Question about praat voice report

317 views
Skip to first unread message

Dan Weber

unread,
Oct 25, 2020, 7:46:04 AM10/25/20
to Parselmouth
Some context:  I am doing a college project for a machine learning class.  I am trying to analyze several thousand short audio clips.  The audio is of supreme court testimony.  It contains only one voice at a time and no background noise, very vanilla.  I'm planning to use the praat voice report to extract features.

I understand that this call makes the voice report:

voice_report_str = parselmouth.praat.call([sound, pitch, pulses], "Voice report", 0.0, 0.0, 75, 600, 1.3, 1.6, 0.03, 0.45)

What are all those parameters?  Are they sort of standard, or should I calculate them somehow for each file?

Thanks!

yannick...@gmail.com

unread,
Oct 25, 2020, 8:01:45 AM10/25/20
to Parselmouth
Hi Dan

Sounds like a very interesting project, indeed!

These parameters are the ones that Praat expects when using the GUI (since this functionality wasn't yet converted as Python method in Parselmouth). So, they are the start time, end time, minimum pitch (Hz), maximum pitch (Hz), maximum period factor, maximum amplitude factor, silence threshold, and voicing threshold. The values for these you've listed are the default values that Praat uses, so should be at the very least a decent starting point for human speech.
The Praat documentation has some further explanation about the voice report: https://www.fon.hum.uva.nl/praat/manual/Voice.html

Meanwhile, do beware that the result will be a long, multiline string. If you want to extract the specific values, you can query them individually, like I've hinted here: https://stackoverflow.com/questions/45237091/how-to-automate-voice-reports-for-praat

If you have further questions, don't hesitate to get back to me! :-)

Kind regards
Yannick

Op zondag 25 oktober 2020 om 12:46:04 UTC+1 schreef Dan Weber:

Yannick Jadoul

unread,
Oct 25, 2020, 8:06:18 AM10/25/20
to parse...@googlegroups.com
Small follow-up:


> Meanwhile, do beware that the result will be a long, multiline string. If you want to extract the specific values, you can query them individually, like I've hinted here: https://stackoverflow.com/questions/45237091/how-to-automate-voice-reports-for-praat

This answer on StackOverflow might be even more useful, if you want to get floating-point values out of Parselmouth instead of the long string: https://stackoverflow.com/a/55209266/2043407


Kind regards
Yannick
--
You received this message because you are subscribed to the Google Groups "Parselmouth" group.
To unsubscribe from this group and stop receiving emails from it, send an email to parselmouth...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/parselmouth/28840824-8984-4ebb-b8bb-740c5990b67bn%40googlegroups.com.

Reply all
Reply to author
Forward
0 new messages