Nero Weirdness?

[Shandra]

I am not sure if that question belongs in this subforum or better would be placed in the GUI subforum, but as the context is audio encoding....

At the moment I am somewhat out of ideas to nail down the following problem and maybe the answer is already known (and I am just unable to find it):

During some DVD Backups I was doing with MeGUI I came to the situation that some Nero AAC encodes on the same track with a lesser quality setting produced larger files then a higher quality setting and did some further (and quick) investigation thereafter. The troublesome quality setting appears to be from -q .4 to -q .37, here are some results of mine so far:


used for encoding: MeGUI (Decoder: NicAudio; Keep Original SampleRate; Normalize Peaks)

Source: 6channel DTS 755 kbps, 539 MB

Nero 1.5.4.0 Setting: Filesize (Overall Bitrate/Max. Bitrate) (as reported by MPC-HC)
<Original channels kept>
-q 0.42: 288 MB (402/457)
-q 0.40: 242 MB (337/402)
-q 0.39: 290 MB (404/479)
-q 0.38: 280 MB (384/463)
-q 0.37: 217 MB (302/367)
-q 0.36: 208 MB (288/352)
<downmix to stereo>
-q 0.40: 90 MB (124/161)
-q 0.38: 104 MB (144/186)
-q 0.36: 71 MB ( 96/129)

Source: 6channel AC3 384 kbps, 313 MB
<Original channels kept>
-q 0.40: 268 MB (328/403)
-q 0.38: 308 MB (384/463)
-q 0.36: 229 MB (280/353)

Source 2channel AC3 192 kbps, 156 MB
-q 0.40: 95.4 MB (115/153)
-q 0.39: 114 MB (139/183)
-q 0.38: 110 MB (132.3/176)
<Normalize Peaks deactivated>
-q 0.39: 110 MB (132.3/181)
<and via DirectShow instead of NicAudio>
-q 0.40: 87.3 MB (105/150)
-q 0.39: 104 MB (127/180)
-q 0.38: 101 MB (122/173)
-q 0.36: 68.1 MB (81.9/130)

Convert via Foobar from 42.1 MB 2channel Wavpack Source, 462 kbps
-q 0.50: 10.8 MB (117/199)
-q 0.42: 8.68 MB (93.7/151)
-q 0.41: 8.37 MB (90.3/145)
-q 0.40: 8.08 MB (87.1/140)
-q 0.39: 7.11 MB (76.5/148)
-q 0.38: 6.44 MB (72.0/143)
<via MeGUI>
-q 0.40: 8.08 MB (87.1/140)
-q 0.38: 6.80 MB (72.0/143)


The Wavpack Source is from an AudioBook, the other ones where DVD Sources and as said: I am out of Ideas of the cause and whether this is a Nero Problem or somewhere in the chain I use for encoding.

[Selur]

you also might waht to check if you output is LC/HE/HE+,...

[Shandra]

Selur: For the first (DTS) Source I checked a wider range and the break between LC and HE was -q 3.0 (3.1 and above LC (*1)). At the moment I am searching through my DVDs for a suitable (short) source to do some more focused tests. And all Episodic DVDs I have at hand (like Robin of Sherwood BBC and such) are Dolby Enhanced or Plain Stereo (from original Mono Source) and I would prefer a short Multichannel Source wich either have a corresponding Stereo only Track or where I can downmix to plain Stereo Wav and then could put both MC and Stereo through some encodes for comparison.

What I have done in addition for the above results so far is to countercheck the results on my secondary PC and my Thinkpad and finally on another ones Windows 7 Home (64bit) Machine - and the results where consistent with the ones from my Main PC.
And I also downloaded the DTS decoder PlugIn for Foobar (1.1.10 by the way) and checked the DTS Source for -q 3.8 and -q 4.0 in Foo and the results are similar to the MeGUI encodes (-q 3.8 higher bitrates and size then -q 4.0).

As said, I am still searching for a good test sample and I maybe should also scan my archives for an older version of Nero to test against. But somehow I am now convinced that I am not doing anything "wrong" here and it actually is the latest(?) Nero which is acting strange. What is puzzling me still is the expected behavior of Nero for the AudioBook Wavpack Source in both MeGUI and Foobar and the not-expected behavior for the DVD DTS and AC3 Sources (So maybe it is related to Multichannel and or Dolby Information - but as the DTS downmix to stereo also showed the effect - sigh....).

Edit: *1: All results in 1st Post where LC encodes.

[hello_hello]

Well I'll be.....

I just happened to have a small DTS sample on my PC when I saw this thread so thought I'd give it a whirl using foobar2000. I tried it with both the DTS file and an MP3. (Nero 1.5.3.0)
The first file size is DTS without mixing down to stereo, the second is DTS mixing down to stereo. The third is the MP3.

Q.36: 2.1MB, 657.8KB, 3.2MB
Q.37: 2.2MB, 687.2KB, 3.3MB
Q.38: 2.8MB, 1022KB, 3.5MB
Q.39: 2.9MB, 1060KB, 3.6MB
Q.40: 2.4MB, 883.3KB, 3.9MB
Q.41: 2.5MB, 914.7KB, 4.0MB

So why do recodes of an MP3 produce "expected" file sizes, while re-encoding DTS doesn't?
I checked quite a few of the encodes with MediaInfo and it seems to be happy they're all LC.

Edit: As an afterthought I converted the DTS file to a multichannel wave file and then converted the wave file to AAC. Same result as when converting the DTS file directly. I even tried downmixing the DTS file to a stereo wave file then converting the stereo wave file to AAC. Give or take 0.1KB or so, the result was still the same. How curious....

[Shandra]

So it is the same with at least one older Version (Haven't found one on my HDDs) ;/ One Parameter that may be of interest is the sampling rate - I haven't thought about that one before - but the "problematic" samples where 48kHz, whereas the ones that result in expected bitrates where 44.1kHz.. Ok a quick Test

Source AC3 48kHz, 6 channel 192kbps (58.7 MiB)
via MeGUI (NicAudio, Keep Channels, no further Options)
Q.40: 113/158 kbps LC (34.9 MiB)
Q.38: 132.3/183 kbps LC (40.2 MiB)

+ Downsample to 44.1kHz:
Q.40: 108/153 kbps LC (33.3 MiB)
Q.38: 88.8/150 kbps LC (27.6 MiB)

Yup - SamplingRate seem to be the culprit here

[hello_hello]

I never thought about sampling rate as I assumed your AC3 test files wouldn't have used a 44.1k sampling rate. If they didn't then why is it still only DTS??

What the current/latest version of the Nero AAC encoder? I didn't know I was using an older version.

[Shandra]

The latest(?) Nero is 1.5.4.0
Yes - my AC3 Files were 48kHz (DVD Source), and they showed the same behaviour as the dts file - what didn't showed the anomaly with Nero was a WavPack Clip ripped from an AudioBook (CD) (44.1 kHz) and the downsampled to 44.1kHz AC3 in my latest post.
But a similar Threat (http://www.hydrogenaudio.org/forums/index.php?showtopic=92756) over at the Hydrogenaudio Forum (Which I came across after I opened this threat here - so no crossposting by intend, just same topic (different users/forum) by coincidence) indicates that sampling rate may not be the cause.

[hello_hello]

Ah yes, I see your AC3 conversion gave the same result. I must have looked at your test encodes incorrectly.
Thanks. I found Nero 1.5.4.0, but at least I could confirm the behavior's not confined to the latest version.
I'll admit without this thread I never would have discovered the anomaly as I only ever encode using a single quality setting (q.50), but if I've got some time to kill tomorrow I might play around with some more encodes to see if I can come up with any clever ideas, although I won't hold my breath waiting to discover something everyone else happens to have missed so far.

[Shandra]

Not that it solves all of the questions at hand (especially why for some sources it works as expected and for others not) - but some friendly user in the linked thread over at HA has posted a link to another thread there that contains an "(semi-)official" answer to the observation

[hello_hello]

Interesting. I understand the answer but don't really know enough about lossy encoding to understand why an encoder works that way.

"For different expected bitrate ranges the encoder uses different parameters that produce different results."

When choosing a quality setting you're generally after a quality of "x" first, and a bitrate range second, so I guess it's natural to assume a higher quality setting will adjust the parameters to give you a higher bitrate. I assume though as the parameters change the quality can still increase while the bitrate can sometimes decrease??

It does seem counter-intuitive (as far as the results of your tests go) but I wonder if it's safe to assume a quality setting of 0.40 (for example) still produces a higher quality than a setting of 0.38, even if the bitrate is lower? You'd assume (hope) so, although I wonder if it's possible to hear a difference anyway.

[MatLz]

I did notice problems since version 1.5.1 iirc.
1.3.3 works fine.
As you said, it mainly depends on samplerate, 32Khz has also problems.

There are "holes" and "jumps" in the quality range.

But the biggest problem is that if you are near to these problematic quality values, you will sometimes get a tooooo smooth bitrate curve wich is near to be CBR instead of VBR.

[Shandra]

... but I wonder if it's safe to assume a quality setting of 0.40 (for example) still produces a higher quality than a setting of 0.38, even if the bitrate is lower? ...

Good Question - and the answer somehow open more question then it actually answers. I am no audio expert only thing I can vaguely remember is old Lame Preset Times and with various Presets different "tweaks" were used (but can well be that I mix up some memories there and just confuse new/old preset switches and combinations of -q/-v, etc.). But Nero here is the first where I notice lower quality settings with higher file size. Naturally one would assume (within one encoder) that higher size means better quality; in difference to compression algorithms like zoo/zip/rar/etc. where we would talk en-/decoding times&horsepower needed versus efficiency - maybe there's some analogy we could use from there in this case, but that is pure speculation on my side.
Back to quality - Ever tried to ABX an 120min AudioClip? That's no Fun with Foobar and then I just have Stereo Equipment at home... but there I must confess that I am unable to hear any differences (at least for .38 to .40, as said the ABX is no fun for those sizes and I just left it there), whereas some friends of mine on our video evenings seem to have noticed some artifacts plus something they described as "whoppling" or loudness oscillation on my encodes below q somewhere lower then ~.48/.50 which I never noticed - though among my friends I am among those which still have some good hearing abilities when it comes to higher frequencies and our age - I still hear those -maddening- devices they use to drive weasels away (could as well be the ages old HiFi Equipment we use there plus the room-position of the observers and the combination with quality settings is pure coincidence or along the line of "Dolby Stuff" - Duno, on such an evening we simply want to enjoy the night & no tests of codecs/decoders/etc.)

But yes, somehow the observations about Nero has lessened my trust in it for backup purposes, but as long as I am not hearing those artifacts and if a higher quality setting produces a lower filesize - hey that is bitrate for the videostream ;)

MatLz: 1.3.3, good to know. Could be interesting to get some changelogs from that to newer versions.