I apologize that I'm not a native speaker

@ greynol. It wasn't my intention to build an oxymoron , sorry for that. I haven't thought about that someone could also think transcoding from lossy to lossless can also mean lossy archiving. But as you can see in my first post in this article, i mean only taking a lossy alternative for backing up music.




Good point. So choosing a lossy format with psychoacoustic model for archiving wouldn't be that good as selecting a no-psychoacoustic one like lossywav and wv lossy. Are there other codecs without psychoacoustics which could be interesting (except gogo with --nopsycho )

Well, let's see first how lossyWAV is...

IMO, there are only two "efficient" options for "lossy archival" - but lets first define two scenarios:

In the first scenario, you may want to do heavy postprocessing to the files at a later time (who knows? maybe some magic "transform 2.0 into 7.1"-dsp or whatever).

In the second scenario, you are certain that you will NEVER do that. You are satisfied with listening to the files "as they are".

In both scenarios, we asume that for listening, you will want them to be playable on anything - which means "transcoding if necessary".


This basically leaves us with just two efficient solutions:

For scenario 1, only a lossy encoder which does not use a psymodel (or is very conservative with it) will provide the necessary robustness for transcoding AND heavy postprocessing. That means stuff like wavpack lossy at 400-500kbps. You may also want to apply a soft lowpass at 19khz or so (you highly probably cannot hear stuff above that anyways, AND those ultra-high freqs require the most bits).

For scenario 2, there is another surprisingly simple solution, which will drive any purist insane: just encode to high-bitrate mp3!!! Perhaps play it safe and go high-bitrate ABR. Why MP3? Well, mp3 plays anywhere - no need to transcode.... if you just want to be able to listen to those files "as is", then you just need them to be in a format which is very compatible - and have them in a soundquality which gives high robustness against rare psymodel mistakes - high-bitrate LAME ABR @260kbps does exactly that. There also are no problems with being future-proof, because mp3 is so damn popular, that by the time mp3 becomes obsolete in like 10-15 years or so, storage-spaces will have rocketed so much, that converting them to a modern lossless format is no issue anymore. All this, an exceptionally low risk of running into soundquality issues - as long as you just listen to the music "as is".

Agree. I went high quality (-v0) MP3 lossy only. Later down the line if MP3 fades into history I'll simply re-encode to lossless.

C.

Indeed. You could even go for a LAME bitrate closer to 200kbps (such as whatever the modern equivalent of "--alt-preset standard" may be), in turn never having to transcode while also maintaining reliable transparency.

There have been solutions to this for years. They're not universally supported, but FLAC isn't universally supported either.

Best:
http://www.hydrogenaudio.org/forums/index....showtopic=35654

...or, for the less picky, just using lame by itself.
...or even letting iTunes handle it.

Cheers,
David.


Lyx,

Yes, "good enough" mp3 is quite a logical choice!

To be confident that any possible future transformation will work, you need true lossless. I would be quite confident with lossywav, though it needs to check the difference channel to make me 99%+ confident (that's not in there yet). mp3 forced to keep the difference channel clean, and forced to keep higher frequencies clean, and forced to over encode everything by a little, would also make me quite confident - except when it ran out of bits, or when you encode known problem sounds/signals.

If you take a lossy file, run it through a vocal cut, run it through some DRC, boost the HF by 20dB, and it still sounds the same as the lossless file treated the same way, then it's probably fairly future proof. It doesn't mean it'll take transcoding though, and it doesn't mean there won't be problem samples.

Cheers,
David.

There is always a possibility that you will buy a better hardware in the future and the fact that you're not able to hear something with your current setup (i.e. it sounds "transparent" to you) doesn't mean you won't be able to hear it with the new one.
"Reliable transparency" is a term i wouldn't use in conjunction with lossy encoding. It's much too relative and "unreliable".
I was quite annoyed when I bought new headphones and a better soundcard and heard that those 192 kbit MP3s are not that "transparent" as I thought they were with my previous equipment. I had to bother my friend to lend me those CDs one more time. Those "birds" were just too annoying in my favorite music.
I wouldn't get too close to 200 kbits and even less using MP3.

Myth.

(Regarding "improving equipment". Switching to WORSE equipment though, may make artifacts visible. Then again, on crappy equipment, encoding-artifacts are the least of my worries.)

- Lyx

Lyx,

Well, if you start with headphones that can't reproduce above 10kHz, and then buy some that go all the way to 20kHz, it's hardly a myth that you'll suddenly hear artefacts.

However, Martel, no one on HA ever suggested using 192kbps CBR mp3 for death metal!

If my favourite genre wasn't mp3-friendly then I wouldn't use mp3. Remember the founder of this board (Dibrom) stopped using mp3 early on for exactly the same reason.

Hearing real problems with mp3 is a perfectly good reason to use lossless.

Cheers,
David.

Thats true in theory - in practice however, when people make such arguments, they aren't talking about THAT crappy equipment. Does something like that even exist i wonder? I mean, i know that theres really crappy stuff.... but THAT bad? I mean, even if i got my headset from fisher price, i'd expect more than that (not much though).

It hardly matters whether you hear more artefacts with better equipment or with worse equipment. The point is that artefacts that you do not hear now may become audible later on different equipment, so alowing some extra margin may make some sense.

I see. So, has this ever been OBSERVED with high-bitrate material, by a person understanding the point of DBTs? I have heard this argument over a hundred times on HA.org.... but haven't seen any posts yet verifying that. Surely, if this is such an important aspect, there must be a obervations of it going wrong, right......... or not?

Didn't you yourself just state "Switching to WORSE equipment though, may make artifacts visible"?

Theoretically, it is more probably, because worse equipment = more distortions = less headroom for the psymodel.

I'm still waiting for any verified and reliable evidence, pointing towards high-bitrate MP3 (200kbps and up) not being enough for changing equipment.... which is what Martell claimed, and which i called a "Myth".

Try some genre with REAL drumkits and properly recorded. The hat and cymbols go right to 20kHz and even over while they're almost pure-noise signal and so hard to encode.
If something lowpasses at 18 - 19 kHz or even produces "birdies" (not likely with current versions of LAME, but used to be a problem with FhG encoder) at these high frequencies, then switching from headphones going up to just 15-16 kHz to ones going up to 22 kHz may make quite an audible difference (it did for me). You might await "better" sound from better gear but you come up disappointed as your collection becomes the limit.
Any tampering with high frequency content may turn that lovely metallic (I don't mean genre) feeling of hi-hat into some fuzzy high-freq noise.
And, please, don't tell me that hi-hat is solely dedicated to metal.

You really think you can hear sound at 21 kHz? Do you ever tried to generate a sine at that frequency and playback it? A friend of mine burned out tweeters of his speakers - he generated 18 kHz sinewave and playbacked through his speakers. As he heared NOTHING he was raising volume and reached the maximum. Tweeters were screaming at maximum and burned while he was not able to hear them. So if you want to try - be careful.

The origins of lossy codecs artifacts are not at ultra high frequencies (if we are talking about such encoder as very well tuned LAME). As I remember - Dibrom made most tunings through his notebook's speakers. So most LISTENABLE artifacts have noting to do with 20 kHz frequencies. Anyway there are lot of old recordings which do not contain sound at at theese high frequencies but these recordings suffer from artifacts as often as new recordings.

I did this to my ears once, to a lesser extent of damage. I heard virtually nothing at IIRC 19k sine, thought what the heck, and pulled the volume control all the way up. It felt as if my ears were pierced with needles, for a moment everything went silent.

The sense of security from a generous bitrate might help you to actually enjoy the music, in case you often find yourself looking for artifacts.

Funny! I did the same thing also with headphones. Cool Edit's meter was all green and yet I couldn't hear anything so I turned my amp up and still nothing. My ears felt very strange afterwards - though not quite temporary deafness.

Questions
a) how do headphones handle this versus a speaker's tweeters?
b) related: why didn't it screw my headphones?
c) what did I do to my ears (since it's outside my hearing range)?

C.

I didn't say I heard 21kHz but I still hear 19 kHz (though with some attenuation, of course) which is a bit higher that 16 kHz that my previous headphones were able to put out. I did this headphone switch some years ago so I was talking about my experience from back then (and many people still used FhG encoder which caused high frequency artifacts). I didn't say that you WILL hear artifact if you use current MP3 around 200 kbits and switch to superior gear, just that the limited content MIGHT (~need not) limit your overall music experience with your new audio gear. So there is sure a reason to leave some headroom when encoding to archive.
And even if your hearing is the limiting factor, there is always a possibility that someone else (with better gear/ears) might use something from your archive.

I doubt that you will find reliable research about that on general sound and music platforms. However, i have two infobits - one is reliable and well known (though, often ignored) - the other is theoretical speculation.

1. Hearing and hearing damage, is neither specifically a "physical" thing, nor a "mental" thing - its both. For example, the sensivity of your hearing is strongly influenced by the overall state of mind, as well as what kind of soundlevels you "normally" hear (so, if for example you move from a city to "somewhere far far out", and then later back to the city, you will temporarily be way more sensitive). The same is also the case the other way around: if your ears take physical damage, your mind will adapt to it. This is important, because it means that mental changes as well as material changes may affect your hearing - and thats the precondition for the next point...

2. The ear itself is in princible just an "air movement and pressure sensor" - and too much "force"... especially longterm, but also shortterm, damages the sensor. For pure raw force to be there, the frequency doesn't matter. Therefore, i suspect that even if one cannot mentally hear something, it may still damage the material sensor, if the soundpressure is just high enough - this typically just is no concern, because in our normal environment, such high soundpressure at freqs outside of the ATH, do not happen. Again, this is pure analytical speculation without any practical evidence.... i'd be very interested if there has been research in that regard (though, i'd suspect it to be hard to get practical evidence - who wants to test becoming intentionally deaf?!? :)

The studies that have looked for hearing damage due to ultrasonic sounds haven't really found any.

Where they have found hearing damage apparently due to ultrasonic sounds, it's turned out that exposure also included plenty of sounds below 20kHz, which fully accounted for the measured hearing loss.

There was a call for more research in this area, since the current ultra sonic exposure safety limits are based on pure guesswork. This is really important for people who use equipment like ultrasonic cleaners etc.

Cheers,
David.

So we should listen to our testfiles on other (friend's) equipment too, and let our wives, girlfriends or daughters do some listening tests. At least I do. MY hearing is certainly the limiting factor here.

please post comments which refer to the topic!

What do you think of MP2? Would it be an attractive archiving codec?

If you can make it gapless, 384k is what you are looking at. I haven't done much tests on it.

Do you know some MP2 codecs? (I only know tooLame, is it a good one?).
I'll try it out.

I'd pick MPC over MP2 since its MP2 based and better tuned. Try MPC --quality 6 - 200k .

As shadowking said: MPC is the better MP2.

The only advantage of original mp2 is: there is a good chance that a mp3 playing DAP (that is in practice: every DAP) plays mp2 as well. Maybe the .mp2 extension has to be renamed as .mp3, and maybe bitrate shows up wrong, but there's a chance it will play.

If you really want to use mp2 I suggest you use at least 320 kbps, or, for best quality, 384 kbps (maybe this will be a problem for a mp3 machinery playing mp2).

The main question is: with which machinery do you want to play your music?

I think the charme of the approach discussed here is: you can have an archive with perfect quality in a practical sense which you can play on PC, or on a DAP, without conversion, and which is future-proof.

In this sense I see two main choices:

a) Lyx' proposal to use very high quality mp3 (for instance Lame ABR 260).
In a practical sense this is the best solution: you can play it everywhere yourself, pass it on to other members of your family, and you are future-safe with this format as is for a very long time from now. Whenever mp3 should die it will be at a time where you can transcode losslessly without pain, so you won't loose quality.

b) same as a) but you choose another favorite codec like Vorbis or AAC using a very high bitrate around or above 200 kbps. The situation is similar to a) with a tiny advantage qualitywise (but probability is close to zero that you will ever run upon a situation where you will take profit from this fact). As a disadvantage you're more restricted towards DAP usage but if you're about to buy a new DAP there are several DAPs to chose from nowadays which play Vorbis or AAC.
Using a lossy variant of a lossless codec like lossyWAV + FLAC or wvPack lossy falls into this category.
LossyWAV has the specific advantage here that in case someday transcoding is necessary transcoding to lossless is not only a lossless process, but you keep the lossyWAV advantage of a higher compression as long as you use one of the lossless codecs that make use of the lossyWAV properties.

[quote name='halb27' date='May 31 2008, 17:46' post='568283']
[quote name='memomai' post='568207' date='May 30 2008, 23:54']
please post comments which refer to the topic!


LossyWAV has the specific advantage here that in case someday transcoding is necessary transcoding to lossless is not only a lossless process, but you keep the lossyWAV advantage of a higher compression as long as you use one of the lossless codecs that make use of the lossyWAV properties.
[/quote]


That is a killer feature - I never knew it.

I've just tested lossyWAV + FLAC. I like it it reduces, for example a 991 kbps FLAC down to 497kbps. So it would make my music collection only to half the size... And I'm not recognizing any artifacts... But I don't have good ears anyway.

I think lossyWAV FLAC would be also a good source to transcode from into MP3/AAC/OGG etc. Or am I wrong?..

If the MP2 codec is not as good as MPC, then MP2 is dismissed ^^

PS: I compared lossyWAV FLAC and lossyWAV TAK (encoded from the same lossyWAV source). Surprisingly, FLAC -5 is smaller (497 kbps) than TAK -p4e (505 kbps) o.O

Why not? And, did you add blocksize=512 option? (-b 512 for FLAC, -fsl512 for TAK)

It should be transparent. I plan on doing tests at some stage but I don't expect to find anything. You have zero preecho + full bandwidth and that should lend to excellent transcoding.

Sounds excellent! It seems that lossyWAV is exactly what I'm looking for.



Uhm, no, I don't really know what purpose this setting has. I thought the presets of the codecs are the best ones and it's not recommended to "tune" the presets... but I'll try it out anyway, thanks for the info.

Yes, it's really great.

(Indeed, since FLAC is lossless, there's no difference between encoding lossy.wav to lossy.tak and transcoding lossy.flac to lossy.tak Maybe this advantage should be mentioned in lossyWAV's wiki page, FAQ section?)

yes, would be very useful.

This setting is specific to LossyWAV, because lossyWAV divides audio to blocks of 512 samples long and processes them independently. http://wiki.hydrogenaudio.org/index.php?ti...c_compatibility

I may attribute a bit to the background:
loosyWAV decides how many least significant bits can be removed (because they are inaudible) on a block basis of 512 samples.
The lossless codec that processes the lossyWAV result looks for the effective number of bits within its own blocks. In terms of making best use of lossyWAVs discarded bits the lossless codec's blocksize should be 512 too.

Ooops, ok now I changed it to blocksize=512. Now TAK delivers me less bitrate than FLAC ^^, thank you very much

Well, as it seems, lossyWAV + TAK will be my solution for my problem.

I have to also say that lossyWAV is a great implementation to lossless codecs to provide an alternative to lossless archiving.

Such a lossy archiving could be very attractive for users who have problems with space when they want to start archving.

Thank you all very much for supporting me, you've helped me a lot

I'm glad you found an encoding format you're happy with for this purpose.

In its very early days, it was dramatically better than high bitrate mp3 (320kbps) to medium bitrate mp3 (-V2) for mp3 problem samples. It wasn't perfect back then though.

A transcoding test with v1.0 final is overdue - I think a lot of use are interested, but none of us has quite got around to it!

Cheers,
David.

From wiki: http://wiki.hydrogenaudio.org/index.php?title=LossyWAV

I think the main point is that this retains both quality and filesize as long as you use compatible with lossyWAV codecs (as halb27 mentioned).
And, maybe lossyWAV should be mentioned at this page - http://wiki.hydrogenaudio.org/index.php?ti...less_comparison ?

Sorry i am on my Mac so i cant test these out yet.

I am just wondering what sort of bitrate would LossyWav + FLAC brings me?

Any has anyone done any ABX test against MP3 , OGG or AAC ?

Indicative bitrate reduction

Apart from the remarks on the 512 sample blocksize of lossyWav, TBeck has noted that with lossyWav setting above -p2m are useless. So in your case you should try TAK -p2e -fsl512 .