78 Sentencenes from SWA recorded, five reps of each. Took ~one hour
Extracted words and syllables by dumpfeats-xwaves
Rebuilt readWavAndWord so it shows both words and syllables
Inspection shows autolabelling by dtw so-so, let's try ehmm.
Autolabelling by ehmm started 11.58 fin 12.49 ehmm actually worse
Combining swa with wpB set (=957 sentences), testing ehmm start 11:23 fin 15:46 this is better!
Now trying PENTAtrainer. Seems to be extremely sensitive to where the boundary is. Target may change from strongly rising to almost equally strongly falling by moving the boundary just a few millisecs - this shouldn't happen.
A relevant pdf.
Here's how I read in targets.txt in R.
enca states that the file is:
Universal character set 2 bytes; UCS-2; BMP
LF line terminators
so first we convert it with:
enca -L none targets.txt -x utf-8
Then open R and read in with:
Apossibly useful command line might then be:
b<-a$height.7 ; shapiro.test (b) ; plot(density(b))