I am analyzing dyadic conversations for speech characteristics through pre-written script. I spoke with an expert in speech analysis prior to data collection as an introduction to praat. He suggested a few parameters I could follow regarding pitch, silence duration, and silence threshold but did not really get into the analysis or the data sets. I have looked at the standard settings given in the script, which are just slightly different than the expert suggested. What is the benefit of adjusting such parameters in one direction or the other? I need to be able to justify whichever parameters I use, which could easily follow the script, as it had been published. However, the expert is familiar with this specific scenario/dataset. Thus, what are the benefits of adjusting and how would one justify doing so with these parameters?

Script parameters: 50Hz pitch, -25dB silence threshold, 0.3 silence duration Suggested parameters: 150 Hz pitch, -30db silence threshold, 0.2-0.25 silence duration

Many thanks.


Those answers are given by Praat documention:

Silence threshold (dB): determines the maximum silence intensity value in dB with respect to the maximum intensity. For example, if imax is the maximum intensity in dB then the maximum silence intensity is calculated as imax - silenceThreshold; intervals with an intensity smaller than this value are considered as silent intervals.

Pitch floor (Hz) (standard value: 75 Hz): candidates below this frequency will not be recruited. This parameter determines the length of the analysis window: it will be 3 longest periods long, i.e., if the pitch floor is 75 Hz, the window will be 3/75 = 0.04 seconds long.

Minimum silent interval duration (s): determines the minimum duration for an interval to be considered as silent. If you don't want the closure for a plosive to count as silent then use a large enough value.

Besides, other parameters may be added :

http://www.fon.hum.uva.nl/praat/manual/FAQ__Pitch_analysis.html http://www.fon.hum.uva.nl/praat/manual/Intensity__To_TextGrid__silences____.html http://www.fon.hum.uva.nl/praat/manual/Sound__To_Pitch___.html


It sounds like you want to separate a recording into interleaved sub-parts: things said by A, and things said by B. TextGrid_to_silences can separate "utterances" from silence, but doesn't tell you who said which part. Depending on the voice characteristics of the speakers, it may not be possible to do this with a script, you may have to manually encode speakers. Amegnunsen explained the relevance of the settings for distinguishing speech from silence, so it might be possible to parse the talking into a series of utterances, but determining who said what is probably impractically hard. The issue that is most likely to cause problems is that there may be very little silence between turns, so a minimum silence interval of .25 may merge different turns into one chunk.

Theoretically, you could compare the parses obtained from different settings, if one speaker has a higher-pitched or louder voice, where one set of parameters gets all of the utterances, and a different set of parameters gets a (substantially smaller) subset of the utterances, because speaker B's utterances are treated as "silence" (below the pitch or amplitude threshold). But I think it's more likely that you will just set incorrect parsing into utterances.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.