@yingtai @meave

these data seem suspicious to me. everything I've seen so far had shown a small difference in the bimodal distribution by language/culture, measurable but by no means larger than the bimodal gap itself.

the first thing to keep in mind is that individual pitch variation is very big, and that most people who didn't sit down to study this topic have only a vague idea of what "pitch" sounds like—e.g. a large speaker with a testosterone-type larynx speaking in a shout or addressing a classroom will have quite a high pitch, yet people describe their voice as "low" because of the resonance. (many times a trans woman has told me "I don't want a high-pitched voice, I wanted to sound deep and butch like this or that celebrity"; I go study the celebrity's voice with trusty praat, she's invariably as high-pitched as your average cis woman.)

so you need to compare data from similar contexts and try to not deviate too much in the other parameters (amplitude, ambient noise etc). below is some data Mark Liberman plotted long ago from CallHome phone calls, in English in Japanese (https://languagelog.ldc.upenn.edu/nll/?p=5908 )

also while I'm quoting Liberman I want to link to these cool visualisations about how the bimodal distribution gets closer with age: https://languagelog.ldc.upenn.edu/nll/?p=31786