ref: 83d6d0c7dfa7c26bc3d1134fd18124c3acde8f08
dir: /TODO/
People are encouraged to pick some of these and implement it. Send all
patches to cbagwell@users.sourceforge.net, or post them on the patches
tracker at http://sf.net/projects/sox/.
SOX includes skeleton format files to assist you in supporting new
formats, sound effect loops, and special-purpose programs.
The full skeleton format, skel.c, helps you write a driver
for a new format which has data structures. skeleff.c is
a starting point for writing a sound effect loop. Sox.c is
a good starting point for new programs. (Someone finally
did this and told me what was wrong...)
In handlers.c, note that many formats set up the header and then
use the raw driver for reading and writing.
o Update all read routines to be like WAV handler and only
return data in multiples of sample_size*channels. This
will prevent corrupt files from causing sox to go into
an infinite loop trying to read just a couple more bytes.
o Make "dither" an alias of "mask" since that's how most
people refer to this process.
o Make "mix" an alias of "avg" since that's closer to what it
really is. For 2->2 mixes, make a 2 option L->L and R->R
shortcut. Similar 4 option for 4->4.
o Merging, e.g. two mono files to one stereo file.
o In avg effect, going from 1->2 should have a 2 number shortcut
so that you don't have to put 100% in each channel. Similar
for 2->4.
o Modify noiseprof and noisered so that they can read
and write from stdin/out. This would allow to run
them in scripts without intermediate files required.
o Several effects have to process a "window" of data at
one time and sox doesn't always pass that amount in.
Make some common routines to buffer up a window of data
instead of requiring each effect to reimplement this.
o Add an interface to libst so that applications can query
what size/encoding/rate/channels a given format handles.
Then applications could modify parameters before an error
occurs. This would allow stripping out override logic
in various formats.
o Effects engine should not have a global "flow" period
followed by a global "drain" period. Each effect should
track which it needs to invoke. Until that's done you
won't be able to stack two effects that both make use
of drain (ie. can't run the "reverse" effect twice
on the same command line and have a good output file).
o Make resample effects idempotent when rates are the same.
o Add MP3 support to WAV handler. Can copy most work from mp3.c.
o Implement speex (http://www.speex.org).
o Modify .dat handler to support stereo data... Probably should remove
time offset column since it's not real useful (could be added by
a simple post-processor). Also, should store information in
header related to original data so that it can be converted back
to similar percision easier. Another idea is to normalize the
data relative to original percision to reduce value ranges
(might result in a smaller text file since percision will
be smaller).
o Add command line options that specify a specific endian
(as apposed to -x which just reverses it from current
machines endian). This would be useful for cdr.c
since some formats like to keep the data in little-endien
(CDDA?).
o Add a FIFO/buffer to audio drivers to prevent losing audio data.
Perhaps mpg123 could be used as an example.
o Lots of effects are using "double" to prevent overflow of
st_sample_t math. With ststdint.h we now have a portable way
of detecting uint64_t... Use that when possible to speed up
math (won't have to convert back and forth from int to float).
o WAV handler is not using word alignment for at least the
main data chunk (expect for GSM).
o Keep sox from using "fail" on errors. Sox was supposed to be
a sound library called "ST" but libraries shouldn't exit a program,
they should return error codes for users to handle. Initial support
for this has been added. Needs to be completed.
o SJB: [11/25/99] (Re. resample effect): Note that upsampling
usually doesn't require interpolation, therefore is faster and
more accurate than downsampling. Downsampling by an integer factor
is also simple, since it just involves decimation if the input is
already lowpass-filtered to the output Nyquist freqency. Get the
idea? :) Also, the original algorithm has apparently been updated,
and might usefully be updated in SoX too.
o WAV handler needs to handle (ignore if nothing else) non-audio
chunks (except LIST which is already handled), especially at the
end of the file.
o All option-parsers should fail if extraneous charcters or arguments
are present. e.g. -v 1dB, vol 3mW, speed 2 3, should all fail.