Oliver Adams

Thanks for the thoughts. Let's leave the preprocessing of diamonds as is for now then.

Thanks for these clarifications. The good news is that because those nasal vowels were in the BI_PHNS set, that takes precedence and they were always treated as such.

Currently GPU usage doesn't work when running experiments via tox because setup.py doesn't specify tensorflow-gpu. However, if tensorflow-gpu is installed but CUDA etc is not installed, then there will be...

Yeah, agreed.

When you do this. Ignore for now run.py, results.py, and lattice.py and of course anything that is already covered by tests (see the coveralls page, as well as what is...

Yeah, this is key. I'll roll this in with a more general explanation of Corpus objects, CorpusReaders and Models so people have more flexibility in expressing the experimental configuration.

Hi Niko, great to see your interest! The Komi-Zyrian documentation project looks promising. What sort of speech have you collected? @fauxneticien @nikopartanen I've written a bit of an explanation on...

> All in all we have around 35 hours of transcribed segments. Oh wow, that's quite a lot of data! (by my standards) > This can be converted quite easily...

> I'm just thinking that the transcription of these doesn't really correspond to normal phoneme sequences in the language, so it may be better if those are not seen during...

> This works very well! It overwrites now the log file, and maybe it could be nicer if it would continue writing into it, just to keep track of how...