Skip to main content ↑ ↓

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: Snow Leopard AAC improvements.. (Read 73722 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Snow Leopard AAC improvements..

Reply #25 –
i wonder if this time around with iTunes 9 we'll finally put our hands on true vbr on Windows. It would be stupid if they don't.

Snow Leopard AAC improvements..

Reply #26 –
Quote from: rpp3po on
I know Soma, but they also serve good quality 128kbps MP3. I wouldn't want to trade that for synthesized low bit rate HE-AAC. Are there really many stations you could not receive without HE-AAC support in iTunes?


There are times when it's nice to have the option of a lower-bitrate but still accpetable quality stream -- sharing the office wireless comes immediately to mind.

Snow Leopard AAC improvements..

Reply #27 –
Quote from: IgorC on
But those clueless/careless average Joes would have to update their firmware to enable HE-AAC support. You understand that's impossible for that kind of people, don't you?


That is actually something that they can do.  However, I think that Apple needs to change the way iTunes operates so that a large message comes up saying "Hey, your iPod's firmware can be updated!  Do it!"  The firmware update notifications are in small text so they tend to not see them.

Quote from: IgorC on
Second reason why people wouldn't use HE-AAC for iPods is that there is only ~5% people using LC-AAC at low bitrate (=<80-90 kbit/s).  Rest of people who use LC-AAC >90 kbit/s won't jump to 64-80 kbit/s. http://www.hydrogenaudio.org/forums/index....showtopic=72265 . If you analyze some numbers around bitrates over internet  more than 95% didn't go lower than 128 kbit/s.


No but they could.  Still, those people would have to educate themselves on any new features added into iTunes/QuickTime and their iPods, test things out, etc.

Quote from: IgorC on
HE-AAC is a big NO-GO for music content on DAP. Its application is for low bandwidth.


I still think that people could benefit from HE-AAC on their iPod shuffle or nano.  It could double, triple, or quadruple the amount of songs that their 4-8GB players carry around.  Couple that with Apple marketing/word play and they could potentially carry around a bunch of tunes that would be high quality to them.

However, my main benefit of HE-AAC compatibility with iTunes is that I will finally be able to listen to the internet radio channels that I want to.  There are a bunch of internet radio channels out there that are broadcast at 32kbps and 48kbps.  They are terrible when playing inside of iTunes yet an HE-AAC capable program plays them back just fine (the differences are so bad that I can even ABX them on the cheap speakers in my HP Mini 110XP).

Snow Leopard AAC improvements..

Reply #28 –
Here is a quality-vs-bitrate graph calculated from some samples. The quality range 0-95 of the new encoder seems to correspond to the range 0-127 of the old encoder.

Snow Leopard AAC improvements..

Reply #29 –
I continued my tests with the new Q127 mode (~300kbit/s). The most problematic samples I have got: EBU SQUAM, Bibilolo, stuff from Kraftwerk & Tool, and whatnot, all perfectly transparent and not a single remaining problem track!

Can anybody confirm this with his ears? In my opinion the new QT AAC is now the best high bitrate codec available. Even lossyWAV still chokes on some samples.

Snow Leopard AAC improvements..

Reply #30 –
Quote from: rpp3po on
I continued my tests with the new Q127 mode (~300kbit/s). The most problematic samples I have got: EBU SQUAM, Bibilolo, stuff from Kraftwerk & Tool, and whatnot, all perfectly transparent and not a single remaining problem track!


Is the PTP sample transparent on the new true VBR mode? I found it to be really easy to ABX at iTunes Plus settings.
"I never thought I'd see this much candy in one mission!"

Snow Leopard AAC improvements..

Reply #31 –
I found it actually damn hard with the iTunes Plus setting (256kBit VBR constrained, highest quality)! But maybe I'm not focussing onto what you are hearing, what is it and what position/range? Have you tried the same version as the Snow Leopard encoder (Quicktime 7.6.3) for the iTunes Plus encoding?

Give those a try, maybe we have finally both found an encoder that we can unconditionally live with.

[attachment=5371:Show_Me_...nes_Plus.m4a][attachment=5372:Show_Me_...VBR_Q127.m4a]

Snow Leopard AAC improvements..

Reply #32 –
Quote from: rpp3po on
I found it actually damn hard with the iTunes Plus setting (256kBit VBR constrained, highest quality)! But maybe I'm not focussing onto what you are hearing, what is it and what position/range? Have you tried the same version as the Snow Leopard encoder (Quicktime 7.6.3) for the iTunes Plus encoding?


Mainly problems appear at the start with the secend synth drum beat. This sample has artifacts all over the place with LAME V2 - V0 and 320 CBR and fails badly at any other Mp3 encoder.

I tried both samples and managed to ABX the iTunes Plus at 15/17 and faild to ABX the Q127 ATM. Mainly because am not focussing, but the new Quicktime encoder does make a huge improvement on this sample, which can be ABXable at any Mp3 encoder at 320kbps CBR.

Quote from: rpp3po on
Give those a try, maybe we have finally both found an encoder that we can unconditionally live with.


Hopefully, so far the reported improvements clearly show that AAC still has plenty to offer. I wonder if Apple will port it to Windows or place it onto iTunes 9, which will be messy for foobar2000 and FLAC users though.
"I never thought I'd see this much candy in one mission!"

Snow Leopard AAC improvements..

Reply #33 –
Thank you for your results, /mnt! I share your conclusions that this looks promising for AAC.

If Apple is still going to withhold this from the Windows platform with iTunes 9, it might not take forever until the Nero developers catch up. They have proven their skills time and again. Let's just hope that their employer still allots them enough time for this.

Snow Leopard AAC improvements..

Reply #34 –
Quote from: rpp3po on
I'm also wondering what Apple's plans for HE-AAC could be. They usually don't bring out stuff like this just for the cause of it.
I think audiobooks could benefit from HE-AAC. It seems that the low bitrate of some audiobooks in iTunes is very annoying.

Snow Leopard AAC improvements..

Reply #35 –
Fcuk the high bitrate, I just replaced my whole AAC collection with the new Q127 encodings!  ~300 kbit/s is still about a third of my lossless average and thus the whole package still fits into an iPod classic.

And zero problem tracks (MP3 has 100's, Nero a few) means zero "lossless sorrow" - you know, the feeling, that albeit your ABX verified rationale that a lossy codec is fine for 99% of your material, this one special track might have sounded better, if you had just taken the lossless version with you.

Snow Leopard AAC improvements..

Reply #36 –
Quote from: rpp3po on
The new XLD version that just came out supports it already!

I have played around with it a little bit and made some interesting observations:

The new Quicktime includes the first AAC encoder that masters the infamous Emese sample transparently! For my ears at least. Nero was never able to handle it even at the highest bit rates and also LAME and Vorbis have always failed badly. This only works at the highest setting of Q127; at Q122 with only slightly less bit rate I can ABX it instantly, which is kind of strange. ABR 320 is also transparent.
Interesting - I can easily ABX emese at Q127:

Code: [Select]
WinABX v0.42 test report
09/06/2009 10:30:47

A file: H:\music\encoder_test_clips\emese.wav
B file: H:\music\encoder_test_clips\emese.qt763.q127.wav

Start position 00:00.0, end position 00:07.0
10:31:30    1/1  p=50.0%
10:31:33    2/2  p=25.0%
10:31:36    3/3  p=12.5%
10:31:40    4/4  p=6.2%
10:31:43    5/5  p=3.1%
10:31:46    6/6  p=1.6%
10:31:50    7/7  p=0.8%
10:31:54    8/8  p=0.4%
10:31:57    9/9  p=0.2%a
10:32:00  10/10  p< 0.1%
10:32:03  11/11  p< 0.1%
10:32:06  12/12  p< 0.1%
10:32:09  13/13  p< 0.1%
10:32:12  14/14  p< 0.1%
10:32:15  15/15  p< 0.1%
10:32:20  16/16  p< 0.1%
Just to make sure we're on the same page, here's the "Tool" string from the m4a file as encoded by XLD and read in foobar2000:

Code: [Select]
X Lossless Decoder, QuickTime 7.6.3, True VBR Quality 127
and the "Encoder Quality" setting was at Max.  I used fb2k to convert to WAV for the ABX test, but I could easily detect the same artifacts when listening in iTunes under Snow Leopard (and I triple-checked that I had indeed encoded from the original emese.flac, as opposed to re-encoding from an MP3 or some other bonehead thing like that).

HOWEVER, I cannot ABX the Show Me Your Spine sample at just Q90 (I CAN easily ABX LAME 3.98.2 V2), which is not too surprising as I've never been very sensitive to pre-echo artifacts.
"Not sure what the question is, but the answer is probably no."

Snow Leopard AAC improvements..

Reply #37 –
Interesting! I think you have just better hearing than me regarding that specific kind of artifact.

What I usually hear with Emese is a quite loud chirping artifact. Nero has it, LAME is terrible, only with the new Q127 it is completely gone.

Snow Leopard AAC improvements..

Reply #38 –
The artifact I'm hearing is small bursts of a "papery" kind of noise, dead center (i.e. not louder in one channel), and to me QT @ Q90 is actually WORSE than LAME 3.98.2 V2.

FWIW, I'm using Sony MDR-V6 headphones being driven by a Symetrix 304 headphone amp from an M-Audio Transit.
"Not sure what the question is, but the answer is probably no."

Snow Leopard AAC improvements..

Reply #39 –
Quote from: Vietwoojagig on
Quote from: rpp3po on
I'm also wondering what Apple's plans for HE-AAC could be. They usually don't bring out stuff like this just for the cause of it.
I think audiobooks could benefit from HE-AAC. It seems that the low bitrate of some audiobooks in iTunes is very annoying.


heh, i had somehow gotten it into my head that that was so obviously what it was developed for that it hadn’t occurred to me that anybody would try to use it for anything other than vocal recordings. [...until i started reading this thread, that is.]

Snow Leopard AAC improvements..

Reply #40 –
I've been testing even further later these days and i just realized that even Q127 for me is not good enough (sounds flat to my ears),compared to Nero AAC q 0,95 for instance. I know Nero may have problems with some samples but i'm sure they will be solved in the next version,but now to my ears i feel more pleased listening to Nero AACs than Apple's ones. Please it's important to note that this is just an opinion of mine,nothing objective like ABX and stuff like that

Snow Leopard AAC improvements..

Reply #41 –
Without blind testing such statements are meaningless. Sounding "flat" is no kind of artifact, that high bitrate lossy encoders would show, anyway. You can get all kinds of artifacts, but usually not that. Flat sound is a common "result" from non level matched tests.

Just give it a try, you don't need anything else than a full Foobar installation for a perfect ABX setup.

Snow Leopard AAC improvements..

Reply #42 –
Sounding "flat" and "my ears are more pleased" really don't mean anything when describing an encoder.  It is all placebo affect unless you actually conducted blind ABX tests.  I imagine that most people would have a really hard time distinguishing QuickTime (at q127) and Nero (at -q0.95) from the source lossless material due to their extremely high bitrates.  That is why blind ABX tests are needed in order to make such claims.  Otherwise people can go around making all sorts of outrageous claims without any experience or tests.

Snow Leopard AAC improvements..

Reply #43 –
 

Okay, most of this is new to me and I'm confused about something...

So now q127 TrueVBR averages 300kbs? I thought, with the previous Quicktime, when using q127 TrueVBR you were basically requesting that the optimum/best (however it should be worded) bitrate be used, no more, no less. So if the new q127 TrueVBR averages 300kbs doesn't that mean Quicktime is forcing an average bitrate instead of using only what is needed?

Snow Leopard AAC improvements..

Reply #44 –
so itunes 9 + quicktime 7.6.4 has also been released to older version Leopard. I've seen that new true vbr is also on XLD in Leopard now

Snow Leopard AAC improvements..

Reply #45 –
Quote from: fferd on
So now q127 TrueVBR averages 300kbs? I thought, with the previous Quicktime, when using q127 TrueVBR you were basically requesting that the optimum/best (however it should be worded) bitrate be used, no more, no less. So if the new q127 TrueVBR averages 300kbs doesn't that mean Quicktime is forcing an average bitrate instead of using only what is needed?

actually, my assumption is that what rpp3po was referring to was the overall average bitrate of an entire library of audio files... and based on even my own experience, I think it's safe to say that the average bitrates of individual files all but regularly fall around 300kbps (I've seen anywhere in the range of 150-350kbps average bitrates for some of my files during testing at -Q 127).

on a side note, @rpp3po- have you noticed that files with vastly different mixes in each channel (i.e., a majority of albums from the 60's and 70's, many classical albums, etc.) seem to have noticeably larger average bitrates than those with similar mixes in each channel?  I've been trying to pinpoint exactly what sounds make the bitrate spike/drop, but so far the most frequent causes for this seem to be the stereo width and/or the difference between mixes in each channel... just thought I'd see if you'd noticed similar behavior with your audio files.
Archive- FLAC (-v 8)
Portable- QuickTime AAC (True VBR/-q 77)

Snow Leopard AAC improvements..

Reply #46 –
Regarding the ipod + HE-AAC discussion.. What I would like to know is whether those of us who have their library stored in lossless format and convert the songs to a lossy format only for portable players really need higher quality than HE-AAC has to offer. I mean yes the difference is there and it can be heard, but is it possible to distinguish the difference when you are outside on the road or in the gym etc? The only tests I've done so far are in complete silence. But I don't think that I would be able to tell the difference in a noisy enviroment.

I'm not sure but I believe that the HE-AAC format is the best for portable players since you can put more songs in your player and have a quality more or less the same in a noisy enviroment. What do you think?

Snow Leopard AAC improvements..

Reply #47 –
Quote from: jmcguckin on
Quote from: fferd on
So now q127 TrueVBR averages 300kbs? I thought, with the previous Quicktime, when using q127 TrueVBR you were basically requesting that the optimum/best (however it should be worded) bitrate be used, no more, no less. So if the new q127 TrueVBR averages 300kbs doesn't that mean Quicktime is forcing an average bitrate instead of using only what is needed?

actually, my assumption is that what rpp3po was referring to was the overall average bitrate of an entire library of audio files... and based on even my own experience, I think it's safe to say that the average bitrates of individual files all but regularly fall around 300kbps (I've seen anywhere in the range of 150-350kbps average bitrates for some of my files during testing at -Q 127).


I can confirm that.


Q127 was indeed intended to provide the optimum/best bitrate for all material. Before the last QT release the developers estimated that this setting, resulting in a whole collection average of about ~190kbit/s, was sufficient. My own testing confirmed this. Still very few tracks remained, that were ABXable, but not necessarily bad. So they included a mode, that vastly scales up the bitrate at the slightest doubt for the paranoid among us. You get the same perceptual quality (transparency!) for ~99% of your collection and <1% less ABXable samples, at the price of 1/3rd more overall bitrate. The old (~190) behavior is still in the code. Just use a lower Q value.

Quote from: jmcguckin on
on a side note, @rpp3po- have you noticed that files with vastly different mixes in each channel (i.e., a majority of albums from the 60's and 70's, many classical albums, etc.) seem to have noticeably larger average bitrates than those with similar mixes in each channel?  I've been trying to pinpoint exactly what sounds make the bitrate spike/drop, but so far the most frequent causes for this seem to be the stereo width and/or the difference between mixes in each channel... just thought I'd see if you'd noticed similar behavior with your audio files.


I haven't actually looked at what content causes what bitrate that much. I just went through some of my problematic tracks and evaluated if the new encoder is an improvement. One thing I noticed it that the new Q127 doesn't drop bitrate for classical music as much as other VBR encoders, when it is a high quality recording with content in the complete spectrum. Low quality monoaural tracks from the 50's indeed scale as low as 150 kbit/s (~100kbit/s with the old Q127).

Quote from: Ilu on
Regarding the ipod + HE-AAC discussion.. What I would like to know is whether those of us who have their library stored in lossless format and convert the songs to a lossy format only for portable players really need higher quality than HE-AAC has to offer.


HE-AAC does not target transparency! It just targets to be the better sounding "good enough" encoder for very low bitrates. For transparency >=128 kbit/s LC-AAC is still the much better choice.

Snow Leopard AAC improvements..

Reply #48 –
Well I think its a weakness to have to use Q127 for codecs supposed to be transparent much lower. Not from a space POV, but from a quality control one. How can one conclude that Q127 is optimal ? AAC is a general purpose codec without specific *transparency* tunings for mid-high bitrates ??. I could be wrong but i think even vorbis is in a similar situation. There are quality settings that produce smaller~higher bitrate in the scale , Yet you don't know at what quality / bitrate a quality-control mechanism kicks in or works well for more difficult material. In that case raising the bitrate doesn't help much and you have to throw bits like crazy. its like saying back in the day that Musepack needed 'braindead' or quality 10 where in fact quality 5 (175 k) had enough control-mechanism for most material and quality 6 (205k) would provide extra safety margin.

Snow Leopard AAC improvements..

Reply #49 –
When you look at the graph by Nao, the bitrate scales almost linearly. There is no specific point where a "quality control mechanism" would kick in. They just made 'insane' parameters available at the upper end and mapped the new range to the old 0-127 scale. I appreciate this. A lower Q value still gives you the excellently performing and sane ~190kbit/s average known from older versions. The new mode improves freedom of choice for HA nuts, who cannot be satisfied with 99% collection transparency and 1% sounding fine but failing direct A/B comparisons. Just because there is an insane setting available, doesn't mean that a sane ~190 average would be inferior.