else:self.o=os.popen("faac -b 32 -P%s -C 1 -o \"%s\" -"%(cond(big_endian,""," -X"),cond(write_to_stdout,"-",outputFile)),popenWB)# (TODO check that faac on big-endian needs the -X removed when we're giving it little-endian. It SHOULD if the compile is endian-dependent.)
elifout_type=="mp3":self.o=os.popen("lame -r%s%s -m m --vbr-new -V 9 - \"%s\""%(lame_endian_parameters(),lame_quiet(),cond(write_to_stdout,"-",outputFile)),popenWB)# (TODO check that old versions of lame won't complain about the --vbr-new switch. And some very old hardware players may insist on MPEG-1 rather than MPEG-2, which would need different parameters)
# Older versions of gradint used BladeEnc, with these settings: "BladeEnc -br 48 -mono -rawmono STDIN \"%s\"", but lame gives much smaller files (e.g. 3.1M instead of 11M) - it handles the silences more efficiently for a start).
# Size regression would probably also happen if we used shineenc or comprec: fixed-point, so likely faster than floating-point lame on low-end ARM etc, but won't reduce bitrate for silences.
# Anyway, on a Raspberry 1 B+, oggenc runs ~5x faster than lame (probably due to the type of floating-point in use), and speexenc 10-15x faster than lame, and both produce smaller files. So if MP3-compatibility is not required (and you're not concerned about further quality loss of any samples already stored in mp3 format) then ogg or spx would be a better choice.
# Typical file sizes for a 30-minute lesson: OGG 2.7M, neroAacEnc 3.0M at 32000 (you might be able to put up with 1.8M at 18000 or 2.2M at 24000), MP3 3.1M, MP2 3.4M, faac 3.7M, WAV 152M
# TODO try AAC+? aacplusenc wavfile(or -) aacfile kbits, 10,12,14,18,20,24,32,40 (or 48 for stereo), but will need a player to test it
# (mp2 could possibly be made a bit smaller by decreasing the -5, but don't make it as low as -10)
