What is your instance regarding Musepack SV8 switches? Options

[Ruse]

Yes, this is well thought out in itself, but it would seem to me this is because of poor managemnt of the Lame developemnt process, not a fault of the switch availability.

Klemm has control over the developemnt of mpc, not a developemnt group (as far as I know). The switches in mppenc are all core functionality switches, and they are far less numerous than in Lame. It should be quite feasable to provide for user adjustment of these small group of switches in a failsafe manner, and promote the presets as the usual manner in which mppenc is used.

[gdougherty]

Ruse and JohnV, I'll try to answer all your questions.

I was using the latest version of Frank's Encoder at the time, I don't recall the date, but it was sometime back in February if I'm not mistaken.

Using Audiograbber (I hadn't yet switched to EAC) I ripped track 1 to wav off the Time Out album which was brand new and still in mint condition.

I encoded and decoded back to wav using standard, xtreme, and insane presets (I mispoke and said it was braindead before)

Using Cool Edit Pro I lined up all four files and cut them into 30 second passages. I randomly selected the various 30 second blocks and arranged them end to end to create the entire song from various 30 second samples. I wrote the splices back out as a wav file.

I then burned a CD with the original wav as track 1, the three preset encodes, and then the splice version.
That CD is what I took over to ListenUp.

Equipment List:
Mark Levinson No. 39 CD player ($6K)
Mark Levinson No. 360S Preamp ($7500)
Mark Levinson No. 336 Dual Monoblock Amplifier ($10K)
B&W Nautilus 800 ($16K)
Not including the speaker cable, which was probably another $500 at least, that's $39,500. The CD player and Preamp have all kinds of features designed to give the best sound possible, ie. jitter reduction and all kinds of technomarketing hype.

We dropped the CD in, and I first played the original wav track as a baseline.

I then played the splice track and asked him if he could identify a difference between the splices. At the time I had the splice order on a piece of paper that I didn't show him, and he correctly identified the sections as better or worse than the preceeding section through the whole song. Statistically it could have been a fluke, but without guidance from me, he identified 13 different sections. I'm not saying it was a blind listening test, but it would be repeatable, and it was enough to convince me that there was something to what he was hearing.

As for me, I couldn't hear any hugely distinct differences and any differences I could "hear" might have been perception from having the list of splices in front of me. I wouldn't exactly describe my own hearing as "golden" though, far from it really, so I don't think my own opinion weighs much.

[gdougherty]

Originally posted by Ruse
Yes, this is well thought out in itself, but it would seem to me this is because of poor managemnt of the Lame developemnt process, not a fault of the switch availability.

Klemm has control over the developemnt of mpc, not a developemnt group (as far as I know). The switches in mppenc are all core functionality switches, and they are far less numerous than in Lame. It should be quite feasable to provide  for user adjustment of these small group of switches in a failsafe manner, and promote the presets as the usual manner in which mppenc is used.

Very much agreed. Lame has more significant issues than the number of switches.

We're not asking that Frank add any switches either, we're simply asking that we not loose any functionality from the previous version. If you want to drop some switches, how about dropping insane, since by Frank's admission it was actually created to give the best "looking" output with spectral analysis instead of giving the best "sounding" output.

[maciey]

@gdougherty: wasn't the CD mastered (normalized) to 98% or like this? If that - this could be the cause - the listener may have (with his good hearing) indentified clipped (encoded-decoded) vs. non-clipped (original) parts... so maybe run a replaygain analysis on the encoded vs. non-encoded track?

Another thing that I think of is that You could repeat this test (if you can't convince your listener to ABX)wiht uneven splice lengths - not all splices 30 sec but, say, first 15 , second 23 etc...

[Dibrom]

Originally posted by Ruse
Yes, this is well thought out in itself, but it would seem to me this is because of poor managemnt of the Lame developemnt process, not a fault of the switch availability.

But it is because of the fact that these switches are maintained (to the detriment of other tasks since this is wasteful in both time and effort), and not dropped when they are unnecessary (same as many of the MPC switches) that continues this situation. Yes that is poor management, but the thing is, you people are complaining about Frank doing the right thing here by removing these switches -- a sign of good management.

[b]Klemm has control over the developemnt of mpc, not a developemnt group (as far as I know).

Andree still has some say and I believe maybe 1 or 2 other people do also.. not entirely certain on that, but MPC isn't solely Frank's project.

[b]The switches in mppenc are all core functionality switches, and they are far less numerous than in Lame.

How do you define "core functionality" switches? Is something like the adjustability of the adaptive noise shaping a "core switch"? How about the ath curve or something else? If so then so are the ath adaptive switches, the noise shaping switches, the psymodel switches, etc in LAME. There's really no difference here, the switches in LAME also affect the "core", if you define that as the encoding engine itself.

Granted, they are less numerous now, but things get out of hand over time. The way to prevent this is to keep things in check and relevant as you go. I think this is what Frank wants to do by removing some of the switches, and I think it is a good thing.

[b]It should be quite feasable to provide  for user adjustment of these small group of switches in a failsafe manner, and promote the presets as the usual manner in which mppenc is used.

I do agree on this. I think the best suggestion so far as been to adopt a system similar to the vorbis -q scale.

[gdougherty]

Originally posted by maciey
@gdougherty: wasn't  the CD mastered (normalized) to 98% or like this? If that -  this could be the cause - the listener may have (with his good hearing) indentified clipped (encoded-decoded) vs. non-clipped (original) parts... so maybe run a replaygain analysis on the encoded vs. non-encoded track?

Another thing that I think of is that You could repeat this test (if you can't convince your listener to ABX)wiht uneven splice lengths - not all splices 30 sec but, say, first 15 , second 23 etc...

I made no modifications to the original wav, there were only the encoding steps after ripping the wav. Time Out was made before the days of heavy mastering compression and produces no internal clipping errors. Correctly identifying 13 splices, even or not, is good enough to convince me that it's more than a fluke. I'd feel a bit odd going back and asking him to repeat the proceedure, though I do love listening to the equipment. I'm really just throwing it out there as an example, as I said, it was enough for me.

G

[Ruse]

Originally posted by gdougherty
Statistically it could have been a fluke, but without guidance from me, he identified 13 different sections.  I'm not saying it was a blind listening test, but it would be repeatable, and it was enough to convince me that there was something to what he was hearing.

As for me, I couldn't hear any hugely distinct differences and any differences I could "hear" might have been perception from having the list of splices in front of me.  I wouldn't exactly describe my own hearing as "golden" though, far from it really, so I don't think my own opinion weighs much.

You say he correctly identified 13 different sections: how many sections were there in the splice? Were the encoder quality presets randomised, or did you start with the wav and work systematically through the quality presets --standard, --xtreme, --insane?

Did he identify the full length, unspliced encoded tracks as being inferior to the wave?

[Frank Klemm]

Originally posted by Dibrom

But it is because of the fact that these switches are maintained (to the detriment of other tasks since this is wasteful in both time and effort), and not dropped when they are unnecessary (same as many of the MPC switches) that continues this situation.  Yes that is poor management, but the thing is, you people are complaining about Frank doing the right thing here by removing these switches -- a sign of good management. 

Andree still has some say and I believe maybe 1 or 2 other people do also.. not entirely certain on that, but MPC isn't solely Frank's project.

How do you define "core functionality" switches?  Is something like the adjustability of the adaptive noise shaping a "core switch"?  How about the ath curve or something else?  If so then so are the ath adaptive switches, the noise shaping switches, the psymodel switches, etc in LAME.  There's really no difference here, the switches in LAME also affect the "core", if you define that as the encoding engine itself.

Granted, they are less numerous now, but things get out of hand over time.  The way to prevent this is to keep things in check and relevant as you go.  I think this is what Frank wants to do by removing some of the switches, and I think it is a good thing.

I do agree on this.  I think the best suggestion so far as been to adopt a system similar to the vorbis -q scale.

Exists since 1.05a. I removed a lot of "do this exactly in this mode" which changes
bitrate again and also changes the meaning of the profiles.
Quality setting selects bitrate between typical 30 kbps and 300 kbps.

--minSMR was removed, it diturbs the current model.

Basic (quality) settings are:

CODE

 --quality    x.x         (0.0...10.0, dflt: 5.0)

 --maxbitrate xxx         (56...2048, dflt: 2048)

 --maxlatency x.xx        (0.05...oo, dflt: oo)

 --minbitrate xx          (0...128, dflt: 0)      (may be!)

Secondary (quality) settings are:

CODE

 --stereoquality   x      (0...10, dflt: 5, stereo imaging quality against sound quality)

 --bandwithquality x      (0...10, dflt: 5, more encoded audio bandwith against encoding noise)

 --temporalquality x      (0...10, dflt: 5, more temporal resolution against better TMN)

Tertiary (quality) settings are:

CODE

 --psychodatabase x.psy   (load alternative psycho database, ca. 2 MByte large)

[Jan S.]

cool!!!

now all the insano tweakers can tweak the hell out of it and it will still be good....if the quality is saved in the file that is

[ancl]

Great!!

[atherean]

Great news, a quality scale, thanks, Frank
Now the only question i have is who's the alternative psycho?

[Jan S.]

who's the alternative psycho

Frank Klemm, Andree being the original psycho?

[AgentMil]

LoL!!

Well Jan S. those two are the best "psychos" I have ever seen, they created one of the best audio coding format ever!

On a more serious note:

When is 1.05a coming out?

Love your work Frank keep it up and you to Dibrom!!

Cheers
AgentMil

PS. Guess no need for that command line anymore.

[Jan S.]

yeah, I agree.
I love mpc... and even more now.

[Frank Klemm]

Originally posted by AgentMil
LoL!!

Well Jan S. those two are the best "psychos" I have ever seen, they created one of the best audio coding format ever!

On a more serious note:

When is 1.05a coming out?

Love your work Frank keep it up and you to Dibrom!!

Cheers
AgentMil

PS. Guess no need for that command line anymore.

Current version is 1.05e.

[AgentMil]

Thanks Frank

But when will that version be available to the public?

Cheers
AgentMil

[CiTay]

I'm very relieved. This was a big step forward.

[gdougherty]

Originally posted by Ruse


You say he correctly identified 13 different sections: how many sections were there in the splice? Were the encoder quality presets randomised, or did you start with the wav and work systematically through the quality presets --standard, --xtreme, --insane?

Did he identify the full length, unspliced encoded tracks as being inferior to the wave?

Go back and carefully read my posts, you're misunderstanding what went on. What I described is all that happened. The song is just over 6.5 minutes, giving 13 splices each from a randomly selected bitrate. I played only the original and the version made up of splices.

G

[YinYang]

Whee..

And I've alreday found my new commandline


--quality (pi-0.004i+0.005*earthsdiameter/2lightyears+AgeofNataliePortman/AgeofCameronDiaz)

[-=Ducky=-]

Frank, you and all the other people who work on mpc keep amazing me!!!!

Everyone was discussing what to do or not to do on the forum, but your switches sound the most logical and easy to use.

To put it in other words : great work, keep it up!!!

But some questions : what will happen to the names like standard, xtreme and insane in the current mpcfiles????

I really would like to convert all my mpc's SV7 to SV8 when it becomes available, I hope the backwards compatibility is still there to "lossless" convert to SV8.
I guess it is, but still wanted a confirmation from you.

[Gecko]

I have recently voted to keep the switches like they are. I was acting on impulse thinking: who the hell do you think you are, to tell me how to encode my music?! I kept thinking and would have rather voted (if it were available): wait and see. We all didn't know what was planned and were arguing based on pure speculation about the future switches. My main concern was that I would be limited to a very small amount of qualities/bitrates. Now that I see what is coming I would like to say: congrats, Frank, throw away those old switches!

I have difficulty understanding the secondary quality switches though. It sounds like you are making a tradeoff each time, sacrificing quality on one end while gaining quality on the other. Is this the way it's supposed to work? Couldn't you have switches like "tonality precision" or "impulse precision" which do not negatively affect other aspects of the encoding process? Please shed some light on this.

I also see a slight difficulty with the quality values, which I guess cannot be avoided. People will be asking: what do the q settings represent? What value do I need to use, if I want xyz enabled/disabled? etc.

[Frank Klemm]

Originally posted by Jan S.
cool!!!

now all the insano tweakers can tweak the hell out of it and it will still be good....if the quality is saved in the file that is

You haven't the slightest idea about the complexity of these tables.

[Jan S.]

You haven't the slightest idea about the complexity of these tables.

nope.

[gdougherty]

To attempt an answer for some users' questions, here's my take on the switches.
Anybody else have better guesses? Anybody in the know, care to clarify where I'm wrong?

Originally posted by Frank Klemm  
Exists since 1.05a. I removed a lot of "do this exactly in this mode" which changes
bitrate again and also changes the meaning of the profiles.
Quality setting selects bitrate between typical 30 kbps and 300 kbps.

--minSMR was removed, it disturbs the current model.

Basic (quality) settings are:

CODE

 --quality    x.x         (0.0...10.0, dflt: 5.0)

 --maxbitrate xxx         (56...2048, dflt: 2048)

 --maxlatency x.xx        (0.05...oo, dflt: oo)

 --minbitrate xx          (0...128, dflt: 0)      (may be!)

So generally we'll use these to adjust our files based on our goals. Most people would simply use the --quality switch, and not bother with the rest unless their specific application needed it. This would be 1.05 at its simplest. Basically replacing the non-tweaked presets with quality levels. Looks like CBR is possible from 56-128Kbps.

Originally posted by Frank Klemm  
Secondary (quality) settings are:

CODE

 --stereoquality   x      (0...10, dflt: 5, stereo imaging quality against sound quality)

 --bandwithquality x      (0...10, dflt: 5, more encoded audio bandwith against encoding noise)

 --temporalquality x      (0...10, dflt: 5, more temporal resolution against better TMN)

More tweaking options, I'd imagine that the quality scale also moves these at the same intervals since 5 seems to be the default for everything. Again, most people will leave these alone and probably use the quality scale.

Originally posted by Frank Klemm  
Tertiary (quality) settings are:

CODE

 --psychodatabase x.psy   (load alternative psycho database, ca. 2 MByte large)

And the final and most likely least frequently modified setting, switching to a differenct psycho accoustic model perhaps? I guess this means we can expect different models to come out with SV8? Perhaps different models focused towards low and high bitrate applications?

[Dibrom]

Originally posted by CiTay
I'm very relieved. This was a big step forward.

I agree.

Good job Frank!