Re: The old MP2 to MP3 conversion issue.
- From: "Bryn Harris" <bryn.harris@xxxxxxxxxxxx>
- Date: Thu, 27 Oct 2005 19:32:33 GMT
<davidrobinson@xxxxxxxxxxxxxxxx> wrote in message
> Bryn Harris wrote:
> > O.k., in general I try to keep the mp2s I save from BBC Radio 3 DAB in
> > 'native' mp2 format. However, it is sometimes useful to have the greater
> > portability offered by mp3s. Now I know that, by and large, mp3 is a lot
> > more efficient in its use of data than mp2 is, but the question of the
> > frequency banding of mp2 has me wondering, "are there some frequency
> > where mp2 offers higher quality than mp3 for the same overall bit-rate,
> > if so, what multiple of the overall bit-rate should I choose for the mp3
> > order to retain all, or nearly all, the meager fidelity offered, in all
> > frequency bands, by the mp2. I'm thinking in terms of 192kbps discrete
> > stereo mp2s as the source, by the way.
> > I know, I know, I could do all the digging around in specs. myself, but
> > likes of Dr. Robertson and Mr. Green might just have such wisdom at
> > fingertips. Here's hoping.
> > :-)
> Oh, if only! ;-)
> Remember that lossy re-encoding can only lower the quality. Re-encoding
> at the highest possible bitrate (320kbps for mp3, unless you use the
> very rarely supported free-format mode) will give the most faithful
> copy of what you have, and when the source itself is a lossy encoding,
> there is no guarantee that further _audible_ loss will be avoided, even
> at the highest bitrate.
> I don't know of any kind of sweet spot at a lower bitrate either, or a
> magic relationship that will help. As a counter example, if you're
> recompressing images, and you know you have 8x8 pixel DCT blocks, you
> can just reverse the transform on each 8x8 block and get back to the
> DCT representation of the first compression (though colour
> sub-sampling, and even simple rounding, usually prevent this in
> However, with audio, almost all codecs use 50% overlapped blocks in the
> time domain, and varying amounts of overlap in the frequency domain.
> Even if you reverse the transform on the samples corresponding to a
> given decoded block, you can't get the "original" compressed data back,
> because 50% of the previous and next blocks are mixed in with the data
> you want.
> Given that you know you want the output to be mp3, just encode as if
> the source was a CD, using whatever settings you normally use, safe in
> the knowledge that the result won't sound as good as you'd expect, but
> might just be good enough. Some people suggest increasing the target
> bitrate to compensate for the poorer source, but often this just
> doesn't work (at least no where near proportionally). That's why I
> mentioned 320kbps - if you're just re-encoding for compatability
> reasons, rather than to save space, why not try it? Other people
> suggest DEcreasing the target bitrate because the source is poorer -
> that'll just make things worse, but if you want to save space and
> sacrifice quality, wht choice do you have?
> My approach (just use the same settings you would with a CD) will give
> you all the faults of the mp2, plus all the faults of the mp3, plus a
> few more faults on top due to the re-encoding. These extra faults could
> be unpredictable artefacts, or very predictable increases in coarseness
> or subtle noise.
> Not good news, but honest!
> P.S. Some people in codec circles are positively allergic to
> re-encoding. However, if I need to, I do it. I always keep the original
Thanks David, it is as I sort of feared. I think I will just tell those who
moan that they can't play the mp2s that they can always use WinAmp to bulk
them out to .wav files, resample them to 44.1k and burn to CD-R to listen to
- Prev by Date: Re: Go to my website a minute
- Next by Date: Re: DAB - The train without any driver
- Previous by thread: Re: The old MP2 to MP3 conversion issue.
- Next by thread: DVB-H T-DMB comparison