HydrogenAudio

Hydrogenaudio Forum => Validated News => Topic started by: menno on 2003-06-17 23:27:35

Title: AAC+SBR listening test
Post by: menno on 2003-06-17 23:27:35
Hello everybody,

For a university project I conducted a little listening test on the new Nero AAC+SBR codec. Other codecs tested were AAC (also Nero codec but without SBR) and Vorbis. The AAC codecs where used at a constant bitrate of 64 kbps and Vorbis at 64 kbps VBR. The results are shown in the follwing graph:

(http://www.audiocoding.com/Vorbis.AAC.SBR_64kbps_MUSHRA_test_results.png)

People that conducted the test: JohnV, Ivan Dimkovic, Case, Moneo, superdump and Garf.

Greetings,

Menno
Title: AAC+SBR listening test
Post by: rjamorim on 2003-06-17 23:43:06
OMG! Menno is trying to steal my job!
Title: AAC+SBR listening test
Post by: ff123 on 2003-06-18 01:22:25
All right, which audiophile rated the reference less than perfect?

ff123
Title: AAC+SBR listening test
Post by: rc55 on 2003-06-18 01:28:41
Wow! Considering that the AAC encode is CBR and Vorbis is VBR, this is pretty stunning. These will be interesting times, and I hope Xiph.org take a second look at Vorbis once their work has calmed down.

As wise Garf always says, Vorbis has a massive amount of tuning opportunity, so get in there if you have the ability.

I'm also wondering about the speed of the encode... AAC for me has traditionally been a slow encode, is it still the case?

Ruairi
Title: AAC+SBR listening test
Post by: Case on 2003-06-18 01:35:38
Quote
All right, which audiophile rated the reference less than perfect?

At least I did. The original sounded worse than 7.5kHz lowpassed, I couldn't give it full 100 points. I gave it 90 points and 88 to lowpassed version.
Title: AAC+SBR listening test
Post by: ff123 on 2003-06-18 01:35:55
I'd like to see the listening comments.

ff123
Title: AAC+SBR listening test
Post by: JohnV on 2003-06-18 01:48:35
Quote
All right, which audiophile rated the reference less than perfect?

ff123

Lol.. You just had to notice that embarrasing detail..  Gotta admit I made one little mistake with the LisztBminor. I'm not gonna start explaining cause you wouldn't believe anyway... 
Though, that piano sample has very little hf content above 7kHz, so you could in right circumstances after adjusting your default audio settings think that what you hear is noise from distortion and not the hidden reference's high freqs, and think that the 7kHz is actually the hidden reference since there's no know original for comparison and it sounds cleaner...    I should have used higher volume on this one, but it was the last sample and I was in hurry...............

But I won't explain..  (blaa blaa blaa)
Title: AAC+SBR listening test
Post by: sony666 on 2003-06-18 08:41:44
Quote
AAC for me has traditionally been a slow encode, is it still the case?

Nero aacenc32.dll 2.5.1.2 (VBR normal, HQ mode, LC profile): 5.0 to 5.5x realtime, bitrate 155-185k
lame 3.90.3 -aps: 2.5 to 3x realtime, bitrate 180-220k
mpc 1.14 --standard --xlevel: 6 to 7x realtime, bitrate 170-180k

Celeron 1.7 with crappy PC100 RAM
Title: AAC+SBR listening test
Post by: guruboolez on 2003-06-18 13:25:01
My own values :

http://membres.lycos.fr/guruboolez/AUDIO/a...ac128/speed.txt (http://membres.lycos.fr/guruboolez/AUDIO/aac128/speed.txt)


EDIT : BTW, interesting test. Thank you for sharing the result
Is it possible to download somewhere the BeautySlept AAC+ decoded file ? Just for curiosity ? Or could someone compare it with a WMA9PRO (VBR -q25) encoding, which is very good on harpsichord ?
Title: AAC+SBR listening test
Post by: sony666 on 2003-06-18 22:33:00
Quote
My own values :

http://membres.lycos.fr/guruboolez/AUDIO/a...ac128/speed.txt (http://membres.lycos.fr/guruboolez/AUDIO/aac128/speed.txt)

a bit off topic here, but..

"
AAC Ahead Nero 5.5.10.35 -streaming... HIGH.....4'27.. x3,69 ... 15.0 MB ... 126 kbps
AAC Ahead Nero 5.5.10.35 -streaming... FAST.....3'17.. x5,00 ... 16.3 MB ... 137 kbps
"

humm.. quite a difference between fast and HQ setting, interesting. Did you ABX these 2 by accident? Would be interesting to know if "fast" gives out bits more freely to keep up the same quality. Or does it spend 11k more bitrate AND produces lower quality?
Title: AAC+SBR listening test
Post by: Cobra on 2003-06-18 23:44:36
I wonder if Ogg Vorbis can be tuned at low bitrates to compete AAC+SBR. Where are guys from Xiph?
Title: AAC+SBR listening test
Post by: atici on 2003-06-19 04:02:01
Can there technically be a Ogg+SBR ?
Title: AAC+SBR listening test
Post by: rjamorim on 2003-06-19 04:25:36
Quote
Can there technically be a Ogg+SBR ?

Technically: Yeah, no doubts about it. You can SBR any music format you want. Just needs the SBR encoder, the (de)multiplexer and the SBR decoder.

Pratically: Unprobable.
Title: AAC+SBR listening test
Post by: atici on 2003-06-19 05:46:37
But why do they refrain from doing so then? Why is the introduction of SBR came this late? Does SBR bring any disadvantages (like slower decoding in embedded hardware) ?

Hmm maybe I should check the SBR threads first  I just know it does a good job at low bitrates. How is that: MPC+SBR

Edit : Shoot. Couldn't find any nice technical SBR thread. Can anyone point me? maybe someone should add it to the Sticky FAQ thread in the General Forum... What is PNS ? 

As far as I read, SBR seems like some kind of a hack for high frequency prediction that are cut-off. Sounds ugly but it's for low bitrate anyway.  Speex+SBR sounds interesting...
Title: AAC+SBR listening test
Post by: jmvalin on 2003-06-19 07:15:59
Quote
Speex+SBR sounds interesting...

Actually, Speex uses an SBR-like (simplified) technique for wideband (16 kHz) below 20 kbps.
Title: AAC+SBR listening test
Post by: rjamorim on 2003-06-19 09:37:03
Quote
But why do they refrain from doing so then? Why is the introduction of SBR came this late? Does SBR bring any disadvantages (like slower decoding in embedded hardware) ?

First, you are right, SBR is an encoder inside an encoder, so decoding is slower.

Other points to consider is break of compatibility with Vorbis 1.0 (not really break, but you would need specific decoders to understand the SBR part, just like MP3/MP3pro); and the patenting issues. SBR algorithms are heavily patented, and they would have to code workarounds to these algorithms.
Title: AAC+SBR listening test
Post by: Cobra on 2003-06-22 14:45:50
PPL from radios will have to decide: pay money 4 slightly better AAC+ or don`t pay for Ogg Vorbis, which is slighly worse. I think that big part of them will choose Ogg Vorbis.

Btw: check this: http://cypress.man.poznan.pl:8000/zloteprzeboje3.ogg (http://cypress.man.poznan.pl:8000/zloteprzeboje3.ogg)

I think that this setting is optimal for Ogg Vorbis radio. Any settings below sucks, because switching form 44.1khz to 22khz will not cause dropping of kbps by 1/2 and quality is really worse. Same for stereo --> mono.

So IMO main goal for guys from Xiph should be tuning 48kbps 44khz stereo - for radio, and Q5+ for music storing.
Title: AAC+SBR listening test
Post by: JohnV on 2003-06-22 15:16:52
Quote
PPL from radios will have to decide: pay money 4 slightly better AAC+ or don`t pay for Ogg Vorbis, which is slighly worse. I think that big part of them will choose Ogg Vorbis.

I'm sure some will choose Vorbis, and it's not the worst choice.

However, aac-he cbr beats vorbis -q0 vbr. aac-he vbr offers even higher quality than cbr, so some quality radio stations may want the best..