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: AAC encoder improvements (Read 4790 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

AAC encoder improvements

Hi guys,

Last week while I was revising in the library, I was listening to Coldplay's Parachutes album. I was listening on my Mac, where my collection is almost entirely ~160kbps AAC. I'm guessing there's about half-and-half split between iTunes/CoreAudio and Nero AAC encodings in there, the Nero encodings being the oldest (transferred across from a PC when I got the Mac in 2006).

The thing is, I couldn't help noticing that it sounded a little bit naff. It wasn't too distracting, and I was still enjoying the music, but somehow it sounded a little muddy. I checked the tags, which confirmed my suspicion: it was encoded with Nero AAC in 2006. Could two years really make that much difference in audio encoding quality? I'd never noticed this in my more-recent encodings, so maybe encoders have improved.

I decided to do a little test. I grabbed the latest Nero Digital encoder (which was tricky, as it's not visible on the site at the moment), and re-encoded the first track from a FLAC version I have stowed away. I did the same using Max, which uses the CoreAudio encoder (basically iTunes). I've always trusted encoder authors, so I used the Nero default (q=0.5) and CoreAudio is set for 160kbps VBR. This gave me a Nero encoding slightly bigger than iTunes (ABR, according to QuickTime = 163 kbps).

Using a very unscientific comparison, between FLAC, Nero '06, Nero '08 and CoreAudio, I came to the following conclusions:
  • The Nero '06 encoding was miserable compared to all other versions. The drums, in particular, sounded closed-in, undefined and mushy.
  • There was barely a noticeable difference between the Nero '08 and iTunes encodings, and little difference from the FLAC too!
  • The iTunes encoding just had an edge for me: I thought the cymbals sounded closer to the FLAC version compared to Nero '08, but there was nothing to tell them apart otherwise.

I realise this is all very subjective, but the point of this post is not to sell a particular encoder. Just from this quick informal listening, I've noticed a significant improvement between two versions of the same encoder. It just goes to show that development is still very much worthwhile, even to improve audio quality at a high bitrate, and that there are benefits from using a modern encoder.

I think this raises an interesting issue though. Would you consider re-encoding your library if you used an 'old' encoder? Would you go as far as doing this periodically, perhaps as major encoder updates are released? I'm wondering whether I should do this, especially for my now-vintage '06 encodings.

AAC encoder improvements

Reply #1 –
Personally, that's why I stick with mp3 for general use. It is an older, but well tested and well supported codec. I also know that any music encoded with 3.90.2+ sounds fine (to me). I can't be bothered to re-encode to AAC/Vorbis etc. I know what you mean though. With each new version of LAME, I wonder whether the improvements are worth the hassle for any noticeable gain.

AAC encoder improvements

Reply #2 –
I am going to suggest this before someone else does: conduct blind ABX tests (unless that is what your test was).  That is the only way to distinguish if you are actually hearing these differences or if it is all placebo.  Otherwise your audio quality claims violate the TOS here at hydrogenaudio.  Additionally, the most recent Nero encoder came out in 2008 and not 2009.

Your question is pretty interesting though.  I used to upgrade my lossy library about once every 2-4 years depending on the most recent encoder.  I started out using Lame 3.90.3 and didn't bother upgrading until Lame 3.97.  I then decided to upgrade to 3.98.2 since it had been a couple of years.  Now I am using Nero AAC and probably won't re-encode my lossless library for another few years unless the next Nero AAC encoder is able to provide substantial increases in quality over the current version.

AAC encoder improvements

Reply #3 –
Quote from: simonh on
Personally, that's why I stick with mp3 for general use. It is an older, but well tested and well supported codec.
[cut]
I know what you mean though. With each new version of LAME, I wonder whether the improvements are worth the hassle for any noticeable gain.

Do you think that MP3's already reached its peak audio quality, or at least near it? I'd be interested to know how flexible the MP3 spec is - could a new encoding technique (perhaps undiscovered as yet) be implemented in MP3 encoding, for example? What is it that authors actually improve with each version anyway; are new techniques involved, or is it just fine-tuning parameters?

Quote from: kornchild2002 on
I am going to suggest this before someone else does: conduct blind ABX tests (unless that is what your test was).  That is the only way to distinguish if you are actually hearing these differences or if it is all placebo.  Otherwise your audio quality claims violate the TOS here at hydrogenaudio.

This was just an informal test, and I tried to make it clear that my comments were subjective and unscientific. I'm not trying to claim one codec is better than another, the point was more about how codecs have improved. I dare say that there's test data available to prove this anyway. I'll see if I can find it later. Also, I've been considering a more comprehensive ABX for ages, to see if my current encoding practices are the best for me.

Quote from: kornchild2002 on
Additionally, the most recent Nero encoder came out in 2008 and not 2009.

Good point, I've edited my original post.

Quote from: kornchild2002 on
I used to upgrade my lossy library about once every 2-4 years depending on the most recent encoder.  I started out using Lame 3.90.3 and didn't bother upgrading until Lame 3.97.  I then decided to upgrade to 3.98.2 since it had been a couple of years.

Did you do this because you noticed a difference, or just because it seemed like the right thing to do?

AAC encoder improvements

Reply #4 –
Hi,

Just curious, what bitrate do you encode your library with previous LAME and current Nero?  As I want to decide the bitrate LAME/Nero for my music library renewal from 128k AAC iTunes. (I mean from the source not 128k to higher bitrate transcoding, of course)

csp

Quote from: kornchild2002 on
...

Your question is pretty interesting though.  I used to upgrade my lossy library about once every 2-4 years depending on the most recent encoder.  I started out using Lame 3.90.3 and didn't bother upgrading until Lame 3.97.  I then decided to upgrade to 3.98.2 since it had been a couple of years.  Now I am using Nero AAC and probably won't re-encode my lossless library for another few years unless the next Nero AAC encoder is able to provide substantial increases in quality over the current version.

AAC encoder improvements

Reply #5 –
Quote from: smiler on
Did you do this because you noticed a difference, or just because it seemed like the right thing to do?


I did it for both reasons.  I wanted to support the developers by using their products, even if they were free.  I also did it because my own ABX tests show improvements over the previous releases.

Quote from: csp on
Just curious, what bitrate do you encode your library with previous LAME and current Nero?  As I want to decide the bitrate LAME/Nero for my music library renewal from 128k AAC iTunes. (I mean from the source not 128k to higher bitrate transcoding, of course)

csp


I decided to go with Nero at -q0.50 after migrating from Lame at -V 2.  I find that -q0.50 files are normally around 0.5-1MB smaller (each) than Lame -V 2 files produced with either Lame 3.97 or 3.98.2.  I had great difficulty ABXing Nero AAC at that bitrate.  I could have stepped down to -q0.40 or -q0.45 for portable listening but I just purchased a 120GB iPod classic and my lossy music consumes about 79GB of that space at -q0.50.  My library would decrease to about 75GB a lower setting.  That and I could still ABX some samples at -q0.45 if I tried really hard.  The best thing that you can do is conduct your own ABX tests to determine what is best for you.  I don't have your ears, your music, or your equipment.  That is why the only true judge that you can trust is yourself.

AAC encoder improvements

Reply #6 –
Quote from: smiler on
This was just an informal test, and I tried to make it clear that my comments were subjective and unscientific. I'm not trying to claim one codec is better than another, the point was more about how codecs have improved. I dare say that there's test data available to prove this anyway. I'll see if I can find it later. Also, I've been considering a more comprehensive ABX for ages, to see if my current encoding practices are the best for me.

Even then, you should conduct a small ABX test (or is it ABC/HR?). That is the only way to make sure there is a difference between the old and new Nero encoding for your ears. And then you can share the results with us 

AAC encoder improvements

Reply #7 –
I think than an ABC/HR test would be the right one if the OP planned on testing two lossy files at once.  I don't know the specifics of that test but I think that type of test is used for more than one lossy file.

AAC encoder improvements

Reply #8 –
Quote from: smiler on
I grabbed the latest Nero Digital encoder (which was tricky, as it's not visible on the site at the moment)

Until Nero AAC page is again back online, please use this one: http://www.audiocoding.com/nero_aacenc.html

Quote from: smiler on
Quote from: simonh on
I know what you mean though. With each new version of LAME, I wonder whether the improvements are worth the hassle for any noticeable gain.
Do you think that MP3's already reached its peak audio quality, or at least near it?

There is some improvement even from version 3.97 to 3.98.2: http://www.hydrogenaudio.org/forums/index....showtopic=67529
Though it is questionable if there is big enough improvement for your ears and worth the hassle.

Quote from: smiler on
The thing is, I couldn't help noticing that it sounded a little bit naff. It wasn't too distracting, and I was still enjoying the music, but somehow it sounded a little muddy. I checked the tags, which confirmed my suspicion: it was encoded with Nero AAC in 2006. Could two years really make that much difference in audio encoding quality?
Quote from: kornchild2002 on
I am going to suggest this before someone else does: conduct blind ABX tests (unless that is what your test was).  That is the only way to distinguish if you are actually hearing these differences or if it is all placebo.  Otherwise your audio quality claims violate the TOS here at hydrogenaudio.
This was just an informal test, and I tried to make it clear that my comments were subjective and unscientific. I'm not trying to claim one codec is better than another, the point was more about how codecs have improved. I dare say that there's test data available to prove this anyway.

We are still trying to improve our encoder, but it is very hard to improve it further at very high bitrates, because there are not many people that can hear differences at these bitrates (most people wouldn't hear differences even at q 0.4, just check the above public listening test and you'll see that many people don't hear difference with mp3 at the equivalent bitrates). Another difficulty is that different people are sensible to different distortions, and there are no personalized profiles.
Your findings, and findings from others, help us determine if we are going to the right direction. But to really help us, we need ABC/HR tests on few samples, or at least on the one where you hear biggest difference. Ultimately public listening test would be the best thing, but it requires too much effort.

Edit: online again! 
http://www.nero.com/eng/technologies.html
http://www.nero.com/eng/technologies-aac-codec.html