We are using Praat to obtain min max and mean pitch from emerging adult narratives about relationship events. We have used the software for many years but lately noted more max scores above 300 (mostly in female participants). We have the range of scores set from 70-300 and when we get a score at or above 315 we check the recording for the places where the score is out of range to make sure there are no non-human sounds but just the participant speaking. In all but 1 instance the recording segment with the high scores were in fact human speech. We welcome any feedback on why this may be happening and what to do. Thanks!
Candice Feiring
Dear Candice,
I am not sure I completely understand your problem. But my suggestion would be to try different settings, such as increasing the high limit of your frequency range and see what happens. Another possibility would be to use a different program/algorithm to obtain the pitch. I know the Swipe' algorithm developed by Camacho et al. is very robust. You can find the Matlab script in his dissertation Article A sawtooth waveform inspired pitch estimator for speech and music
.Hugo
Dear Hugo- thanks for the reply. The human voice is supposed to range from 75-300HZ so that is why we set the PRAAT parameter to range between this minimum and maximum. So we don't want to raise the high limit frequency- although I do need to do some research on whether the human voice can/does reach pitch higher than 300.
Best,
Candice
Well I suppose it IS possible that in speech the fundamental frequency goes beyond 300 Hz, especially in women. If you are having several occurrences of pitch going above 300 Hz, then it is likely that it does. My point about using a different algorithm is to check that it is really the "right" value. When this happens, you could also try to listen to the pitch and try to find the corresponding note by hear, and then find the associated frequency. If it sounds like 300 Hz then it is.
The human voice does go above 300 Hz in singing, and to be honest even in speech, this value does not seem so high to me. I would say the typical value for women is, in average, around 210 Hz, which means that 300 Hz would only be a fifth above, which sounds completely attainable to me. Therefore, I do believe that you should increase the range.
I suppose the question is mostly about what you are trying to measure. I see that you are recording people narrating relationship events, is that correct? If that is the case, I would expect those to be rather emotional, therefore you might expect a larger range of pitch - that is my assumption.
So, definitely, I would personally increase the pitch range, because you want to catch all the possibilities.
The range setting in PRAAT is mostly to increase the accuracy of the detection - the narrower the range, the less likely you will obtain frequency doubling or halving, for example. If some parts of your audio ARE above 300 Hz, then PRAAT will probably give you values that are just above (315 Hz is a good example), or will probably half the "real" pitch.
Hi Hugo-
I really appreciate all this useful information! And thanks for looking at our work- indeed narratives about negative relationship events might be more arousing- than narrating happy events- in fact I have some students looking at pitch in narrations in the same partner about met needs (positive events) and unmet needs (negative events). I don't have comparison data for our couples of narrative arousal vs interaction with partner arousal around the same topic---- always another study that can be done.
Before I read you reply I had gone back to review some of the papers on couple conflict interactions that used pitch as a variable to index emotional arousal and Praat to measure pitch. I found that the scores reported varied widely and the range of pitch when reported went well above 300 for women.
Taken together I am now less concerned about the scores we are getting above 300-- and we have been checking the audio files to make sure the higher scores are indeed the human voice and they have been.
You suggested that we set the max higher-- although it does seem that if the voice goes above 300HZ the program does yield values -- so far our narrative date as high as 318.
If you have any reference about female voice Fo range in singing or otherwise that would be great-- or perhaps suggest search terms for me to use to find such papers?
Thanks again for taking the time to share your expertise with me and my research group.
Candice
- Badges
- Science topic
- Similar topics
- Filing