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: Source for the perceived quality vs file size chart in the LAME wiki? (Read 4987 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Source for the perceived quality vs file size chart in the LAME wiki?

Hi,
Just joined a couple of days ago.  I've enough background in science and engineering to be completely on board with the general tenor of things on HA that if you're using a lossy format anyway, then why not on settings that are effectively 'as good' or 'good enough' and save disk space etc. 

Until about a month ago I had a jumble of bit rates because of the history of the collection - when it was ripped, under what circumstances (mostly in a hurry), and because I hadn't really given it much thought.  So there's everything from 5 years ago 128kbps CBR to (more recently) 320kbps stuff on there.  It's only lately ripping vinyl, and particularly with the appearance of the Beatles remasters and playing MP3s over living room stereo speakers that I've started to notice, then research (hence HA) and finally try to decide what the 'policy' will be from here on in.

Anyway, did a very quick ABX of original CD vs -V0, -V2, -V4, and more or less as I expected I could tell -V4 (11/13), but not the other two: I knew I was guessing right away, so not much need to carry on - I know that's not very scientific, but man, ABX-ing is tedious, and I did just know I couldn't tell the difference, I didn't need to knowingly guess 20 times, for the statistics to tell me I was 'probably' guessing (or rather, unlikely to not be guessing).  Life's too short!

The conclusion from this, presumably, would be I should go with -V2 (or thereabouts, maybe -V3, subject to another test, if I can bear it).

So... anyway, researching further, I came upon the HA LAME wiki page, and in particular, the perceived quality vs file size chart.

Question: how was this produced?  Whether I can tell the difference or not, the perceived quality curve is monotonic upwards as bitrate goes up, with -b 320 top of the heap, even if the rate of improvement wrt file size tails off (after -V3, as it happens, if you plot the values in the chart on an x-y scatterplot).

See... even if I can't perceive the difference, I have children with sharper hearing who might, other people listening to the music who might, and I have plenty of disk space, so I'm not that bothered at using almost twice as much of it for -b 320 relative to -V2 if, on some scale of measurement (the one on that chart), it actually is 'better' than -V2.  In other words, in terms of a cost-benefit calculation, the marginal cost of -b 320 is effectively nil (I already have all the disk space I need) and there may be some intangible benefit, even if it's just that it makes me feel better!

So... what's the source of the data in that chart?  Can anyone enlighten me?

Apart from its relevance to my decision, it seems like sourcing it would be in the spirit of HA!

Source for the perceived quality vs file size chart in the LAME wiki?

Reply #1 –
Gabriel, a LAME developer, came up with the quality figures here:

http://www.hydrogenaudio.org/forums/index....st&p=329935

And I cam up with the chart here (a few posts further down):

http://www.hydrogenaudio.org/forums/index....st&p=329974

Note that the graph is based on suggested quality figures from Gabriel, and file size data for one album.  I posted the graph to visually demonstrate an example of Gabriel's values against real-world data.

When it really comes down to it, ABX - plus maybe a little headroom - is still the best way to determine what setting you should use.

Edit: I guess the only other thing I can add is that most people would say that 320kbps CBR is total overkill, and if you are really concerned about quality -V0 would be a lot smarter option.  I guess I could also touch on the fact that, if quality is a real concern for you, perhaps keeping a lossless copy of your CDs, which your kids can transcode to -V1 or -V0 if they ever need to, would settle your mind.  Rip to lossless, encode to -V4, and then if you have a problematic track, album, child or friend use your lossless version to create another higher quality/higher filesize lossy version.
I'm on a horse.

Source for the perceived quality vs file size chart in the LAME wiki?

Reply #2 –
Your conclusion is quite right. But if there is "infinite" diskspace, you should go lossless, then the calculation would go in favor of that, as there has no compromise has to take place.
If you own a portable device (which IS very probably limited in diskspace - 80 gb devices aren't THAT common), you COULD go with a lower setting (320 CBR is waste most of the time, -V0 is good enough probably, even the aforementioned -V2 would be transparent in more than 99% of all cases - even for the kids).
Take this as an input of ideas on your situation.


Source for the perceived quality vs file size chart in the LAME wiki?

Reply #3 –
Thanks for the speedy replies.

Huh... so the perceived quality numbers are pretty much made up?!  That's not to criticize anyone, since any measure of 'quality' is likely to be subjective to some degree, but it seems like some information to that effect should be in the wiki!  Was there any group listening test or something from which the quality numbers were derived?

FWIW here's my take on those numbers - just replotted as x-y data (haven't figured out how to embed an image yet).

That suggests there's a big change in terms of bang for the buck around -V3.  But it's obviously completely dependent on the numbers on the vertical scale!

I think the general advice that lossless rips of one kind or another (certainly for the vinyl) so they can be archived (the CDs are their own archive) and something saner than CBR320 for living room media center playback - somewhere in the V2 to V0 range makes sense.  Of course heading down that path, ultimately it makes some sort of sense to rip it all losslessly, so that it can be batch-transcoded to whatever as required, even for the CDs.

Lossless unfortunately makes my disk space not infinite: although extra drives aren't expensive money has recently been spent on the system, so I can't rush out and spend more right away.

Off to the lossless forums to find out about tagging then...

Source for the perceived quality vs file size chart in the LAME wiki?

Reply #4 –
Quote from: davidos on
Lossless unfortunately makes my disk space not infinite: although extra drives aren't expensive money has recently been spent on the system, so I can't rush out and spend more right away.

You can get an external 1.5TB drive for $100 and store more than 3000 CDs in lossless format for about $0.03 per CD. To me this seems like a no-brainer.

Source for the perceived quality vs file size chart in the LAME wiki?

Reply #5 –
Quote from: pdq on
Quote from: davidos on
Lossless unfortunately makes my disk space not infinite: although extra drives aren't expensive money has recently been spent on the system, so I can't rush out and spend more right away.

You can get an external 1.5TB drive for $100 and store more than 3000 CDs in lossless format for about $0.03 per CD. To me this seems like a no-brainer.


You're right of course (although the maths is different in NZ$). 

Also, it occurred to me after I posted that, that I won't be getting everything (re)ripped any time soon.  Everything is already ripped at quality levels we can live with, and a 'background-task' re-ripping-to-lossless will take some while before it means we need more storage anyway.  So I can make that 'policy change' right away, just as soon as I figure out all the ins and outs of taggable lossless in Windows MC (the frontend to the music on the family-friendly HTPC).

Source for the perceived quality vs file size chart in the LAME wiki?

Reply #6 –
Relatively younger audiences than you may not notice the difference on casual or even hi-fi listening, unless you also subject them to rigorous ABX tests.

Enjoy your music transparent to your ears.  The perceived difference (in percentage) between -V2 to -V0 may not be equitable to bitrate difference.

The chart kinks from -V3 onwards; -V2 maybe the safer setting for you.  As posted earlier, -V2 is transparent in 99% of cases, I think the remaining 1% should not bother you to worry about the kids or storage space, even if it's cheap.
"Listen to me...
Never take unsolicited advice..."

Source for the perceived quality vs file size chart in the LAME wiki?

Reply #7 –
In the context of lossy codecs quality refers to the percentage of tracks in the universe of music with audible deviations from the orginal as well as of the amount of annoyance audible deviations can bring.
Nobody really knows that of course as nobody has tested the universe of music. There is sufficient experience however usually to get a reasonable feeling about codecs' quality. This is what the graphs show.

Whether you choose -V2, or -V0 or whatever high quality setting depends on your degree of acceptance of rare audible issues of a more or less small amount of annoyance.
It also depends a bit on your musical taste. If you're a lot into electronic music with sharp impulses you're better off using an extremely high quality setting. The same is true if you listen a lot to harpsichord or music with castanets.
The less file size is a real issue the more natural it is simply to use the highest quality setting you can afford.

If you don't have to care about file size you may want to consider to use lossyWAV preprocessed lossless codecs like FLAC or TAK. File size is higher than when using mp3 but a lot lower than when using pure lossless. Signal path however is the same as with pure lossless (no approximated transformation of the time domain to the frequency domain and back). lossyWAV just reduces the accuracy of the wave sample representation to the locally required amount of bits. 16 bits are needed to allow for a good accuracy of the wave sample representation in a global sense with respect to the entire dynamic range. Locally it is usually sufficient to use less than 16 most significant bits, and lossyWAV takes care of this.

lame3995o -Q1.7 --lowpass 17

Source for the perceived quality vs file size chart in the LAME wiki?

Reply #8 –
Quote from: halb27 on
In the context of lossy codecs quality refers to the percentage of tracks in the universe of music with audible deviations from the orginal as well as of the amount of annoyance audible deviations can bring.
Nobody really knows that of course as nobody has tested the universe of music. There is sufficient experience however usually to get a reasonable feeling about codecs' quality. This is what the graphs show.

Interesting to think about how you might measure this! 

That does makes things a bit clearer,although given that the quality numbers are in the 0 to 10 range, they must have been rescaled from the percentages you're talking about (and also reversed, since the higher quality settings would have lower percentages with audible deviations), it's still not clear that the 'elbow' in my chart a couple of posts back, or the 'levelling off' in the wiki chart, is actually meaningful.

Which brings it all back to ABX tests for individuals, I suppose!