[Ffmpeg-devel] alsa input / output

Måns Rullgård mans
Wed Mar 21 16:26:33 CET 2007

Vladimir Mosgalin said:
> Hi Michael Niedermayer!
>  On 2007.03.21 at 12:53:30 +0100, Michael Niedermayer wrote next:
>> you said "(believe it or not, but SRC_SINC_FASTEST sounds better than
>> SRC_SINC_BEST_QUALITY and all other SRC_SINC libsamplerate filters to
>> me)" which of course you cliped away from the quote
> I can say that again, so what? It's not like BEST_QUALITY has bad
> quality. It's just that I see no use of it.
> I believe in math, but I don't believe in pure theory. I believe my
> ears.  Another example - everyone says that vorbis is better than mp3,
> however I can hear that it is much worse! True, for very low-bitrate

The output from vorbis is measurably closer to the original than that of mp3.
It really is that simple.  This is independent of any psychoacoustic models
that might have been used during quantization.

> audio vorbis can be better, however when you increase it, mp3 quality
> improves, and vorbis still contains same audible artifacts, just like on
> lower bitrates. Does it mean that I should use vorbis instead of mp3
> just because it is "supposed" to be better? (I'm talking about same
> bitrate in 256/320kbit range)

Obviously you find distorted music more pleasing to listen to.  A resampler
should give as accurate results as is mathematically possible, although
slight deviations can be tolerated for performance reasons.  You can always
apply an effect filter separately if you think it makes an improvement.

>> ffmp3 has dithered output by default you just have to belive it for
>> the placebo effect to work that is unless you use artificial tests with
>> the volume turned up and the content being very silent
> I'm glad for ffmp3, but mad's quality is better. At least it was better
> a few years ago. mad hasn't improved since then, but has ffmp3 gotten
> better?

The only difference I've noticed between FFmpeg and mad is that mad
used to handle bitstream errors different.  FFmpeg would sometimes make
a loud chirp where mad would not.  I'm guessing it dropped a few more

> Anyway I don't think you can get more from mp3 sample than what mad
> gives to you.

Any half-decent mp3 decoder will give the correct decoded samples.

M?ns Rullg?rd
mans at mansr.com

More information about the ffmpeg-devel mailing list