Peter Olufsen po@dsinet.dk wrote:

1. Is it correct that the clipping appears in the decoder even though the

file was encoded "correct" without clipped samples ?

Yes, because of the nature of the MPEG audio transformations.

2. What is the "Other"-errors in the info-box, it seems that there is 1 og

2 in every file ?

They are anything not synchronization or CRC errors. :-) They usually indicate some sort of corruption of the bitstream. For example, if you use a program to split an MP3 into multiple files, the first frame of each resulting file other than the first may have a bad main_data_begin field pointing into the bit reservoir contained at the end of the previous file. Frames with such problems cannot be properly decoded.

3. I think it would be nice to have some form of auto-attenuation (and also

auto normalize-to-peak or even amping/compressing), but to make it perfect wouldn't you have to scan all your files first, or could you make some setup for that ?

The expected mode of use would be that an appropriate setting would be learned the first time a song is played, and recalled when the song is played again.

I have testet some files and they typically have clipping about 1db, but i found one file with 6db clipping in one sample (?), and if you auto-attenuate that the result is not good, so you should have some form of setup to prevent that.

This is a good point, but also a tough decision... how much clipping is acceptable?

Are there any audio technicians on the list (Anthony? :) who could go into details with me about sound energy levels, compression, normalization, etc.?

I have also tested some auto-amp-plugin and the result wasen't good because you could hear that the sound-level was adjusted when the songs changed in level.

Attenuation in the MAD plug-in is performed in such a way that this problem is hopefully avoided. Attenuation is not applied to the output PCM directly, but rather to the decoded subband frequency samples that are fed as input to the synthesis decoding step. In this way, state already accumulated in the synthesis filters allows the transition to be made more gradually.

Also, once the appropriate attenuation level is learned, it can be applied to the song from the beginning the next time it is played.

4. Would it be possible to use MAD's 24/32-bit output with some DSP-plugin

like DFX ? Or do you loose the advantage from MAD when you use DSP ?

If both the DSP and the output plug-in support 24- or 32-bit output, you may not lose very much in terms of quality.

Where you will lose is with 8- or 16-bit output because the DSP plug-in will receive low-resolution dithered output from MAD. It would be much better if the DSP plug-in could get undithered higher resolution samples, and then afterwards MAD could scale and dither the result for the output plug-in.

I might consider an option to do this in a future release. It will incur additional overhead though, and not all DSP plug-ins may be able to handle high-resolution samples.

Cheers, -rob