> > > > Take a look at LAME code, it has very good (for MP3, that is) M/S > > implementation and you can identify the parts you quoted in the psymodel.c > > code. There are some additional issues that need to be solved - i.e. to > > avoid constant toggling of M/S on and off (in MP3 that causes additional > > artifacts), and to verify does the bit allocation yield in bitrate > savings, > > etc.. > Is there any mechanism on avoiding the toggling effect in Lame. Also, what > is the effect on the constantly toggle >
IIRC, we average the switching criterion over several granules. Either the average has to be greater than some moderate value, or the un-averaged criterion has to be very large to trigger a M/S frame. It is not perfect, but it does prevent most cases which would trigger a single M/S frame surounded by lots of stereo frames. Mark _______________________________________________ mp3encoder mailing list [EMAIL PROTECTED] http://minnie.tuhs.org/mailman/listinfo/mp3encoder
