Guys, I am having a discussion with someone regarding the best mp3 encoder and settings to use for audio files that consist of a single voice (e.g. a lecture).

I assume that LAME, as usual, is the best encoder?

And is VBR still the best mode to use, say using
-V 9 --vbr-new
?

The LAME settings guide
http://www.hydrogenaudio.org/forums/index....showtopic=28124
simultaneously says
"At a given bitrate range, the quality scale is following: VBR is higher quality than ABR which is higher quality than CBR"
as well as
"For very low bitrates, up to 100kbps, ABR is most often the best solution."

So for a lecture, instead of a low VBR mode like -V 9, should I instead do something like
--abr 64 -mm
(where the -mm while make it mono)?

Note that streamability is probably not important.

I tried searching these forums for any previous discussions and had no luck, but feel free to point out anything that I may have missed.

Stepping back from specific recommendations, what is the optimal way to go about deciding "best" in this context anyways? I understand from these forums that using FFTs/spectrum analysis, at least with simple visual interpretations, is a disaster for music, and that the best thing is to conduct many ABX tests on actual sound files. Is this correct, or are there any quantitative/computer/mathematical tools that can help you determine quality?

Some relevant links:

http://www.hydrogenaudio.org/forums/index....showtopic=39500

http://www.hydrogenaudio.org/forums/index....showtopic=35214

http://www.hydrogenaudio.org/forums/index....showtopic=36879

Thanks much for the links. Looks like I should have searched for "speech" instead of "voice"...

It looks like the best advice comes from a guy named esa372 who has found good results with LAME with these settings:
-V3 --vbr-new --lowpass 8
which for speech seems to produce files in the ~48kbps range.

One disturbing thing is that in this discussion
http://www.hydrogenaudio.org/forums/index....showtopic=35214
Sunhillow makes the claim that
"many portable mp3 players do not support sample rates < 32 kHz (mine does not even support 32 kHz)"
which led him to use
-V7 --resample 44100
for speech. Note that the above setting will still use stereo even for speech since it allows reverberation effects that make it sound more natural.

Whoa! Anyone know how common of a problem it is among portable mp3 players to not support various features of the mp3 standard like alternative sample rates?

Also, from my original posting, is doing ABX tests the only way to determine sound quality?

Just to be clear... I use those settings with mono source files.


I've heard of bit-rate limitations on portable players, but not sample-rate limitations. I've never had any problems playing 22.05kHz mono files on my Rio Karma.

Just try a few mono samples @ 22.05kHz on your player; that's the best way to see if it will work.


It's certainly the best. After all, the final judges in the world of audio are one's ears.

(Quote from an very old HA-post of rjamorim):
"The frequencies supported are different for each MPEG version:
MPEG1: 48, 44.1 and 32kHz
MPEG2: 24, 22.05 and 16kHz
MPEG2.5: 12, 11 and 8kHz."

A mp3-player wich doesn't support 32 kHz sampling frequency, is by def NOT a mp3 player - all mp3-players should minimum support MPEG1 Layer 3...

LAME supports Layer 3 of MPEG1 + MPEG2 + MPEG2.5. Even though most of the recent portables (with good reputation) also supports this, it is wise to chek out what sampling rates your portable actually supports - at least if you are planning to make GBs of MPEG2.5-files...

(MPEG 2.5 is not formally a (ISO-) standard, but has become a defacto standard).

I highly recommend you to try the following settings for LAME 3.97b2

-V7 --vbr-new -mm --resample 24

It averages at 24kbps. If you raise it to -V6, it will probably average at 30kbps.
My source is a language learning CD with a mild noise reduction.

all of these presets are using 3.97b1, and b2
for voice....i use

--preset 13 --noshort --scale 1.05 -q0 -mm --resample 8 --lowpass 3.9

or if the above is just a tad bit too muffled, then try

--preset 22 --noshort --scale 1.05 -q0 -mm --resample 16 --lowpass 7.0


if you want lowest bitrate possible, try this (about telephone quality):

--preset cbr 8 --noshort --scale 1.05 -q0 -mm --resample 8 --lowpass 3.3 --lowpass-width 0.4 --highpass 100