Oliver Adams

Nothing has changed. Empty means there is no actual WAV data. It's just a header with duration 0 WAV. Another related thing would be to skip utterances where the number...

That would be good but I agree with your labelling this low-priority. Calling sox for now is fine.

Sure, since it's pretty orthogonal to everything else.

There's no data: the distinction between `filterbank` and `filterbank_and_pitches` is only that pitch features are extracted from the audio additionally, so a test for pitch extraction could use the same...

Yes, that's right.

Yeah, perhaps we should just enumerate current documentation change requirements in the checklist above, unless its really big enough and requires implementation that somehow warrants its own issue.

Okay, I guess the hashability isn't a big deal, but perhaps if the classes hash function was based on the state the original data is in (among other things), then...

Yes, that's right. The idea I had was to make Corpora immutable and the API should be consistent with this behaviour. (However, I'm open to counter-arguments to the underlying design...

The way the diamond is handled by that above code is to treat it the same way as the vertical bar is treated: remove it if we are not predicting...

As for the OP: I don't think it's directly related to #157. The reason the parameters are different is that the Na data has some features that may not be...