While analysing pitches with Praat, I'm often faced with the problem, which is, I must modify the F0 data manually since there are always some octave up points or other points which is impossible to be there, as well as some impossible silence.
However this kind of modification seems to be casual, therefore I wonder if there are any operation standards for me to guide the modification so as to make this kind of operation more rigorous and scientific.
UPDATE:
I'm sorry for my poor description, and there are some pictures:
There are four common conditions which I always met:
Condition A & D are the impossibly high or low F0 points.
Condition C is a short series of F0 points which should be there but are treated as silence by the Praat algorithm.
Condition B is more complicate, which is, there isn't a falling part at the end of the syllable in terms of transcription and my own perception, but there is according to Praat (therefore I don't know if I should delete this part or leave it there). I know the falling part at the end of a syllable as well as the rising part at the start may be account of the physiological reason and always can't be noticed by the hearers, but I don't know if I should hold these kind of features in the picture of my report.
I don't know how to treat these kind of conditions. Maybe I should ignore the bad-condition syllables but these conditions are common. To modify them manually is just, however, not of science, I think.
Voiced / unvoiced cost
parameter, and all those annoying gaps disappear. As for condition B, I'm agree with you and just leave it there! BTW, why don't you put your comments in an answer therefore I can accept it.