What's the best way to re-encode existing MP3s (mostly 320, 256 and LAME -V3 VBR) down to a lower bitrate so I can fit more music into a phone MP3 player's more restricted memory? I know it would be best to go back to the source WAV/CD and re-encode from scratch, but that's not always possible.

Is there a utilty that does this with the minium of extra loss, or should I just simply pass the existing high-bitrate files to LAME and get it to create new files?

Andre

None. No. Yes.

I'd try dbPowerAmp Music Converter. Easy to use, preserves ID3 tags. Free. Supports batch conversion. Uses LAME. Did I forget anything ?

Lionel

Yes. Transcoding sucks.

We all know that.

As the original poster stated:

"I know it would be best to go back to the source WAV/CD and re-encode from scratch, but that's not always possible."

I would think transcoding sucks less if you stick to the same format/lower bitrate as a target.
This is just a theory of course, but let's say you want to transcode a 320 Kbps MP3 to 96Kbps. The information lost between the original and the 320 Kbps would be lost if encoding from lossless/CD to 96Kbps anyway.
Obviously there is only one way to find out, ABX...

1. original (if available)
2. 320 Kbps MP3
3. 96 Kbps MP3 from the 320K
4. 96 Kbps MP3 from the source.

Or course YMMV but I could not tell the difference between #3 and #4 on the two samples I had tested (used lame VBR when applicable): Wagner's Ride of the Walkyries (or sthg, it's Walkürenritt), and Ferry Corsten's Punk (Live version).

Anyone like to comment on that? Anyone else tried this?

Micky

transcoding always sucks...no matter if you transcode to a higher or lower bitrate...

Ive tried this and couldnt hear the difference either, when i tried really hard i guessed the wrong tracks on my portable - but when i transcoded them i also resampled to 32Khz which it might play better -and my ears dont mind the difference.

Besides having easy ears to please, I do write code a bit and have some understanding of mp3s lossy process. My understanding, however incomplete, does not lead me to expect the disasterous results from transcoding that is insisted on here.

Consider how lame has been tweaked to deal as best it can with every type of weird sample that can be found to throw at it. The biggest danger from re-encoding is not loss on loss - if you are transcoding to a lower bitrate the second loss should mostly just eclipse the first. The danger is from effects/patterns/virtual stitches reinforcing each other, but i expect that most of such potentialy dangerous signatures - that can hide below hearing in the first encode, but could potentialy damage the next, well most types of oddities have had the heck tweaked out of them by all the test samples anyway. You know? theres bound to be the odd trace of virtual milli-voodoo in some of the CDs out there, but its been assimilated by successive versions and tweakfests - if the music corps could put pre-encode lame poison in there it would be part of the mastering process, so why should lame choke on its own copy? I dunno

So.. to make mp3s for my portable, i use foobars diskwriter and ssrc to resample to 32k lame --preset 96 or --preset cbr 112.

I wonder are there any listening tests archived documenting the deadly transcode effect....

Right on. I played around trying to get more files on my MuVo and found that abr 128 with a lowpass of 15.5 and resample 32 was the only thing that was at least listenable.

Otherwise the artifacting became like nails on a blackboard for me.

Eventually I quit even doing this as it just doesn't work.I have no particular skill hearing mp3 artifacts but the transcoding destroyed the quality.

Mp3 to ogg is much much much better.Mp3 to wma was also terrible.Never tried aac.

I've been transcoding MP3's music for my phone's ringer using this command line:

lame --preset 32 -m m --resample 32 --mp3input input.mp3 output.mp3

The --resample 32 keeps the resulting file as an MPEG I, as the phone seems to have issues with MPEG II.

The -m m downmixes to mono, since there's only one speaker on the thing anyway, it's not stereo.

Anyway, for this specific case, it works very well indeed. Considering the quality of the playback speaker, I honestly can't tell much, if any, difference between that and the original mp3 half the time.

Yes, transcoding is bad, but if you go low enough in quality then it's much less noticable than the quality difference that you hear from the low bitrate itself.

Assuming that you're going to transcode ANYways, I've found the most convenient program to be dBPowerAmp. It's ridiculously easy to transcode with, does batches without fuss.

At any rate (no pun intended), "it sucks" lacks the accuracy we all seem to cherish in this forums... In the present case, it can be loosely interpreted as "degrades the audio as compared to re-encoding from the original source" - which may be true with lossy algorithms in most cases from a purely mathematical standpoint, but may be untrue in many cases from a perceptual standpoint (read: as demonstrated by ABXing the results), and may not even be relevant when the target file is going to be low-bitrate anyway.

As some wise person said: All generalizations are false - including this one.

I've had to transcode a few mp3's for my dap and I've noticed some strangeness. The process I use is:

1. mp3gain a 192-320CBR mp3 down to ~89db if its not already there
2. EAC/decompress to get a WAV
3. EAC/compress (lately using 3.97a7 -V[4,5] --vbr-new) to mp3

Oddly, when the initial mp3 was a Lame encode (according to EncSpot), the final V4n or V5n mp3 is really quite listenable (to my tin ears) with a resulting bitrate of ~130-170. When the initial mp3 was a non-Lame encode, the final mp3 is usually gawd awful with a resulting bitrate of ~80-110.

I've only done this a few times, but the results are fairly consistent. When starting with a Lame encode, the result is usually ok. When starting with a non-Lame encode, the result is usually poor.

Just wondered if anyone else had seen this. Are some mp3 encoders lossier than others? Even at 320CBR?

Yes I'd love to see some listening tests on transcoding to lower bit rates. Right or wrong good or bad the fact is that sometimes people dont have access to the original source and may want to recode to smaller size for a portable device.

I guess that one difficulty with a transcode listening test would be sheer number of different possibilities to test. Never the less it would be great to see a "quality based" listening test (as oppossed to a simple "yes I can ABX it" test) of say Original source encoded to either ABR 128 or -V5 versus Lame -aps transcoded to ABR 128 or -V5. This would be a reaaly interesting test IMHO.

As I mentioned above, I cannot ABX them at 128 Kbps with MP3... but then it might just be me.
Also keep in mind that with an MP3-capable phone you have other issues than the encoding when it comes to sound quality, like ambient noise, the DAC quality, the earphones you're using, etc.

Yes I just transcoded a couple of files from Lame3.97-a7 --vbr-new (approx 170kbps) down to --abr 130 (128kbps) and certainly they sound ok to me. I compared them with encodes (also --abr 130) made from the original source and there are no obvious problems jumping out at me, though no doubt theese are not "problem samples".

My personal opinion is that the majority of users would not notice any significant loss of quality with these particular transcodes (compared with enocodes from origianl source) when listening on your average portable, especially in the typical listening environments that these devices are used.


Edit: Well I just did some more serious ABX tests between the above encoded/trancoded tracks and yes I can ABX them and yes the transcoded version sounds inferior. So this does support the widely held view here that transcoding is generally a bad thing. I still believe that the differences I heard (in the particular samples I chose) are not so obvious that the casual user of a portable device would likely notice.

I just made some neat&tricky samples to compare... just seeing how to upload them now, ill make a fresh thread I think - distance myself from my speculation ~