How dangerous is it if you decode MPC q6/q7/q8 and then reencode it to the same quality level. How likely is audable loss of quality?
Is MPC smart enough to see that everything that can be removed already is removed?

If you're wondering why i'm asking, it's because mpc doesn't support direct cutting.

Madman2003.

Lossy encoding is not only about removing audio data, but also about adding quantization noise. If you transcode a file 10 times, you will add additional quantization noise 10 times.

Unfortunately it´s not that easy. The encoder doesn´t simply "remove" parts but adds noise by coding "unimportant" parts less accurate. When decoding a lossy compressed file the resulting signal is not more simple than the original input.

There are similar effects on JPEG-images (compression is achieved in a "similar" way): You will see "noise" especially on hard edges - making the image "more complex" than before.

edit: too slow

(Will)/(how likely is it that) that noise become audable after one transcode?
And will the extra headroom on higher than q5 help in this case?

I know transcoding shouldn't be done as a regular thing, but i did it once and i couldn't hear the difference, now i'd like to know if that is a case of luck, bad ears or normal that you don't hear it after one transcode.

Madman2003.

The higher the -q quality level the lower the quantization noise and thus the lower the chance of hearing artifacts on a transcode. Generally. However, I don't think you can specify a quality level where it is "safe" to transcode. Even if you have sufficient headroom, mppenc employs other techniques like temporal postmasking and noise shaping that don't "translate" well on the second run around, even if you are using -q 10.

Hi mithrandir, good to see you 'round these parts again!

I just did a quick test of Musepack's transcoding ability. Picked a random song, 311's "No Control", encoded once from FLAC, then again from the encoded copy. Both encodes were -q5, mppenc v1.15r.

I used WinABX to compare the first-generation MPC to the second-generation MPC. First, I tried the opening seconds, and scored a wonderful 4/8. Then I tried a part in the middle, and scored 18/24, for a grand total of 22/32 (pval 2.5%). The twice-coded version had a noisier tambourine that didn't sound quite as "real". Of course, if I didn't have a copy of the original to closely compare with, there's no way I would have known the second-generation encode had anything wrong with it. I'd give it a 4.5 on the 1-to-5 scale of ABC/HR (perceptible, but not at all annoying).

My conclusion: transcoding MPC->MPC is unsafe. Perhaps later I'll try starting from higher -q values, or even testing the first-generation MPC against the original. I did an MPC transcoding a couple years ago, using an older encoder version (v0.90o, perhaps), and I remember the artifacts were much worse, even going from --insane to --xtreme. So MPC has certainly progressed in transcodability, and presumably in overall quality as well.

(Below are the details of my test setup)

For the first encode, I used Foobar's converter with replaygain disabled, but the DSP enabled (with crossfeed, convolver, and a few negative dB's of volume). I turned the DSP off for the second encode, and used Foobar's converter on the first-generation MPC. Foobar's replaygain says that there is a 0.14dB difference between the two tracks, the second encode being louder. Sound Forge also reports 0.1dB difference between the two, both in peak amplitude and RMS volume. I didn't apply replaygain to the second-encode because, really, there shouldn't be any difference in volume except for any additional noise introduced by the second transcode.