Skip to main content
Topic: Musepack setting for very high quality? (Read 2622 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Musepack setting for very high quality?

Hi,
This is my first post here, but I was a long time reader and finally decided to join.
I'm going to rip my cd collection and I want very high quality lossy archive that should be also good for occasional transcoding to other lossy (vorbis/aac) but I dont want lossles + lossy because maintaining two archives is too demanding for me.

As I understand using additional switches --nmt --tmn generally improving quality.

In few old threads I've read that using q7 --nmt 16 --tmn 32 (~320k) was a popular setting and gives very good quality even for trancoding which guruboolez also said in one thread long time ago.

I think that something like q9 --nmt 18 --tmn 36 --bw 18600 should be enough for really high quality. Probably very good for transcoding. This yields bitrate of 350k on average which is my target bitrate.

Do you think this could be also enough for problem samples and very high quality in general?

Also, is --bw switch good to use all the time? It is basically a lowpass filter as I learned. Is it degrading quality in some other way except cutting frequency above 18600 Hz?
Am I on the right track?
Sorry for long post. :)

Re: Musepack setting for very high quality?

Reply #1
and why not directly encode in vorbis/aac ? in >=320kbps no matter: musepack/vorbis/aac/opus. not a mp3/wma - good. there is also a lossyflac/wavpack-hybrid. but, if you still want a musepack - use q10 without any. imho.

Re: Musepack setting for very high quality?

Reply #2
Be careful when you're looking for very old information and settings in this forum. I'm not familiar anymore with MPC but I'm quite sure these switches are outdated and are coming from a time when MPC wasn't that flexible. Long long time ago --insane was the highest setting available and people tried some tricks to higher the bitrate or waste some free time. Now you have a much better VBR scale with higher settings (q8…10) to cover all usage without doing anything wrong.
MPC encoders are also well-tuned. Additional settings are not necessary and benefits are highly hypothetical… and probably not audible at all at such high bitrate.

As you mention my test (thank you!), it's also very old (15 years):
https://hydrogenaud.io/index.php?topic=32440.0
Ogg Vorbis was at this time at least as good as MPC for reencoding purpose (it ends with a better score). In the meantime Opus appeared ; Vorbis and AAC improved ; lossyWAV also create an opportunity for good lossy source.
I second m14u's answer. Go for a format you can directly use in many devices (at high bitrate they are all very similar and could be a better source than MPC) and you can eventualy reencode later if you really need to spare some space. Opus, Vorbis and AAC offer your targeted bitrate (350 kbps). You can also take into consideration WavPack Lossy or even LossyFLAC (I'm not familiar with it).

Re: Musepack setting for very high quality?

Reply #3
Thanks a lot for your answers.
But I forgot to mention that I found this thread few days ago:
https://hydrogenaud.io/index.php?topic=96108.25

On page 2-3 I noticed that magicgoose and shadowking discovered that standard setting with --nmt 18 became transparent (or very close to it) but magicgoose was able to ABX Q10 but he couldnt ABX q5 --nmt 18.
So, I think --nmt 18 is definitely improving quality.

As for WavPack, I think it needs higher than 350k for piece of mind and probably strong setting like hx4 which is very slow.

@guruboolez
Do you think if I use vorbis/opus @350k that I'll have a quality headroom even for problem samples?
Shoud I use -q (quality) or -b (bitrate) setting for vorbis? Is there any difference?

Re: Musepack setting for very high quality?

Reply #4
Sorry, I thought you were mentioning much older information.

Quote
On page 2-3 I noticed that magicgoose and shadowking discovered that standard setting with --nmt 18 became transparent (or very close to it) but magicgoose was able to ABX Q10 but he couldnt ABX q5 --nmt 18.
So, I think --nmt 18 is definitely improving quality.
It's improving quality on one sample or kind of samples and with one version of the encoder.
Some tuning are reducing some known artifacts but they may also lower quality elsewhere. I'm not saying --nmt 18 can lower quality, but from a logical point of view you should be careful.

To answer your questions:
I know very well this scary feeling of insecurity that come with the decision of replacing lossless by high bitrate lossy: is it enough? can I improve it? Should I increase the bitrate further to be very very sure that nothing could ever happen? What if a golden ears person appears and find a problem sample that pushes the frontier of known audibility?

I personally think that this mindset is a real poison. You can never be sure that your encoder and setting of choice are perfect. Especially if you rely on someone else's experience.
But you can be sure of something: even if someone comes with a problem sample at 350 kbps, the audible difference will probably be  very slight, not disturbing on a daily basis listening experience (ABX tests are extreme listening conditions, often focused on short/very short parts), and very rare.

At this point, there's a simple alternative from a logical point of view:
  • [1] as a probability still exists to discover an audible issue with X bitrate, you feel the necessity to increase the bitrate further, over and over…
  • [2] as you understand that 350 kbps is maybe the double of what you really need and is still not enough to warranty your peace of mind, you'll start to decrease the bitrate, over and over… to a certain point.

— Do you think if I use vorbis/opus @350k that I'll have a quality headroom even for problem samples?
I'm sure it has enough headroom to handle 99,99999% of the total duration of your music even if you're a critical listener with platinum ears. And the missing 0,00001? You can never be sure, whatever the format/setting/switch you chose.

— Shoud I use -q (quality) or -b (bitrate) setting for vorbis? Is there any difference?
-q is a general and common recommendation. However it hasn't been tested at such high bitrate because it's insane and humanly impossible. Maybe -b can handle transparently 99,99999% of your music and -q 99,99998%? Who knows? Nobody… So stick with the recommended settings (-q)


Re: Musepack setting for very high quality?

Reply #5
To answer your questions:
I know very well this scary feeling of insecurity that come with the decision of replacing lossless by high bitrate lossy: is it enough? can I improve it? Should I increase the bitrate further to be very very sure that nothing could ever happen? What if a golden ears person appears and find a problem sample that pushes the frontier of known audibility?

Yes, exactly. That's what I'm facing right now.
I'll think about your recommendations. Very helpful. Thanks.
I'll decide between MPC Q10 or Vorbis -q 9.2
Also would like to know more about WavPack @350-400k... (not higher)
Could be also very good alternative. As I understand It's very good for transcoding also.
Maybe I'll post a question in wavpack section... I'll think about that.
For now I think I'll go with Vorbis or mpc.
Thanks. :)

Re: Musepack setting for very high quality?

Reply #6
I've a small experience with WavPack Lossy (I submitted one or two problem samples in the past). The only known artifact is to my knowledge noise, very basic noise. Even if you hear it on a daily playback basis, it's not something that may alert you. It's a bit like grain/noise on a digital photo. If your compressor add a tiny amount of noise you can spot from time to time, you won't  be able to say if it comes from your camera's sensor or not. Same with audible noise.
Of course if you're very maniac, each time you'll hear noise there's a risk that you'll think it always comes from WavPack  and not from the recording process :D


But from my experience, the noise is very very difficult to hear. And even if you hear it, well, it's just a slight amount of hiss :)

Re: Musepack setting for very high quality?

Reply #7
Of course if you're very maniac, each time you'll hear noise there's a risk that you'll think it always comes from WavPack  and not from the recording process :D
I think this could be an issue. :D
What setting would you recommend for Wavpack lossy that is considered high quality?
is x4 necessary for high bitrate Wavpack? I would like to avoid slow encoding if possible.

Re: Musepack setting for very high quality?

Reply #8
"that extra processing can be done during encoding to provide better compression, but with NO corresponding cost to decoding performance"
use x6, don't be greedy.

Re: Musepack setting for very high quality?

Reply #9
What setting would you recommend for Wavpack lossy that is considered high quality?
is x4 necessary for high bitrate Wavpack? I would like to avoid slow encoding if possible.
As said before, my experience is really small (and old, outdated: the encoder improved with time).
Take a look at this thread:
https://hydrogenaud.io/index.php?topic=117610.0

About speed: use what's comfortable/possible for you.

Re: Musepack setting for very high quality?

Reply #10
As said before, my experience is really small (and old, outdated: the encoder improved with time).
Take a look at this thread:
https://hydrogenaud.io/index.php?topic=117610.0

About speed: use what's comfortable/possible for you.
I missed that one. Thanks.

@m14u
x6 is extremly slow. I doubt is any better than x4 and encoding with x6 is really slow. I think is not worth it.
But thanks for the info.

Re: Musepack setting for very high quality?

Reply #11
i need correcting:
x-key increase compression in lossless mode, olways; in lossy mode MAY be negotive effect. source depends.

Re: Musepack setting for very high quality?

Reply #12
What's stopping you from going lossless? Do the tedious process of ripping and tagging once and make sure you never have to do it again. It really suxx and is a major investment of your time (so also create a backup).

High bitrate lossy with huge headroom is a lost cause, imo. It's wasteful for the most part and in those instances where the psychacoustic model breaks, it may still not be enough. Tweaking parameters may give you that comforting feeling of "taking control", but the reality is that you are still using something inherently imperfect.

Go lossless and have peace of mind or embrace the occasional artifact with lossy encoding at roughly your transparency threshold.

Re: Musepack setting for very high quality?

Reply #13
Quote
It's improving quality on one sample or kind of samples and with one version of the encoder.
It definitely shouldn't hurt all other samples, as evident from documentation. (it just raises 1 specific parameter which should raise Signal/Noise ratio in a subset of situations) unless there are some undiscovered bugs related to that.
And there aren't really any other versions of the encoder, and probably won't be, it seems none is developing this codec anymore, let's be real.
some ANC'd headphones + AutoEq-based impulse + Meier Crossfeed (30%)

Re: Musepack setting for very high quality?

Reply #14
I'd also advise against WavPack lossy because it targets bitrate (constant, except it won't pad data with useless bits if it already can encode a piece losslessly with lesser bitrate), not quality, so it has to be wasteful AND you still may get artifacts when there's something too complex for the target bitrate. If you really want to use this kind of codec, I'd recommend LossyWAV instead. (which targets quality, not bitrate)
some ANC'd headphones + AutoEq-based impulse + Meier Crossfeed (30%)

Re: Musepack setting for very high quality?

Reply #15
Thank you for your answer, magicgoose. I haven't checked in MPC documentation what --nmt or --tmn are really doing; I knew it but I totally forgot it with time. I also agree that MPC won't probably rise again. My comment was much more a general speaking statement about lossy encodings and I haven't read the thread mentionned by synclagz (started in 2012 and ended in 2019).
And to be fully honest, I'm sure that using MPC with q9 --nmt 18 --tmn 36 --bw 18600 instead of pure q9 or q10 alone won't change anything to the listening experience. It's more an opened door to forgotten practice here (remember r3mix ;) )

Re: Musepack setting for very high quality?

Reply #16
(please delete this message, accidentally sent 2 times and cannot delete myself)
some ANC'd headphones + AutoEq-based impulse + Meier Crossfeed (30%)

Re: Musepack setting for very high quality?

Reply #17
for me, --nmt 18 resolves all problem samples that it had without this, so I don't see any value in further tweaking.
but I don't really have good ears, and they only get worse with age. (well, maybe they're still more sensitive than average, idk)
someone else might want to explore this perhaps.
some ANC'd headphones + AutoEq-based impulse + Meier Crossfeed (30%)

Re: Musepack setting for very high quality?

Reply #18
I'd also advise against WavPack lossy because it targets bitrate (constant, except it won't pad data with useless bits if it already can encode a piece losslessly with lesser bitrate), not quality, so it has to be wasteful AND you still may get artifacts when there's something too complex for the target bitrate. If you really want to use this kind of codec, I'd recommend LossyWAV instead. (which targets quality, not bitrate)

It's theoretically true but efficiency of both formats is probably not the same. At a given bitrate the quality-based format can be worse.

This is what Shadowking wrote two years ago:
It could be true. Wavpack at same bitrate is better than lossywav with objectively less noise .  Lossyway noise moves in 6db steps while wavpack moves infinite steps from what I read in the past. OTOH Wavpack doesn't have a yet a true quality mode while lossywav does. This also might not have any impact if comparing strong settings like lossyway --extreme to wavpack -b550x4 .. wavpack is also expected to be 100% transparent and offering better quality [objective].  At more moderate bitrates (400k)  subjectively , Lossywav may have the advantage in very rare cases due to the quality mode . At lower rates like 250k wavpack may have a general advantage too .

From this post: WavPack lossy should be better at low bitrate as a consequence of advance coding techniques; and with no surprise at very high bitrate both WavPack and LossyWAV sound perfect. And between: LossyWAV has a true quality mode but lower quality coding technique; WavPack lossy better coding tool but no true quality mode. I don't see a clear winner here (except perhaps WavPack lossy at  ~200…250 kbps) in the absence of listening tests.
You don't recommand WavPack because it's wasteful and with possible audible artifacts on possible situation. But is it really different from true VBR encoders like MPC --nmt 18, or from Ogg Vorbis 350, or anything lossy that always stays at very high bitrate?

Just for the anecdote, I fed my classical bitrate table with some extreme bitrate (low and high) recordings. And I have data for MPC and low bitrate stereo CD:
Code: [Select]
                                                                                     FLAC -8       MPC -7        MPC -10
Feldman [Schleiermacher] The Late Piano Works, Vol.2 (MDG, CD, 2009)                 219 kbps      235 kbps      337 kbps
Mompou [Perianes] Música Callada (Harmonia Mundi, CD, 2006)                          274 kbps      250 kbps      361 kbps
Silvestrov [Blumina] Piano Works (Grand Piano, CD, 2013)                             255 kbps      217 kbps      332 kbps
VA [Haochen Zhang] Schumann, Liszt, Janacek & Brahms. Piano Works (BIS, CD, 2017)    308 kbps      235 kbps      335 kbps
                                                                          AVERAGE    264 kbps      236 kbps      341 kbps

MPC is a true VBR encoder, quality-oriented and efficient… and it can wastes bit as well (MPC uses up to 50% bitrate more than FLAC, not LossyFLAC, true FLAC!) on some occasion. And even when bitrate exceeds the needed bitrate for lossless compression you still can't be sure that transparency is fully reached because every piece of the signal is transformed by the lossy algorithms. Anyway, these examples can also prove that MPC can be as wasteful as WavPack Lossy and probably as unsecure for maniac people (same apply to Vorbis and Opus, but AAC FDK, FHG and iTunes are more flexible here…).


To recap, problem is unchanged and Gecko perfectly summarized it. If we except MP3 which has native flaws at high bitrate (pre-echo handling), everything is considered as wasteful but transparent… at least until something breaks and wait to be discovered. For peace of mind lossless is therefore the best and maybe the only choice ; for high quality at 350 kbps almost all formats are fine. There is no need to tweak anything with additional switches or to recommend one format over another.

Re: Musepack setting for very high quality?

Reply #19
I see a lot of new stuff to learn since yesterday. :)
First of all I would like to thank you all for giving me deatailed answers so I can decide which way to go.
@Gecko
Yes, lossless is the best way to go but, as I said, I don't want to maintain lossless + lossy for mobile. It's kind of time consuming and tedious to rip/convert two archives and update+backup twice all the time.
What I wanted to see if there is any lossy that is of high quality to store my music as single archive making it more simple to manipulate.

@magicgoose
My setting for MPC came mostly from your experience with --nmt 18 switch and I wanted to bump the quality a bit higher so potential problem samples shouldn't cause issues (at least from my findings/perspective). I'm glad you participate here. :)

@guruboolez
Your quote of shadowking's notation regarding wavpack @550x4 is very interesting. He obviosly have much experience in wavpack (as I also noticed reading a lot of threads here).
If Wavpack setting at this very high or even extreme setting @550k could be considered objectively transparent, there is no need for lossless in my opinion and 550k is good 30-40% smaller than tipical pop/rock flac's which are in 900-1100k range usually.
I doubt that there will be "killer" samples that will sound bad @550k. If this setting is realiable enough for normal music (I don't care much about artificial sine waves and stuff) this is also very good solution to archive my music.
A lot to think about... :)

Re: Musepack setting for very high quality?

Reply #20
@magicgoose
Do you think that --nmt 18 switch can be used all the time?
Did you notice any degradation in quality on other than problem samples?

Re: Musepack setting for very high quality?

Reply #21
No, I didn't notice any degradation from it, and by my (limited) understanding of the underlying algorithms, there shouldn't be any degradation. (however it inflates bitrate to ~266 kbps on average (on a large collection of music, mostly rock and metal); still less than what LossyFLAC or WavPack would need to achieve transparency on all samples but I suppose if it was something smarter, it could be much less than that)

Unrelated note, what I like about MPC is that it supports 44.1k and 48k sample rates as-is, and so I can just encode tracks individually and things will remain gapless and will start and end exactly where they did. (kinda useful if occasionally playing it in random order)
If Opus could do this, I'd probably just use it and forgive it the fact that there's a risk of audible distortion on some samples, but Opus complications with gapless requirements are a bit too annoying for me for now.
some ANC'd headphones + AutoEq-based impulse + Meier Crossfeed (30%)

Re: Musepack setting for very high quality?

Reply #22
Thanks a lot.
As I understand, nmt and tmn swithes  should incease sensitivity to tone/noise (signal to noise ratio). Maybe not technically correct, but I think that end result is inceased sensitivity.
So, my initial setting q9 --nmt 18 --tmn 36 should only be better than standard --nmt 18 (cannot be worse)?
What do you think?


Re: Musepack setting for very high quality?

Reply #24
Hello synclagz and everyone.

For the last few years I've been experimenting with a different direction.
The motivation was similar; to maintain one collection and avoid lossy only
dillema. Also to maximise portable storage.

I've been using WV 265k -x4c and -hx4c and the results are very good.
On the PC the WV files are decoded as losssless. I have Android phones
since 2012 and even the 2011 model easily handled all the WV lossy modes.
On the phones, I copy the WV files only and not the WVC correction files.
I am pleased with lossy quality , never had an issue so far. As there's only theoretical 'noise',
to me -personally- its lossless or close to it.
If you need transcode to mp3 etc; you do it on the PC and it would work identically
to flac  / ape transcoding -after all the source is decoded lossless

So there's one way of having one lossless  / lossy archive .

Regarding a lossy only  'quality 1st approach'  while not using insane data rate , I would go for
WV 350k -hx4.

Regarding MPC --nmt 18.  It remains an unknown . I know Shy from musepack.net advises against these tweaks.
F Klemm was also discouraging these. So quality scale was made
In another instance, the --ms 15 switch which was touted by some users actually made one problem sample
FSOL much worse for me using --standard --ms 15.  The problem spot devloped a new 'chirp'. It was assumed nothing would
go wrong since -- ms 15 only inflated bitrate. But F Klemm was against that too & suggested --quality 6 or 7 for DOLBY or
vocal cuts.
It could be that tweaking --standard can give worst of everything in certain scenario - Inflated bitrate but still
'standard'.
wavpack 4.8 -b3hx4cl

 
SimplePortal 1.0.0 RC1 © 2008-2020