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: Documentation for LAME v0 and previous listening tests (Read 9412 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Documentation for LAME v0 and previous listening tests

Hello world.

I am to design a listening test on the MP3 format, and I wish to test for audible differences between v0 quality LAME encoded MP3 and a lossless audio format. I am currently perusing whatever I can in order to find information on this, and I thought I'd ask here as well, in case any of you know any good sources I haven't discovered. Specifically I'm looking for previous studies comparing different audio formats, especially MP3, plus some documentation on LAME v0.

Any help is much appreciated.

Thanks
eiterorm

Documentation for LAME v0 and previous listening tests

Reply #1
wish to test for audible differences between v0 quality LAME encoded MP3 and a lossless audio format.


You might want to use the search word "killer sample". If you expose a few random persons to a few random samples then you must expect to work for long to get a result that beats the coin.

Documentation for LAME v0 and previous listening tests

Reply #2
If you just pick normal samples you probably will not find a difference as v0 is generally transparent. You can pick killer samples that break lame, but the results may not generalize to the real world.

What is your goal for this listening test?

 

Documentation for LAME v0 and previous listening tests

Reply #3
At the risk of seeming overly obvious, searching for previous threads in Listening Tests and the two subfora for MP3 ought to provide you with plenty of existing information. Since you have specifically asked here, should we presume that you already searched and did not find what you were looking for, or enough of it?

Documentation for LAME v0 and previous listening tests

Reply #4
You can download some killer samples from here.
eig_essence, lead-voice, harp40_1, herding_calls, trumpet_myPrince are samples where you might hear a deviation from the original.
lame3995o -Q1.7 --lowpass 17

Documentation for LAME v0 and previous listening tests

Reply #5
Thanks for your replies.

Unfortunately, samples is not exactly what I'm looking for. I'm simply looking for textual sources (references) showing related work (listening tests). Here are a couple of examples:

[1] Pras, A., R. Zimmerman, D. Levitin, and C. Guastavino (2009). Subjective evaluation of mp3 compression for different musical genres. In Proceedings of the 127th Convention of the Audio Engineering Society. New York, USA.

[2] C. Lee and A. Horner, "Discrimination of MP3-Compressed Musical Instrument Tones" J. Audio Eng. Soc., vol. 58, pp. 487--497 (2010 June).



[1] is a related study which compared CD audio to LAME encoded mp3 audio of different bit rates. This study suggests that there is no audible difference between CD audio and LAME encoded mp3 audio with CBR 320 and 256 kb/s. However, this is only a convention paper, which means that it has not been peer reviewed. [2] is a peer reviewed, published study regarding the discrimination of mp3 compressed single tones at different bit rates, but the stimuli are only encoded with bit rates of 160 kb/s and below.

So [1] is not peer reviewed (albeit very relevant for my study) and [2] is peer reviewed, but only remotely relevant, because the stimuli had bit rates only in the lower part of the spectrum. I'm trying to find more relevant references like these. I have a personal motivation for my study, but I also need an academic foundation to initiate my study. Therefore I am asking you whether you know of any relevant studies.



I would add this to my initial post, but I can't seem to find an "Edit" button.

Thanks
eiterorm

Documentation for LAME v0 and previous listening tests

Reply #6
Peer reviewed references about the audio quality of 20+ year old audio formats are really unusual bitrates may not be very easy to find.

Documentation for LAME v0 and previous listening tests

Reply #7
[1] is a related study which compared CD audio to LAME encoded mp3 audio of different bit rates. This study suggests that there is no audible difference between CD audio and LAME encoded mp3 audio with CBR 320 and 256 kb/s.


It baffles me how some "specialized" sources seem to blatantly refuse to acknowledge the existence of Lame VBR in their findings.

Quote
I would add this to my initial post, but I can't seem to find an "Edit" button.


Now you can only (and rightly so, I must say) edit a post up to two hours following its creation.
Listen to the music, not the media it's on.
União e reconstrução


Documentation for LAME v0 and previous listening tests

Reply #9
Doh! After failing to find it on the TOS list, my crappy memory and me could swear it was two hours.

Sorry & Thanks.
Listen to the music, not the media it's on.
União e reconstrução

Documentation for LAME v0 and previous listening tests

Reply #10
While not peer-reviewed, the public ABX listening tests from Hydrogen Audio are probably the best you're going to get.  Though be careful, because many are done with pre-selected killer samples, which biases the results.  I'm sure there was a random music-sample test done at some point...  Can anyone more active remember?

Additionally, I remember past high-bitrate tests illustrated the difficulty of distinguishing even many "killer" samples.

Documentation for LAME v0 and previous listening tests

Reply #11
Public listening tests of V0 on non-killer samples is probably pointless. Only the most extraordinarily talented listeners will hear any difference. Listening tests are more about things that the average listener cares about.

Documentation for LAME v0 and previous listening tests

Reply #12
You can download some killer samples from here.
eig_essence, lead-voice, harp40_1, herding_calls, trumpet_myPrince are samples where you might hear a deviation from the original.


(Bumping an old thread, I know.)

Have just downloaded them and encoded them using LAME 3.99 (64-bit) with options -b32 -q0 -V0 -mj.

I can't hear any differences between FLAC and MP3 using Winamp.

Am playing via a USB SPDIF interface, feeding an Audio Alchemy DAC-in-the-box, Chord Cobra II interconnect, Yamaha DSP-A592 amplifier (running in straight stereo mode), Audioquest 14/4 speaker cable, Mission 760iSE speakers (bi-wired).

I think it's safe to say that LAME (with the right options) is pretty much transparent.

Documentation for LAME v0 and previous listening tests

Reply #13
Lame -V2 (or perhaps -V3 or -V4) would probably give the same results without all the silly switches. Try it with foo_abx.

Documentation for LAME v0 and previous listening tests

Reply #14
As for lower -V settings and the samples I mentioned:
Relatively recently I did a personal listening test in order to see what quality is like for lower bitrates than those I'm used to (and to see how other codecs like AAC behave in comparison).
My conclusion was:

-V3: quality is quite acceptable even for these hard samples, but problems are usally easily audible
-V2: no significant improvement vs. -V3
-V1: quality is very good even for these samples. The difference is still audible for the harder samples, but it's more or less negligible to me.

I used my lame3100m variant of Lame 3.100a2 for the test, but with the exception of eig I don't think my variant is essential here compared to Lame3.100a2. I think however 3.100a2 has the edge over 3.99.5 with these settings when it comes to problems of the trumpet_myPrince type.
lame3995o -Q1.7 --lowpass 17