Skip to main content

Topic: 'Waiting' is MPC killer clip (Read 12903 times) previous topic - next topic

0 Members and 1 Guest are viewing this topic.
  • Garf
  • [*][*][*][*][*]
  • Developer (Donating)
'Waiting' is MPC killer clip
Waiting clip in FLAC format

At the moment he sings 'for any_th_ing at all', the 'th' is distorted. This is rather easily audible.

mppenc 1.15r

I tested standard, easily ABX-able. Xtreme is better, but I can still ABX it.

  • ErikS
  • [*][*][*][*][*]
'Waiting' is MPC killer clip
Reply #1
Quote
Waiting clip in FLAC format

At the moment he sings 'for any_th_ing at all', the 'th' is distorted. This is rather easily audible.

mppenc 1.15r

I tested standard, easily ABX-able. Xtreme is better, but I can still ABX it.

Hehe.. Your too sensitive! 

Ok it's audible but is it really horrible enough to call it a killer clip?
  • Last Edit: 09 June, 2003, 05:48:24 AM by ErikS

  • sld
  • [*][*][*][*][*]
'Waiting' is MPC killer clip
Reply #2
Well, it's another step for improvement on MPC. Good job Garf. Too bad mpc SV8 at this stage is ready only when it's ready.

  • honz318712
  • [*][*][*]
'Waiting' is MPC killer clip
Reply #3
Wow, this one is tough for me to ABX.  I have been listening to it for awhile now straight, I’m thinking the distortion is more noticeable in the right channel for some reason.  Perhaps there is a channel separation problem during encoding?  The guitar is louder on the left channel, but I notice the distortion on the right channel mostly.  Let me know Garf is you notice this as well..

I’ve only tested mppenc 1.15r, I plan I trying 1.14 later.  I think this sample proves how good Garf’s hearing is.
  • Last Edit: 09 June, 2003, 06:25:43 AM by honz318712
Learn how to setup EAC and Musepack here Download Foobar 2000 here EAC Naming Scheme %N - %T - %C - %A
--quality 5 --xlevel --ape2 --artist "%a" --title "%t" --album "%g" --releasedate "%y" --track "%n" --genre "%m" --comment "rip by HoNz / EAC + mpcenc 1.15r q5" %s %d

  • Garf
  • [*][*][*][*][*]
  • Developer (Donating)
'Waiting' is MPC killer clip
Reply #4
Quote
Ok it's audible but is it really horrible enough to call it a killer clip?

Anything I can ABX at standard is a killer clip in my book. MPC was designed not to be ABX-able at this setting.

  • Garf
  • [*][*][*][*][*]
  • Developer (Donating)
'Waiting' is MPC killer clip
Reply #5
Quote
Wow, this one is tough for me to ABX.  I have been listening to it for awhile now straight, I’m thinking the distortion is more noticeable in the right channel for some reason.  Perhaps there is a channel separation problem during encoding?  The guitar is louder on the left channel, but I notice the distortion on the right channel mostly.  Let me know Garf is you notice this as well..

Yes. But to me its as if there's a 'fllt' that 'shoots' from the right into the center and left channel. It's also there in the original, but in the mpc I hear it louder in the left channel. Or something like that

  • guruboolez
  • [*][*][*][*][*]
  • Members (Donating)
'Waiting' is MPC killer clip
Reply #6
Quote
Anything I can ABX at standard is a killer clip in my book. MPC was designed not to be ABX-able at this setting.

That's why a lot of people chose to encode at extreme or insane : not really for transcoding or extreme samples, but for some hypothetical 'flaw' in the perfection of the mpc encoder.

I found myself some passages I can ABX (less easily that Waiting) with some perseverance at standard, and not at higher setting (sometimes extreme solve it, sometime insane...). I never worried about them : --standard is standard, very high, though, but definitely standard. If you want more security, use mpc at 200-230 kbps ; a lot of people are doing it, and a lot of people are blaming them for it...

Anyway, thank you for pointing this flaw of the mpc encoder.
  • Last Edit: 09 June, 2003, 07:52:49 AM by guruboolez

  • Garf
  • [*][*][*][*][*]
  • Developer (Donating)
'Waiting' is MPC killer clip
Reply #7
Quote
That's why a lot of people chose to encode at extreme or insane : not really for transcoding or extreme samples, but for some hypothetical 'flaw' in the perfection of the mpc encoder.

I found myself some passages I can ABX (less easily that Waiting) with some perseverance at standard, and not at higher setting (sometimes extreme solve it, sometime insane...). I never worried about them : --standard is standard, very high, though, but definitely standard. If you want more security, use mpc at 200-230 kbps ; a lot of people are doing it, and a lot of people are blaming them for it...


<begin pet peeve>
I'm one of those people

Almost everytime MPC fails at standard, the higher settings do so as well to some extent. What's the use of a higher setting then to feel 'safer'?

Sometimes --xtreme will save your *ss, sometimes --insane does so. Sometimes nothing does. And in the meantime, you're using bits and bits and bits...

It's like WavPack @ 400kbps. Sure, there's not much stuff you can ABX it on, but it's 2.5 times bigger than an MPC standard encode, and it's going to have failure cases as well.

If you want security, go lossless.
</end pet peeve>
  • Last Edit: 09 June, 2003, 08:58:25 AM by Garf

  • Pio2001
  • [*][*][*][*][*]
  • Global Moderator
'Waiting' is MPC killer clip
Reply #8
There is a blip in the right channel, this is what is distorded, isn't it ?

How about mppenc1.95z67 "post SV7 pre SV8" ? It seems to sound better.
  • Last Edit: 09 June, 2003, 08:49:07 AM by Pio2001

  • Garf
  • [*][*][*][*][*]
  • Developer (Donating)
'Waiting' is MPC killer clip
Reply #9
Quote
There is a blip in the right channel, this is what is distorded, isn't it ?


Yes.

Quote
How about mppenc1.95z67 "post SV7 pre SV8" ? It seems to sound better.


Not really. Again very easy ABX.
  • Last Edit: 09 June, 2003, 08:56:48 AM by Garf

  • NoMoDo
  • [*]
'Waiting' is MPC killer clip
Reply #10
Same here - very easily ABXable. I agree that the it is far more easy to ABX by listening only to the right channel than listening only to the left channel. Not sure if this indicates anything about the mpc encoder's failure, or the way the album was produced.

  • Pio2001
  • [*][*][*][*][*]
  • Global Moderator
'Waiting' is MPC killer clip
Reply #11
It seems the difficult sound is just in the right channel. Thus the distortion too.

  • Gecko
  • [*][*][*][*][*]
'Waiting' is MPC killer clip
Reply #12
Yikes! I tried to abx a different position: right at the start where he sings "I('ve) been". It sounds more noisy. ABXed without actually trying hard 8/8. I seem to be particularly sensitive to this kind of artifact: distorted pure voices. I can hear it in a lot of OGGs. Maybe it's the resampling of the soundcard? Well, that Terratec Aureon will soon be mine, so no more worries about resampling.

Focusing on the "th" of "anything" was harder for me to abx: 10/12. Sounds like he's spitting more. If in that part I focus on the roughness of the voice, it's easy again.

Quick test with the new foobar ABX component and resampling/replaygain enabled was still abxable, so it isn't my soundcard's resampling.

  • den
  • [*][*][*][*]
'Waiting' is MPC killer clip
Reply #13
Thanks for the sample Garf. I haven't had a chance to check it out, but it's always good to find out about these.

Quote
Sometimes --xtreme will save your *ss, sometimes --insane does so. Sometimes nothing does. And in the meantime, you're using bits and bits and bits...

It's like WavPack @ 400kbps. Sure, there's not much stuff you can ABX it on, but it's 2.5 times bigger than an MPC standard encode, and it's going to have failure cases as well.


Offtopic:

I regularly get surprised about the big deal made about wasted bits in this forum. 

Who cares about the bits and bits... really... It's not like you can take these "wasted" bits and save a few trees, significantly reduce greenhouse emissions or feed some starving children in a third world country. If an enthusiast is prepared to use the space for piece of mind, it's their choice. 

As for Wavpack, I could list reasons why I use it, that have nothing to do with ABXing, and give multiple examples where it outperforms Vorbis, LAME and Musepack consistently, in my everyday usage, even when they are all used at a 320 kbit rate, but there is no need. This is an MPC topic, and it's been discussed before.  B)

Back to topic:

Seriously though, thanks for the sample, I'm going to have a play with it a little later. 

Den.

  • Dibrom
  • [*][*][*][*][*]
  • Administrator
'Waiting' is MPC killer clip
Reply #14
Quote
Who cares about the bits and bits... really... It's not like you can take these "wasted" bits and save a few trees, significantly reduce greenhouse emissions or feed some starving children in a third world country. If an enthusiast is prepared to use the space for piece of mind, it's their choice. 

The problem is not the fact that people have a choice, it's the fact that people often protray these ultra high bitrates as being necessary, which is misleading.  This forum was started to provide relevant and valid information, but when people start going around saying that you need to encode your MPC's @300kbps or so it goes against this.  In fact, it's not all that different from the people that say you shouldn't use joint stereo or that you shouldn't use vbr, "just to be safe."

I personally don't care what someone uses, but I do care with what they tell others to use when they do not provide a strong case supporting their stance.  Being "safe" isn't one.

  • Garf
  • [*][*][*][*][*]
  • Developer (Donating)
'Waiting' is MPC killer clip
Reply #15
Quote
As for Wavpack, I could list reasons why I use it, that have nothing to do with ABXing, and give multiple examples where it outperforms Vorbis, LAME and Musepack consistently, in my everyday usage, even when they are all used at a 320 kbit rate, but there is no need.

I'd actually like to see and hear the examples (clips =/= claims). I've never actually seen them. If they were in another thread, please point me in the right direction, thanks.

There is relevance to this thread because I was actually comparing WavPack and MPC when I ABXed this sample. I could ABX WavPack @ 320 kbps on a reasonable amount of stuff (but not at 400kbps), and I couldn't ABX MPC --standard on most, with some exceptions, like this one.

For me, this rules out WavPack for normal usage. I've seen several claims it's superior in transcoding, but I could not reproduce them so far, and as far as I know no concrete examples were ever given.
  • Last Edit: 10 June, 2003, 07:28:08 AM by Garf

  • DickD
  • [*][*][*][*]
'Waiting' is MPC killer clip
Reply #16
Garf: Thanks for mentioning how you came across this sample. I'd wondered how the more subtle artifacts were discovered.

Often without the original or knowing the original sound in serious detail with a very good sound memory for it, it is hard to notice artifacts when they don't sound really ugly (though I guess this sample's extra noise at 'th', being stereo-specific, does sound out of place (at least on headphones) and even in a reasonably unsilent environment I found it fairly easy to ABX from a mppenc 1.14 encode even if it wouldn't sound especially disturbing. I just wonder if I'd have ever spotted it in the first place in normal listening. I suspect not, though I'm a beginner.

I'm also fascinated by Wavpack lossy/hybrid and would love to hear some of the samples where it fails at ordinary volumes and a valid comparison with Musepack and lossless for transcoding to format like MP3 or MiniDisc ATRAC (short play).
It would also be good to see how it compares to Musepack at comparable average bitrates.

Apart from soundcard hiss, I haven't noticed major problems going from MPC standard --xlevel to my MiniDisc Walkman (rather old and only supporting short play) via analogue line in, though I haven't made conscientious listening tests.

It appears, from recent threads in the Lossless Codecs section, that David (a.k.a. bryant) is keen to implement something like strong ATH noise shaping of the residual noise in Wavpack 4's lossy hybrid mode, which ought to help improve the performance for any given bitrate (whereas the current triangular noise shaping reputedly makes things worse than leaving noise shaping off).

Sorry for the length of the next bit of the post. Skip if not interested.

I'm not sure if it's simply the noise that you're ABXing, Garf (possible since you say it's usually OK at 400 kbps) or effects of the "inadvertant psychoacoustics" used by Wavpack lossy. (e.g. the allowed noise increases after transients or anything else that causes the predictor's error to increase suddenly and the increase in allowed noise has about a 6 ms decay time. While this appears to be an inadvertant side-effect of the bitrate control, it appears to be inadvertantly allowing greater distortion during the post-masking period of transients, which agrees with psychoacoustic principles, at least qualitatively.

However, if something that is not like a normal loud transient also causes the average prediction error to increase markedly (prediction is based on the previous 16 lossy sample values at most, but the rolling average of the error is over about 6 ms), it might cause the allowed distortion/noise to become much larger for at least a few milliseconds and exceed the psychoacoustic masking thresholds that a psychoacoustically-aware lossy codec like MPC would recognise.

I'm thinking that prediction error might increase without a source of temporal masking being present if, for example, a new frequency appears or there's a rapid phase shift, maybe a fast tone sweep or rapid vibrato spring to mind as plausible causes, or perhaps a change between tonal and atonal/noiselike signals of similar loudness would cause the prediction error and then the future allowed lossy distortion to increase during a time when there's insufficient masking. Dynamic compression or peak clipping/limiting might also cause prediction error to increase. Crazy levels of inaudible ultrasound, like the udial sample might also cause dramatic prediction errors far bigger than the audible portion of the signal, though they hardly count as normal music!

If it's just a brief increase in time-averaged prediction error that subsides, I guess it could sound like a burst, swell or pop of noise above the normally more consistent (and inaudible at normal volumes) noise floor. If the perceived noise level is reduced by good noise shaping, this may become solved at a lower bitrate. I haven't played with Wavpack enough to try to identify samples of this sort, but perhaps I have some idea of what sounds might cause audible problems. I guess starting with a lower bitrate than 320 kbps is also good for getting used to what sorts of scenarios tend to cause artifacts.

I'd have thought that Musepack ought to be perceived to be better than a lossy hybrid at the same average bitrate, with the exception of when the predictor becomes remarkably accurate (and then the lossy hybrid needs becomes practically lossless because the sample is highly predictable). I presume that the sub-band coding of Musepack doesn't use a predictor (and can't do so very usefully), so it just quantizes each subband as coarsely as allowed by masking and ATH, then packs the quantized data using Huffman coding, Rice coding or similar.

Exceptions might be if a psychoacoustic codec cut out important frequencies when they're not masked for all people (not a problem with Musepack at high quality levels) or when the psycho codec introduced transient smear or something similar. I'm not sure this happens in MPC, because of the sub-band splitting and reconstruction process being lossless, I believe.
The time-to-frequency-domain transform and windowing only applies to the analysis phase, where pre-transient distortion might exceed the pre-masking threshold in the time before a transient if time resolution weren't high enough. MPC is pretty darned excellent in this respect, but guruboolez seems to notice differences for harpsichord and baroque, so perhaps here Wavpack lossy could win, but then again, would MPC at an average 320 kbps solve it and sound better than WV at 320 kbps on the same track?
I guess also that relying on tonality measurements for masking thresholds might be considered a risk.

But once you've thrown about 160-190 kbps at all the right areas (averaged over a whole track), the 130-160kbps you throw at making quantization distortion far below the predicted masking threshold to hit an average 320 kbps and match Wavpack ought to cover up what's left rather better than the 320 kbps thrown at quantizing the error-signal of the Wavpack predictor in a relatively untargetted manner. In a like for like comparison with the same total file size for a track, I'd expect Musepack to win from a sort of information theory standpoint.

Wavpack lossy might have a more consistent bitrate (for unbuffered streaming?) and might beat a Musepack that's constrained not to use instantaneous bitrates >320 kbps. E.g. Fatboy Slim's Kalifornia intro, known as fatboy.wav is likely to become non-transparent (it uses about 500 kbps in MPC standard according to Frank), while it's just plausible that Wavpack's 16-sample predictor would cope OK at 320 kbps by not being overinfluenced by prior samples.

That's just gut instinct. It's rather harder to verify.

  • den
  • [*][*][*][*]
'Waiting' is MPC killer clip
Reply #17
Garf:
I haven't had a chance to try the sample yet, Real Life™ and my two little daughters keep getting in the way... 

Quote
I'd actually like to see and hear the examples (clips =/= claims). I've never actually seen them. If they were in another thread, please point me in the right direction, thanks.

There is relevance to this thread because I was actually comparing WavPack and MPC when I ABXed this sample. I could ABX WavPack @ 320 kbps on a reasonable amount of stuff (but not at 400kbps), and I couldn't ABX MPC --standard on most, with some exceptions, like this one.

For me, this rules out WavPack for normal usage. I've seen several claims it's superior in transcoding, but I could not reproduce them so far, and as far as I know no concrete examples were ever given.


I posted some stuff about my tests and findings here.. This is when I first tried Wavpack just to see what our feathered friend was talking about.  I'm not at my home PC at the moment where the samples reside, but if required, I can make up a sample or two and post them later. A classic test sample is the beginning of New Order - Blue Monday (original mix). The repeated drum machine intro doesn't transcode well in most cases. Different faults introduced depending on the codec.

For those who have read the above, they may wish to skip the rest of this post.

The majority of the time, I listen to music on Minidisc as I travel a lot with my job. It has proven itself to be a very convenient and reasonable portable format for me, but the downside is that everything must be encoded into ATRAC. I also have other audio needs, so I do not wish to keep all of my music archived in ATRAC on my main PC, as I can't use it for anything else. I also can't go lossless as this stage as I simply do not have the space available, and won't for a while. I had been encoding my CDs using Lame alt presets (mainly standard), and then dabbled in Vorbis q5/6 (including your cool tuned versions) and finally settled on Musepack q5/6 as I was starting to detect the odd transcode artifact in my music from Lame and Vorbis. Musepack was not without artifacts either, but much less frequent or annoying than the others.

Musepack, Vorbis and LAME are all transparent to me with the above settings, most of the time, when used to directly play back music. They only present problems when transcoded into ATRAC.

When I tried Wavpack on some of my music, and some classic problem samples, I quickly found that I can ABX it every time at 265, 320 and sometimes 400 kbits if I get into "the zone" while testing. I am not hearing artifacts/music defects as such, just the slight hint of additional background noise (hiss) within the spaces in the music when listening at high volume. I seem to be sensitive to this. It is not annoying and I can easily ignore it, but it's there.

I then took some samples of music, encoded them into Wavpack 320 bits, LAME standard and extreme, Musepack q5 and q10 and Vorbis q5 and q10. I then transcoded all of these into ATRAC3 132 kbits and onto Minidisc. Wavpack had a slightly higher background noise level in quiet sections at high volume, but no other problems. LAME and Vorbis had regular noticeable artifacts, including pre-echo/smearing. Musepack had some slight changes on some repeated percussive transients but they were much harder to pick up. The original wav was like Wavpack without the noise.

I found that I can ABX the above against wav direct into ATRAC3 if I have my wife select tracks at random on my MD behind my back, except the Wavpack where I sometimes get it wrong.

I repeat the above test from time to time for my own curiousity, but the findings have been consistent for LAME and Vorbis, but sometimes Musepack is transparent as a transcode source. I'm yet to test AAC.

As a result, I now keep all my music on my main PC as Wavpack lossy at 320 kbits with the joint stereo and high quality switches, and trasncode these to other formats (mainly ATRAC) as I require, with very pleasing results so far. For irreplaceable recordings, I also keep the correction file to make it lossless. B)

I'm truly looking forward to Wavpack 4 also, as it sounds like David has some cool, but relatively easily tricks to implement that will improve the bitrate, such as smart switching joint stereo and how the additional noise is distributed. (It's easy for me to say that when I'm not he one doing the coding.) 

Sorry for the long post, but hopefully it makes sense, and tells you enough to check it out further if you wish.

Den.

  • Garf
  • [*][*][*][*][*]
  • Developer (Donating)
'Waiting' is MPC killer clip
Reply #18
Quote
but if required, I can make up a sample or two and post them later.

Yes, please.

  • Garf
  • [*][*][*][*][*]
  • Developer (Donating)
'Waiting' is MPC killer clip
Reply #19
Quote
I just wonder if I'd have ever spotted it in the first place in normal listening. I suspect not, though I'm a beginner.




The spreading through the channels sounds unnatural enough to pick it up without much problems. But you have to know the track, of course.

Quote
I'm not sure if it's simply the noise that you're ABXing, Garf (possible since you say it's usually OK at 400 kbps)


Yes. In some sections, there's a noticeable noise increase. At 400kbps it usually 'sinks in' the music. But at 320k it's sometimes obvious enough that WavPack is 'out' for me.
  • Last Edit: 11 June, 2003, 03:48:58 AM by Garf

  • den
  • [*][*][*][*]
'Waiting' is MPC killer clip
Reply #20
It sounds like you are hearing the same noise as myself Garf. It isn't too obvious in everyday listening, but if you know the music, it stands out.

As for a sample, I will get something posted somewhere either tonight, but more likely tomorrow night when I have some scheduled home PC time... 

How do you want the sample? I can make part of the wav available as a flac or even Wavpack lossless    but are you set up to then make this into ATRAC and compare against the other transcode options? Or are you more interested in how it transcodes into another lossy codec sych as Vorbis? I'm just interested to see how others do these types of tests.

All of my testing was into ATRAC, and auditioned off my Minidisc. It's hard to get this into something you can ABX with software such as WinABX. I know Guruboolez has been known to download the Sony software and then record the ATRAC output using Sound Recorder or similar, but this is not without its problems...

Den.

  • den
  • [*][*][*][*]
'Waiting' is MPC killer clip
Reply #21
OK, sorry for the delay.

First sample is the infamous Roland drum machine intro for Blue Monday - New Order. I get artifacts in this section when I transcode from high bitrate encodes of this into ATRAC3 132 kbit for LAME, Vorbis and Musepack, but not via Wavpack lossy.

Blue Monday Intro as Wavpack Lossless

Later in the same piece (New Order - Blue Monday from their Substance compilation double CD), is the following "jet" sequence. Also has the repeating drum machine, but adds some hihat style percussion.

Blue Monday Jet Sequence as Wavpack Lossless

Throughout Blue Monday, you can find examples of the various evils of transcoding, but some parts and codecs are worse than others, and each lossy codec puts it's own spin on it. 

I have many other samples I can show, but these are a start. I listen to lots of Australian rock music, which tends to feature a distinctive style of rock percussion, which can show weaknesses in various codecs, particularly if you need to transcode... I only have limited web space so check these out first, and then I can place some more up. 

Examples include various tracks from INXS - The Swing, including the title track itself and Original Sin.
Midnight Oil - The Power and the Passion (drum solo) off the album 10,9,8,...,2,1.
The Models - I hear the motion. (Has very basic drum machine line initially, but it also has naturally played hihat through out which often gets lost/smeared)  B) 

I look forward to hearing any comments regarding these samples and anyones findings with regards to how various codecs suit/don't suit transcoding.

Den.

  • KikeG
  • [*][*][*][*][*]
  • Developer
'Waiting' is MPC killer clip
Reply #22
Doesn't 132 Kbps ATRAC3, trash quality on difficult samples, by itself?

  • den
  • [*][*][*][*]
'Waiting' is MPC killer clip
Reply #23
Off topic, but here goes...

Quote
Posted on Jun 12 2003 - 06:33 AM
--------------------------------------------------------------------------------
Doesn't 132 Kbps ATRAC3, trash quality on difficult samples, by itself?


Yup, but it is as high as you can go with Minidisc if you wish to use the direct transfer (NetMD) at 32x from your PC. The software actually has a SP mode which is 292 kbits (normal Minidisc bitrate), but if you dive into the software, it actually rips and encodes at 132 kbit, archives this on your HD for transfers, and then transcodes it to SP (292 kbit) when you select it to transfer. 

Sony's design reasoning for this has continued to be one of the most debated and slammed topics on various Minidisc boards. Perhaps they thought they were saving disk space? 

Anyway, I personally find ATRAC3 (LP2) at 132 kbit a reasonable compromise for portable music. It allows the high speed transfers, and provides 2 hours 40 minutes per disk. In real world portable style listening it sounds very good on the whole through street style head phones, and to my ears is superior to mp3 and vorbis at 128 kbits, but I'm not so sure about AAC. It consistently has less pre-echo than LAME and Vorbis with the stuff I listen to. I would liken it to somewhere around LAME @ 160 kbits with my music.

ATRAC3 LP4 at 66 kbit is another story. :x Don't go there...

Some time in the future, I plan to invest in a sound card (probably USB) with digital out that I can use from my home PC or laptop, and record in ATRAC-R (292 kbit) in real time ex foobar.  B) I tried some as an analogue recording, and it was very good if I ignored the slight hum from using crappy cables...

Den.

  • bubka
  • [*][*][*]
'Waiting' is MPC killer clip
Reply #24
thats funny, i had trouble with ogg vorbis on this album, i wonder what it is?
Chaintech AV-710