[LAU] What is the best MP3 encoder?

[Peder Hedlund]

Quoting Ralf Mardorf <ralf.mardorf at alice-dsl.net>:

> On Tue, 2013-04-02 at 11:51 +0200, Peder Hedlund wrote:
>> Quoting "Jostein Chr. Andersen" <jostein at vait.se>:
>>
>> > On 04/02/2013 09:31 AM, Peder Hedlund wrote:
>> > ...
>> >> You really should try doing one and check if you *really* can hear the
>> >> difference between the original wav and an mp3 produced by, say, "lame
>> >> -V4" ( which would be ~160kbps) or if it's just your mind fooling you
>> >> into thinking you can. Never underestimate the power of belief :)

> MP3 is pure crap! It might work for the modern pop
> music my neighbour blast the whole day, were the wav files already sound
> like crap, because the whole production and composition of this music
> already is pure crap. MP3 is bad, loss always is audible.

Prove it in an ABX test. I think you'll be surprised how low in  
bitrate you need to go before being able to tell which is the "crap  
mp3" and which is the "pristine, shimmering wav" :)


>> Lame does have problems with certain types of samples, and good
>> problem samples are listed on
>> http://lame.sourceforge.net/quality.php,
>
> And this wasn't sent at April the first ;). So there are problems with
> _some samples_. What is the definition of a really good ear? my
> definition is that you should have healthy ears.

No psychoacoustic encoder is 100% foolproof which means there will  
always be certain things it'll have problem coding properly and since  
MP3 is an older technique, as compared to newer algorithms like  
Vorbis, AAC, Opus and such, it might be slightly more prone to fail on  
those particularly problematic samples.
That's not to say the average listener would be able to hear that anyway.

Good ears obviously implies not having been standing next to a  
jackhammer for your whole life but in the case of audio testing good  
or "golden" ears often means people who are really good at hearing  
stuff like encoder added pre-echo, phasing problems and such.


> Note even without using a lossy codec, some high quality recordings
> sound equal at 48KHz and 32KHz, but some audio material can't be
> recorded at 32KHz without getting a recording that is that bad, that you
> can't listen to it. _Some samples_ will cause issues and other material
> doesn't cause issues.

You should read Monty's (of Ogg/Vorbis and Opus/CELT fame) 24/192 post  
: http://people.xiph.org/~xiphmont/demo/neil-young.html
When recording you should obviously go as high as feasible but 16/44.1  
is more than adequate for listening.

- Peder