbit/s FBR
q7 => 210.7 Kbit/s FBR
q8 => 238.8 Kbit/s FBR
q9 => 309.2 Kbit/s FBR
q10 => 439.5 Kbit/s FBR

I’m not sure which one to choose: -q7 or -q8.

The other two codecs tested were at 256.1 and 217.3, so why not use -q8?

Also, since the test is at 256 kbps nominal, -q8 is the nearest to this value (-q8 produces a nominal bitarate of 256 kbps at 44.1 KHz stereo).

Actually I’ve already added Vorbis q7. Going to create “224 Kbit/s” group and move there Lame and Vorbis from 256 group.

According to first results Vorbis q7 has substantial perception margin - http://www.soundexpert.info/coders256.jsp

What to choose next?

OK. Which aoTuV version have you used, 4 or 4.5? Your web page says:



but in the previous post you say 4.5.

Thank you. Right you are – it is 4.5 as stated at RareWares: Oggenc2.6 using aoTuVb4.5 2005-11-10 where I downloaded the codec from. I'll correct the page now.

I'd like to see aoTuV beta4.5 tested at low bitrate, but maybe this is off topic.

Anyway, thanks for your tests.

Will be for sure.

If you download and grade a test file from SoundExpert (or a few ones) these tests will be yours as well

Can you consider using Gogo 3.13 -b 256 -q 2?

Gogo's behavior would be very interesting as compared to Lame's:
- it uses gpsycho as the underlying psychoacoustic models which has some merits with respect to pre-echo. It would be interesting to see whether or not gpsycho's disadvantages are relevant with 256 kbps.
- though with -b 256 transport frames are all the same size this is not true for audio content within a frame. Other than suggested by common sense Gogo's cbr 256 mode is struggling for constant good quality and can achieve it by quantisizing audio content of a difficult frame with a resolution that for instance Lame 3.97b's -b 320 mode is not able to provide. This is why cbr 256 mode is to be favored over vbr mode when using Gogo.

As mp3 is still the most wiedely spread format this might be a good reason to consider several encoders.

Actually I’m a bit confused with the proposal, because I know very very little about this codec. Is it really popular?

If you suppose it to be better than Lame at high bitrates then it may be interesting to compare it with Lame at 320 kbps bearing in mind the question of achieving the best possible quality with mp3 format. I’m not sure if this question has much sense. Any ideas?

Not at all.

In case you will give 320 kbps audio samples for testing IMO Gogo would be better there than with 256 kbps.

BTW Gogo is a development based on Lame 3.88 with special tuning to very fast encoding. Used with CBR Lame 3.88 is pretty much the same as more famous Lame 3.90.3, but using bit reservoir to a larger extend.