I think first you need to define "pronunciation complexity." Is it a measure of the amount of time it takes the average L2 learner to correctly pronounce the word? Or, a measure of the diversity of phonemes in the word. Operationalizing your variable will help immensely in clearing up your goal.
Suppose you want to measure complexity by the diversity of phonemes in the word. You could then find a dataset of phonemes for each English (or other language) word. Then you could do something like, calculate the entropy of each phoneme set per word. Or maybe the Shannon diversity (although this might not perform well for sets with very low specie occurrences). You could also vectorize each word and perform some measure of similarity between words.
Of course, this is all hypothetical. Once you update your question with a clear goal in mind it will be easier to help you more concretely!