mp3 to aiff

October 27, 2006 8:32 AM

mp3-aiff: What's the best way to preserve the questionable quality of mp3s when converting to aiff so that they'll play on a mix cd?
posted by OmieWise to media & arts (15 comments total)

Umm, what?

Are you trying to preserve the bad sound of an mp3? If that's the case, any mp3-aiff convertor will make that happen. When you convert from mp3 to aiff, you'll have a wonderfully accurate file describing all the mp3-ness of the original file.

Or are you trying to keep the mp3 from degrading any further? If that's the case, the only way to make it worse is to re-encode it. Otherwise, the above holds true, and converting to aiff will give you a high-quality aiff of a low-quality mp3.

And really, why are you doing this by hand? Shouldn't you have a program that does all the conversion magic for you? Just wondering.
posted by god hates math at 8:43 AM on October 27, 2006

converting to aiff will give you a high-quality aiff of a low-quality mp3.

If that's so, that answers my question. I was under the impression that converting from mp3--->aiff could lead to further degredation, although I'm not sure why that would be now that I think it out in writing. I guess aiff isn't compressing the mp3 any, so why should it degrade it any?

Thanks for helping me to see the light.
posted by OmieWise at 8:46 AM on October 27, 2006

Decoding should always give you the same result. The step you apparently can't control, encoding, is what you would have tweaked. Or just not gone through the process at all.

Ah, you might be tempted to normalize the volume on the mix CD. If you really care, then don't. You'll have to turn the volume up and down on various tracks, but it would preserve the dubious quality.
posted by adipocere at 8:54 AM on October 27, 2006

I have questions about this too. I know that with mpeg decoding, while the internal format is standard, the exact method of decoding is proprietary and some are slightly better than others for the same source materials. This can depend on hueristic reconstruction algorithms or on the precision of the data representations.

Is this also the case with mp3s?
posted by StickyCarpet at 8:59 AM on October 27, 2006

For gain correction use the free mp3gain program (google it) to adjust the volumes before converting. This will only change the offset in the header, and not any of the data in the soundfile.
posted by StickyCarpet at 9:06 AM on October 27, 2006

MP3s are part of the MPEG format family: the name means "MPEG-1, layer 3."

There are differences in both encoders and decoders--the LAME encoder is thought to be especially good. So yes, the format is standard, but there are some ways of getting to it and from it that are better than others.
posted by adamrice at 9:08 AM on October 27, 2006

As an example of how one converter could be better than another:

The converter may or may not add a randomly dithered signal as the least significant bit (better if it does.)

The dithering may or may not have a weighted distribution that takes harmonic musical perception into consideration.
posted by StickyCarpet at 9:11 AM on October 27, 2006

The point here is that AIFF is lossless (and uncompressed, but that's not that relevant). So you can convert *to* AIFF without losing any information.

Are you sure that MPEG (the original one at least, MPEG 1) output isn't standardised Sticky? I've seen reference outputs for the audio codecs of MPEG 1, and I was under the impression that they existed for the video codec and multiplexing too.
posted by fvw at 9:11 AM on October 27, 2006

Mp3 gain adjustment without reencoding doesn't exactly modify any offsets in the MPEG data by the way, it just adds a tag that specifies the gain to be used on playback, and if your mp3 player supports it it'll read the tag and attenuate or amplify as needed.
posted by fvw at 9:14 AM on October 27, 2006

After reading adamrice's comment I should add that I'm only talking about the decoding of a certain MPEG stream, which I think SC was talking about. Adam is right that encoding is an entirely different kettle of fish and much more of a free-for-all.
posted by fvw at 9:16 AM on October 27, 2006

Yeah, encoding is a huge free-for-all. I've used two codecs on the same audio and gotten wildly varying results - one would sound exquisite (for an mp3), and the other would be totally unlistenable. Certain codecs even allow you to make a speed/quality tradeoff within a certain bitrate.

Personally, I think LAME is fantastic, and MP3pro is also pretty darn good. Your ears may say otherwise.
posted by god hates math at 9:32 AM on October 27, 2006

Sticky: Yes, but the differences between decoders are much smaller than the differences between encoders. Decoding an MP3 stream to samples is a pretty much deterministic process, but as you mention, different decoders may have slightly different tweaks for speed, attempts to hide the MP3 artifacts, different numerical roundoff, or whatever. Encoding an MP3, on the other hand, is full of judgment calls that are totally up to the discretion of the encoder (and its model of the human perceptual system): different encoders may make very different decisions about how to best represent a given soundfile as an MP3.

In practice, I doubt there's any perceptible difference between one MP3->AIFF conversion and another.

On preview, what everyone else said.
posted by hattifattener at 10:12 AM on October 27, 2006

It sounds like people are talking about MPEG video, where 'better' often means more post-processing..
posted by unmake at 10:12 AM on October 27, 2006

unmake: better can also mean better, as in more like the original signal. If the MPEG video encoding says: this block of fixed-pattern data just switched to this other fixed-pattern data, the decoder could 1) just switch, 2) crossdissolve, 3) cross-correlatate and morph, where 3 is much harder to implement but may better match the input signal.
posted by StickyCarpet at 10:19 AM on October 27, 2006

From man madplay:

Among the special features of MAD are 24-bit PCM resolution and 100% fixed-point (integer) computation. Since MAD is implemented entirely without the use of floating point arithmetic, it performs especially well on architectures without an FPU.

Because it decodes the compressed data to a waveform as 24-bit data, in order to pump it out your 16-bit soundcard it has to throw away some detail. madplay offers a dithering option (as StickyCarpet described) that should make for a nicer sound than pure truncation from 24 to 16 bits and should sound 'smoother' than plain 16-bit rendering to the ear, the way color dithering can make gradations smoother in image files.

The MAD MPEG audio decoding library is often thought to be among the best sounding decoders, and among the fastest.
posted by xiojason at 1:26 PM on October 27, 2006