Determine how to align SAA text with phonetics

[cimbriano]

Entire transcriptions in string form can be split into words, (see example in make-test-dataset.rb)
If the size of the split isn't the same as the number of words either:

• There's a bug in the splitting
• An utterance was produced twice and transcribed
• The speaker skipped a word
• Error in transcription or in spacing to signify word boundaries

[cimbriano]

One option is to only use the transcriptions that match in length to the number of words in the "Please call Stella" paragraph. split-transcription.rb prints the matching and mismatching transcription filenames