[Flac-dev] pcspkr wave encoding

[Jan Engelhardt]

Hi,


there is an interesting case when the FLAC encoder (using 1.2.0) is
given simple waves. Simple waves means: I have a list of {frequency,
duration, pause} tuples that define the monophonic tune. In other
words, exactly one frequency is played at a time.

This is the original dataset from 1989 (driving a PC speaker back
then):

$ ls -l ihold.sd
-rw-r--r-- 1 jengelh users 20616 Aug 14 00:57 ihold.sd

Since driving the PC speaker is mostly a privileged operation these
days, one would just convert it to PCM and use soundcard output. The
program "sd" I have written does just that. The two numerical
parameters a and b denote the square and sine rate.

# 0:1 rate <=> 0% square, 100% sine generation
$ ./sd ihold.sd 0 1 | sox -r48000 -s -w -t raw - ihold-sine.wav

# 1:2 rate <=> 33% square, 66% sine generation
$ ./sd ihold.sd 1 2 | sox -r48000 -s -w -t raw - ihold-square1-sine2.wav

# 1:0 rate <=> 100% square, 0% sine generation
$ ./sd ihold.sd 1 0 | sox -r48000 -s -w -t raw - ihold-square.wav

Now this is what happens when encoding to FLAC:

$ flac -8 ihold-sine.wav 
ihold-sine.wav: wrote 2422225 bytes, ratio=0.353

$ flac -8 ihold-sine2-square1.wav
ihold-sine2-square1.wav: wrote 5474921 bytes, ratio=0.798

$ flac -8 ihold-square.wav 
ihold-square.wav: wrote 5573016 bytes, ratio=0.813

As soon as square waves are mixed in, the FLAC compression ratio
drops rapidly up to the point where it equals the ratio of typical
day-to-day music.

While I do not request FLAC to compress the wave back to its original
20KB, I think it could improve on the sine and square compression.
Are there any plans to address this?

If you would like to have the data files and/or transformation
program for testing, let me know and I'll upload :)


thanks,
	Jan
--