The sample Badvilbel is known to cause drop outs with Lame alt-presets (Discussion : http://www.audio-illumination.org/forums/i...=ST&f=16&t=1059 ).

The frequency analysis shows a frequency modulated sine around 18 kHz. The picture represents 0.6 seconds only.

http://perso.numericable.fr/laguill2/pictu...el/original.jpg

Nothing special visible in the spectrum of the 3.90.2 alt preset standard version :

http://perso.numericable.fr/laguill2/pictu...dvilbel/aps.jpg

Substracting the APS version from the original makes audible the drop outs : a background noise is suddenly switched on, meaning it was suddenly removed by the MP3 encoding :

http://perso.numericable.fr/laguill2/pictu...ancellation.jpg

The sonogram features a rectangular dark zone (very sharp on better pictures than these jpeg close-up), meaning first that what's above 18 kHz was removed, that is the normal inaudible behaviour of the lowpass, but secondly that something suddenly triggered the removal of an audible noise around sample 462,000.

Anyone knows if it is related to the behaviour of the FM sine at this point ?

Edit : there is not much difference with Lame 3.94a12