Choose MP3 bitrate when writing

[lminer]

trafficstars

Now that mp3 support has been added, it would be great to be able to choose the bitrate when writing to mp3. What is the default btw? 128?

[bastibe]

Does libsndfile have this feature? We can probably support it if there's an API for it.

[jimmieliu]

is bitrate now supported?

[bastibe]

Does libsndfile provide an API for it? If so, we can support it.

[sammlapp]

Couldn't find an API so I asked here: https://github.com/libsndfile/libsndfile/issues/1008

[sammlapp]

I don't really understand the syntax but over in that libsndfile thread 1008 evpobr linked to this documentation: https://github.com/libsndfile/libsndfile/blob/c81375f070f3c6764969a738eacded64f53a076e/docs/command.md#sfc_set_bitrate_mode

[bastibe]

I think that's just a flag on how the bitrate is to be interpreted. But how do we set the bitrate itself?

[sammlapp]

Hm, I don't know in that case

[sammlapp]

@bastibe more details provided by arthurt now https://github.com/libsndfile/libsndfile/issues/1008#issuecomment-2135784810

it seems like exposing the SFC_SET_COMPRESSION_LEVEL in the python-soundfile API would be useful. The argument takes values 0-1.

For VBR, it translates to lame_set_VBR_qaulity(). 0.0 -> 0.0 (highest quality), 1.0 -> 10.0 (lowest quality)

behavior is different for average and constant bitrate methods, see aurthurt's comment for details

[bastibe]

That sounds great! Would anyone want to prepare a pull request?

[sammlapp]

I gave it a try, but did not see any effect. I first added the constant defining SFC_SET_COMPRESSION_LEVEL to soundfile_build.py:

SFC_SET_COMPRESSION_LEVEL		= 0x1301,

then running pythin setup.py install, which made this constant visible in the soundfile._snd.SFC_SET_COMPRESSION_LEVEL API.

However, with the following code, the value assigned to pointer_compression_level has no effect on the file size produced:

pointer_compression_level = _ffi.new('float *')
pointer_compression_level[0] = 0.0
with soundfile.SoundFile('./compressed.mp3','w', 44100, 1, format='MP3') as sfc:
    _snd.sf_command(sfc._file, _snd.SFC_SET_COMPRESSION_LEVEL, pointer_compression_level, _ffi.sizeof(pointer_compression_level))
    sfc.write(data) # data is numpy array of audio samples

I also tried a different syntax for setting the float value:

pointer_compression_level = _ffi.cast("float *",0.5)

but the file sizes do not change with pointer_compression_level.

I don't have any grasp on how ffi really works so I'm not sure how to proceed. I feel it would be easiest for someone more familiar with the python-soundfile library to take it from here.

I also tried with flac format and saw no difference in file size.

Documentation: https://github.com/libsndfile/libsndfile/blob/c81375f070f3c6764969a738eacded64f53a076e/docs/command.md#sfc_set_compression_level

[bastibe]

Thank you for looking into this. I think the pointer_compression_level should be a double, not float. What do you get as return value from _snd.sf_command? It should equal SF_TRUE or SF_FALSE.

Also, check if your version of libsndfile is new enough to support the SFC_SET_COMPRESSION_LEVEL API.

[sammlapp]

I tried 'double', no change. The returned value is 0 (not SF_TRUE or SF_FALSE). I also updated libsndfile to 1.2.2 with brew install libsndfile, no difference.

[bastibe]

If you installed the wheel, you'll need to remove _soundfile_data from your site-packages before it'll pick up the system libsndfile. Or uninstall the wheel and install the source distribution, which does not come with _soundfile_data.