Add support for multichannel audio data

[iver56]

shape=(num_samples, num_channels) should be supported as well as shape=(num_samples,)

[francislata]

When is the expected delivery for this one @iver56?

[iver56]

Hi Francis, and thanks for your interest in this. I have no clear plan for this. I guess it's whenever I have time. If you want to speed this up, contributions are welcome!

By the way, yesterday I figured that shape=(num_channels, num_samples) is also sometimes something people use for multichannel audio. Should we support both?

[iver56]

Librosa (see librosa/librosa#1130) will use the convention (num_channels, num_samples). I think I prefer this.

The other dimension ordering, (num_samples, num_channels), is used when loading and saving multichannel wav, where stereo samples are interleaved.

[francislata]

@iver56 - I think sticking to one specific ordering like num_channels, num_samples just like in librosa would be much better.

[iver56]

Agreed.

I have started some basic work on multichannel audio support here: 59707fd

Contributions are welcome

[iver56]

Gain, PolarityInversion and PeakNormalization support multichannel audio now. I guess we can create separate issues for adding multichannel support to other transforms.