orianakc/dissertation

Source files for dissertation-related scripts

dissertation's Issues

Matching phones to words/syllables in extract-switchboard-stp.py

The STP files have on each line the end time, the number 121 (?), and a phone/syllable/word.

Currently, extract-switchboard-stp.py matches the phone with its syllable or word with this code:

for s in speaker.wrd:
                if s['time']>=i['time'] and speaker.wrd[speaker.wrd.index(s)-1]['time']<i['time']:
                    valDict['wrd'] = s['value']
                    valDict['wrdTime'] = s['time']

Sometimes the times seem to be off, in that the end time for a word (e.g. "THAT" at 4.401200 from file 2151-A-0003.wrd) is earlier than the end time given for the phone or syllable (4.401687 for the corresponding [t] in 2151-A-0003.phn and 4.401690 for the syllable [dh ix t] 2151-A-0003.syl). This means that the script matches the following (wrong) word.

Matching phone with the syllable that contains it

Syllable tags have a nite:child attribute which lists the range of phones associated with that syllable:

<nite:child href="sw4168.A.phones.xml#id(ms6A_ph1)..id(ms6A_ph3)"/>

How to efficiently match, e.g. phone ms6A_ph2 with the syllable above?

Check if nite:id for phones is a reliable indicator of word boundaries

[q] and [dx] not from /t/

(reposting something here that Oriana sent in an email)

Also -- some glottal stops and flaps come from sources other than /t/, of course. 
What do you think would be a good way to separate them out? 
Maybe use orthography to get underlying form from a pronunciation dictionary? 
Then I would really need to figure out the issue with word/phone alignment.