obviously, AC3 is lossy, but DTS is as well. however, the problems in theatrical AC3 is not present in DVD AC-3 because higher bitrates can and usually are used, as far as the stuff i've encountered (i wish i could borrow/rent Kings of Comedy to triple check that). well, the Garbage music video, TWINE @ 96kbps being an sadly grieved exception to that. if high enough bitrates are used in AC3, it should at least be "on par" with DTS (which i thought i read was not the most efficient codec).

i guess what i'm saying is, is that AC3 can be just as good of a contender as DTS, since they're both lossy.


oh, and i wouldn't suggest a straight music video DVD for LPCM, unless they mastered the sound specifically for the DVD. they seem to use 2nd generation audio from an analog-seeming transfer or something as most don't even compare to the CD mix.

that sounds like a good sample, at least based on my experience that killer samples seem to have synthy or distortion related attributes to choke the codec.

I really like my caffeinebattlecry sample and it would probably cause problems with some of the encoders (haven't tested yet):
caffeinebattlecry.sample20sec.flac
caffeinebattlecry.sample5sec.flac

It did really well as a sample in my private (F)AAC tests, but I guess people could complain about the test becoming too muck rock and not enough classical stuff. I will look for a good live sample later today.

Hrm... I can already hear people complaining that sample gives them headaches

Also, I don't think it represents any specific style well. I can't even identify what instrument is playing that!



Do you think that sample would be good to replace the other sample you submited? (mybloodrusts). I think it's kinda too agressive and "messy" (for lack of a better word), I already got comments from several people complaining they don't know where to focus to try to detect artifacts on such kinds of samples (gone's final part is very difficult in that aspect too)

Identifying artifacts in mybloodrusts is really easy in my oppinion (and according to the ratings from the prior tests most people seem to think similiar), but I wouldn't have a problem with replacing it with caffeinebattlecry, which might represent punkrock/emocore/hardcore music more accurately, since it's less noisy and features more "classical" artifact spots (hard attacks etc.).

I'd really like to have them both featured, but I see that you can't do that. There are some interesting spots on albums I bought recently, which I haven't looked into any further yet, but I promise there'll be a shitload of new heavy samples for the next test set ready.

http://sjeng.org/ftp/work/Peccatum.flac

Old sample from Dibrom. IMHO very useful for codec testing. (Posted in other thread)

In case it's not too late yet:

I've finally found a good/hard to encode live sample. It's from
James Brown - Greatest Hits - Body Heat (live)

Uploaded here.

I don't think that biniou.wav is a killer sample. Nero AAC have problems with it in the past - but other encoders sound find, close to transparency.
Brass instruments in Mahler3 are more terrific, for all encoder at this bitrate

still interest in a movie sample (speech will low volume background music)?

if yes, how should i treat the multichannel dts source? it should be possible to downsample with azid to stereo...

Indeed, very interesting. Do you think it can be used to replace gone maybe?

@tigre: Thank-you very much. I'll replace Scars with it, since that seems to be another sample listeners don't like much. Also, I didn't have any funk/soul sample

@bond: I probably can accept, but you will need to hurry. I'll start encoding the samples very soon. Also, tell what sample should be replaced :B

thanks to gam3r and fr4nz i now have 2 sources:

1) speech only
2) speech + some pumping sound (from drums?) in the background

which of the two is prefered?

Speech can usually be compressed down to 16kbps while keeping a quite good quality (look at Speex, ACELP, etc.), so I think the version with pumping will be better. I'll test it to see if it is too obviously easy to encode, and I'm afraid it might be.

Downmixing with DPL would do, no problem.

And resampling to 44.1kHz would also be very important

Also, again: What sample will be replaced?

ok here it is: http://8ung.at/bond/NinjaScroll.flac

i created it that way:
1) decoding the source .dts (from the anime Ninja Scroll, thanks gam3r! ) with the Intervideo Audio dshow decoder (which is the only stable DTS decoder i know) to .wav (default settings: 2 speaker mode - dolby surround compatible, vocal options: both)
2) cutting the .wav in cooledit
3) saving as .fla with case's flac plugin (default setting: compression level 5)
4) rename to .flac (i assume thats correct)

i hope its interesting


hm also again, no clue as i dont really know the samples which will be used atm

Server down?

I can't get it from here nor from the 3 SSH servers I have access to.

Can you please put it at the upload thread?
http://www.hydrogenaudio.org/show.php/showtopic/18632



http://www.hydrogenaudio.org/forums/index....ndpost&p=185286

Downloaded, k-tnx

Some problems I find:

-The volume is too low
-I couldn't find any part of it that would stress codecs. My guess it that most of them would reach transparency there.

Opinions from others?

well the volume is always that low on movie dvds (normally someone would gain the whole thing, but i avoided that of course to not change the source)


hm if you think so

Last minute change: Fatboy will be replaced with Roel's velvet.wav. Compaact behaved too badly on Fatboy (270kbps!). And velvet is pretty evil neverthless.

So, the final frozen list will be:

BigYellow
bodyheat
DaFunk
gone
Hongroise
Mahler
mybloodrusts
NewYorkCity
OrdinaryWorld
Quizas
velvet
Waiting

Regards;

Roberto.

Ugh! I can't do ABXing with velvet, one constant beat with melody tends to get to me pretty fast, one constant beat on its own . . . . my ears just can't take that.

hello.

I'd like to ask a favour from someone that understands well the unix shell.

This is one of the .bat files I'll use for this listening test:



Can someone please provile a .sh script that would perform the same?

I will expect the test participants to have FLAC an FAAD somewhere in the path. There's no way I can provide binaries for Linux, Solaris, MacOS, BSD... specially considering all the dependency issues.

The echo off isn't needed.

Thank-you for your help.

Regards;

Roberto.


Also, so that critics can start complaining right away that Nero bitrates are too high:
http://pessoal.onda.com.br/rjamorim/Bitrates.txt

Omg, they're too high! I understand that bitrates tend to rise for problem samples and for overall music they stay closer to their target (as I once argued), but I doubt this is the case here....I'll encode a few CDs to check.

They really made a special Guru-optimized build, after his classical test results.

I'd use it to replace whatever sample had the least discriminating power/highest scores and that's still in. Qua genre it has a little of a lot, if you know what I mean.

I tested 272 samples which are a mixture of completely different music styles. I have 132.5kbps average for Nero, which is actually lower than for Compaact1.2b2 VBR5 which was 1 kbps higher average.
It depends on the music style pretty much.

JohnV:

Compaact 1.2 beta 3 has been out for almost a week I believe. Alexander informed me of it. Perhaps he forgot to update his beta thread with that link. I think it has various optimizations, so you might want to get a copy instead of using beta 2 for testing.

Do you really think he didn't send me an e-mail already about beta 3?

Edit: Erm? That post is directed at JohnV?


So, anyone willing to take a look at my batch script to convert it to bash-friendly?

If Nero and Compaact were so close, maybe the fatboy sample where Compaact reached 270 kbps should be reconsidered? :B

Hold a sec...

BigYellow 143 123 130 124 111


ok..> I encoded BigYellow...

Original: 4320784

MP4: 419946 - RATIO 10.288 - BIT RATE - 137.1 Kbit/s
Audio Track Only: 413488 - RATIO 10.4495 - BIT RATE - 135.02 Kbit/s

(Audio track only is the actual encoder bit rate measured in bit allocation, you could get it by extracting track)

So I think claim 143 + 5-6 kb/s is little bit out of bounds

Indeed, partially my fault. foobar randomized the input, and I forgot to sort it properly.

These are the bitrates foobar reports for the sample suite, encoded with the AAC plugin v. 2.6.2.0, high quality, -internet, LC:



The average is still 141kbps.

The final proof:

All 12 wav files summed: 52.895.248 bytes divided by
All 12 aac files summed: 5.264.357 bytes equals 10,0478

Final bitrate = 1411 (CD bitrate) / 10,0478 = 140,428

So there you have it.

rjamorim,

Yes that reply was directed at JohnV replying to his post mentioning beta 2 of Compaact 1.2... Sorry if you missed that.

[quote=rjamorim,Feb 17 2004, 09:05 AM] [QUOTE=Garf,Feb 17 2004, 08:37 AM]
@tigre: Thank-you very much. I'll replace Scars with it, since that seems to be another sample listeners don't like much. Also, I didn't have any funk/soul sample [/quote]
nooooo! i liked scars! and i liked it even more once i found out it was from Chrono Cross

Nero on Internet :: Medium (High quality) is fair, over 500 full audio tracks (2G encoded mp4) it came out 2.5% above 128Kbps.

Compaact comression is running right now, in about 7 hours it will be completed and I will post the results all together.

well, here's a quick translation, I haven't tested it

Simply copy this into a blank file with LF for newlines, and set it to be +x

What faac setting did you use? Default seems to give considerably less than 128 on average sample.

-q 115. The average bitrate ended at +-132kbps

I didn't use TNS, since it is disabled by default and not recommeneded at the Audiocoding Wiki page.

Mammouth encoding task is over, 3 days worth - lets just say central heating was not required

Here are the raw figures for final file sizes:

500 lossless Ape files converted to Mp3 @128Kbps came to 1,980,037,004 bytes


FAAC (1.23.5) Quality 125 2,074,364,607 bytes 5% over
FAAC (1.23.5) Quality 100 1,881,780,853 bytes 5% under

Quality 115 should be spot on



Nero Internet::Medium high quality 2,029,931,171 bytes 2.5% over



Compaact q5 1,922,327,404 bytes 3% under

Guessing q6 would be 15% over (would have liked to try q6 compression, but we ran out of time).

Best of luck Roberto!

These bitrates sound fair

Thank-you very much for your help, Spoon

The test will start any minute now.

Thanx, I didn't even know it's so high.
Edit: I mean I didnt know the requred quality setting is so high.

Well, the required bitrate is 128. That's it


Test is now open:
http://www.hydrogenaudio.org/forums/index....ST&f=22&t=18863

Please participate

I'm coming a bit late.
~250 discs are currently stored on my HDD in lossless format. Classical music only. I've loaded them with foobar2000, randomized the playlist, and converted them with VBR streaming (aacenc32 2620 HIGH). I've stoped after 40 hours of processing.
1495 files were converted. Musical length: 105 hours of music.



=> Bitrate is superior to 128 kbps with classical music. But the deviation isn't too high.

I'm using the Internet VBR profile in the test... :/

Stux, you should guard these "rm -f" with a condition so that the users don't lose their samples. Something like:

Would be safer, Roberto will appreciate

Actually, I would just get rid of the rm's

oups... typo error.

I used VBR internet too (streaming would be clearly > 150 kbps).

Ah, OK.

Thank-you very much for testing bitrates.

http://lists.mpegif.org/pipermail/mp4-tech...ary/003312.html

Anyone wanting to conduce listening tests?

Sample 7 is next up.
Most of the samples have been transparent to me.
Some very few, have been almost transparent. But since the 128 multiformat test, where everything but mp3 and ogg were transparent, I didn't expect to actually hear diferences.
Using onboard sound and walkman headphones at the job, but hey, it's the only way I can test...

Is this a bad time to mention my PuntAAC codec (which says it encodes at 128kbit but in fact runs losslessly while using 5000kbit) and my normalisation rule



where:

S = Final rating
F = Rating given by listening tests
A = Normalized bitrate quotient = b / B (where b = track bitrate and B = 128)

example
If a track with the chosen quality setting (aimed at 128kbit, obviously) turns out to sound 'transparent' (5 out of 5) to everyone, but has a bitrate of 700, then without my normalisation rule it would've won with the perfect score, while with my rule, the final score would be:

S = 5 / (700/128) = 0.9

which sounds fairer to me...

yes, yes, I know that the quality settings are chosen such that the SAMPLE SET averages near the intended bitrate, but:

- It still doesnt paint an absolutely fair picture to encoders that consistently end up slightly below the target bitrate
- It doesn't mean that nothing can be learned from the INDIVIDUAL SAMPLE statistics, especially if a sample is more representative of the type of sound you usually encode.

*ducks chairs thrown by absolutists*

Here I come to bitch again.

So far, I got samples from 6 (six) different listeners. And most of them didn't submit complete result sets.

Obviously, that's not enough to generate statistically valid results. Specially considering I didn't screen the results yet for ranked references.

So, I'd like to know, from the ones that tried to participate, the reason of such low response:

-Having problems with ABC/HR Java?
-The samples are too transparent?
-Getting tired of so many tests in a row?
-Another reason?

I don't plan to extend this test. If I don't get enough results by sunday night, I'll probably upload whatever results I got, and give up.

Regards;

Roberto.

PS: No troll feeding, please ^^

Here: I'm doing 1 or 2 samples a day. When I'm finished I'll submit the results. Besides this, I should save my results more often - I've lost results when I've been almost finished with a test 2 times already, because my PC crashed (not related to ABC/HR)