Way back when, EarGuy's Digital Ear had indicated that mp3 produced poor quality near the beginning of files. For example, see:

http://ff123.net/earmodel.html

where the last picture shows what I'm talking about.

Just today, I had the occasion to test applaud.wav, to figure out if someone's encoder was using FastEnc. I encoded my sample of applaud.wav with EasyMP3 (which uses FastEnc), and noticed that it sounded much worse than the one I received. Then I noticed that the sample I received had 0.5 seconds of silence at the beginning, whereas mine had none. When I inserted the same amount of silence into my applaud.wav sample, the encodes sounded the same!

Also, if I copy and paste applaud.wav and make a double-length sample, the second one sounds better, even when as much as 2 seconds of silence have been inserted before the beginning of the first one!

Strange.

ff123

Could this have something to do with the psymodel not having info about what precedes the first audible parts?

--
GCP

Tonality estimation in ISO Psychoacoustic Model II (Used in LAME and derived out of FhG's psymodel) requires two past frames to estimate tonality well. One possible workaround is to set tonality in the first two frames to the maximum (set unpredictability to 0.05)

Also, some other variables might also depend on past frame (or two).