I've been reading stuff on this forum for some time now and well, now I've decided to register and ask some questions.

First of all, I realize that LAME 3.90.3 is the most tested one, and prolcaimed as "the best". I have also tried version 3.96 and 3.97 alpha, but I didn't like the fact that they always cut off some frequencies (about 20kHz and onwards) even if I give them a --lowpass 22500 parameter (it was verifyed that I can hear frequencies up to 21.5 kHz so I'm a bit sensitive about that). Why was this change made, and is it possible to turn this option off? (Is it even recommended to use any other version than 3.90.3?)

Next, again about the frequencies, why does --alt-preset standard (yes, I am "another noob asking about that "damn" --alt-preset standard again") use a lowpass filter and can I turn it off? Is there really no possibility that one could make a custom command line that would be better that --alt-preset standard?

I have also heard (and once or twice read) something about the objective tests used for audio compression codecs such as mp3, ogg and similar, in order to determine sound quality and fidelity. Is it possible for someone such as myself to run these tests? If so, how? If not, why?

Many thanks. I hope I don't get on your nerves (a lot)

LAME cuts off high frequencies because they are harder to hear, and the bits spent encoding those frequencies are better off used to encode lower frequencies, if the overall goal is to achieve transparency. Having tested that you can hear frequencies up to 21.5 kHz doesn't mean you'll hear them in regular music. They are very quiet compared to the rest, and thus much harder to pick out than when testing to determine if you can hear a certain frequency. If you think you can notice a difference when high frequencies are filtered, please read up on ABX testing methods if you'd like to be able to prove this.


There is a possibility that on a problem sample where there are still audible artifacts with --preset standard, a modified command line could reduce some of those artifacts. However, the chances are very very small, as the presets are very tuned and already use internal tweaks not available in any other command lines.

Officially, the Hydrogenaudio recommended Lame version is 3.90.3. Like you said, it's the most tested one. Nevertheless, many are already using 3.96.1 (it was released yesterday) since it wasn't shown to have significant problems and has better quality in some bitrate ranges (using the presets, iirc). People who use it may also end up contributing to Lame development, if they report issues.

You can tell the encoder to keep all frequencies using the -k option. But you should also know that's counter-productive, since the encoder may be "taking" bits for the high frequencies that would be better used elsewhere. Plus, if you use the presets (--alt-preset standard, extreme, insane), you shouldn't use this option, since they are optimized to work properly without extra swicthes (at least with 3.90.3)

Finally, just because you can hear isolated frequencies up to 21.5 kHz doesn't mean you can distinguish real music "lowpassed to lower" frequencies. It's rare to find an Audio CD with such bandwidth, and they usually are lowpassed to ~20/21 kHz, as far as I know. You should do an ABX test, with real music and not some isolated tones, to see if you can note such lowpass.



Objective forms of evaluation of lossy audio codecs are normally only useful for developers. The real tests are subjective (listening tests), since lossy formats use psychoacoutic models, among other things and because no tool can totally replace the human ears.

You can also read:

http://www.hydrogenaudio.org/forums/index....showtopic=24125
http://www.hydrogenaudio.org/forums/index....showtopic=11286

Cya

Heh, that's almost impossible (usual top for trained musician is about 19Khz, as far as I know, read and heared rurmors ) so if it is so really, then you are very lucky

Masking effect. Low frequency tone masks high freq tone with the lower energy and so on... Means pshyhoaqoustics
Read about blind tests (m0rbidini provided a link) and compare for yourself, if you can hear difference between encoded material and original...
Hope this will help a bit...
EDIT:

Lame 3.96.1 is unofficially recomended (at least by it's developers) for bitrates about 128Kbit.

It is impossible to improve the high frequency hearing of someone by training. The maximum audible frequency depends on the physical state of the ear. So hearing 21.5 kHz doesn't make him a genius, just someone with unusual ears.

Or a newborn.

In all seriousness, do an abx test and see if you can really hear a difference between --aps and the source.

OK. So, here goes a dumb question (but I can't think of a smarter one, or of a way to get the answer myself).

Is there a way to determine the actual difference between the original and the compressed audio? If so, how? What I mean is, will these tools mentioned above help me determine what is lost during the compression?

Thanks again.

ABX test.

Determining what was lost is not pertinent as that is the nature of a lossy codec. If you can't abx the original from the encode, then it is transparent. It doesn't matter what was lost.

http://www.hydrogenaudio.org/forums/index....howtopic=16295&

Thanks, guys. I appreciate the help. I don't guarantee that I won't ask more questions soon, though

It's virtually impossible for someone to survive childhood with that high of a hearing capability. Musicians have some of the worst hearing because their instruments are usually so loud or stage amplification so intense that their hearing is often worse than the average person.

Did you live in a sealed laboratory or on a farm with no electric tools up until now? If you live in a city (where most people are regularly exposed to 80db sounds) then there is no way you can hear up to 21.5k let alone 18.5k. I can hear up to 17.2k and everyone I have tested so far can't audibly hear a signal tone from ff123's website beyond 17k to 18k. 17k is the only usable range I can personally identify in all people I've given the test to and the 18k was only heard by a few people out of 50 or 60 that I have run this by on quality PC speakers. Nobody heard passed 18k that I've tested.

So, when you make the claim that you an hear 21.5k, a little backround to prove it would be nice to see.

mp3

The ability to hear a tone at 22Khz and the ability to hear a lowpassed music file are not the same thing. I am 43 years old and can still pick stuff up to about 19.5 but still fing APS transparent on everything.

As far as altering the command line, Lame just used to obey the last argument it got. So something like --alt-preset standard --lowpass 20 should work AFAIK although your bitrate will bloat a tad.

-k might induce a bit of ringing if your hearing is really so acute.

Alt preset insane might be the way to go for you if your hearing is exceptional.

When I had a hearing test, I wondered about what is the highest frequency others can hear. The person administering the test was kind enough to show me charts of other adults, showing reception up to 22 kHz.

MP3fan, maybe your PC speakers can't play above 18 kHz, or maybe they are extremely directive at this frequency and one can't hear the tone unless he is exactly in front of the speaker.
I rather trust medical tests done with calibrated hardware than all we do with soundcards and hifi devices.

Did he say it was a medical test? It's unheard of (if you'll pardon the pun!) for standard medical hearing tests to test such a high frequency. 8kHz is the bog standard limit, though some special tests may go higher. Indeed, higher tests are now encouraged, but I don't think they're standard practice over here yet. They certainly don't go above 20kHz! (Or even 16kHz?)

Anyway, the point is (and I can't believe it's not in the FAQ) that just because you can hear frequency X doesn't mean that you can detect a low pass cut-off just below frequency X (unless the source material contains a very strong unmasked harmonic component at frequency X, which is fairly unlikely).

I reckon there's usually a 1-3kHz gap between the highest pure tone you can detect and the highest low pass you can detect on "normal" music, or even white noise.

Other posters already covered the need for a lowpass filter when lossy coding.

Wintershade, how did you test your hearing?

Cheers,
David.

My hearing was tested through some medical tests (which, although not very accurate, especially not in my country, showed exceptional hearing), and some time ago i tried to listen to isolated frequencies (up to 22 kHz) on a friend's system (with wide freq range headphones), where I was able to hear 21,5 kHz (well, not very well to be honest, but I still heard it), while I wasn't able to hear 22 kHz .
(I am buying myself such a set of headphones soon.)

Neverthelss, this wasn't the main reason why I posted the questions here.
I'm just trying to find a way to get the (at least near) size and the quality of say, --alt-preset standard, and still preserve all the trebles (not just below 19,5 kHz or so).
(I'll be asking some more in the new topic).

Thanks again for your help, guys.

I just wrote tone://22000,50 to foobar's "add location" box and pumbed up the volume to the max and, alas, I was able to hear it, but the volume must have been something like 100 dB. So that doesn't really mean that I can hear it.