I been thinking about mp3 conversion lately. I don’t know exactly how it works. I know the general theory but I don’t know anything else. One of the questions I have always wondered, “Is it necessary to use bit reduction and convert a 24 bit song to 16bits before mp3 conversion?”
After thinking about it, I realized that it probably doesn’t matter. It would most likely depend on the encoder’s ability to deal with 24 bit files. Just going through the simple logic of it, any gains by using dither would be negligible due to the fact that the file is being converted into a lossy format. Which mean any distortion caused by truncation at the 16th bit, would probably not be noticible for two reasons. Reason 1 is that it would mostly likely be masked by the degradation caused through the encoding process. Reason 2, either way the effect of the dither will be un-noticible except in extreme circumstances of bit reduction or in certain listening situations.
My answer to the question of whether a 24bit wav should be converted to 16bits with dithered applied prior to conversion to an mp3 would be, no. Recently I had to make an mp3 out of one of my projects in Logic Studio. The project was a combination of audio and virtual instruments. I do everything in 24bit when I write music. I didn’t even bother to bounce it out and convert to 16, instead iIjust did a straight bounce to mp3 conversion that is in the bounce options. I did not notice any terrible “nasty distortion” from the truncation to 16 bit. All I heard was the usual lossy sound of a 128kbps mp3.
I’ve wondered about this too. I know some MP3 encoders are limited to 16-bit (or at least PCM) input, but I would use a higher-resolution source by preference as there would be less quantization noise for the encoder to have to deal with. I think the standard Ogg Vorbis encoder can accept floating-point waveform input, for example.
The problem I’d envisage with dithering before encoding would be that there’s no guarantee that the effect or character of the dither would be preserved in the encoding. This is especially true for noise-shaped dither, which has more energy in the high frequencies and would likely be completely discarded by the encoder.
AFAIK, most lossy codecs operate on floating-point internally, so there may be some benefit to using dither (or decoding to more than 16-bits) on playback. There are certainly decoders that implement dither, and I assume they wouldn’t bother if it weren’t beneficial.
Good sonic food for thought!
Cheers.