I don't think the best use case for BK-trees is spell-checking and words. The area in which they are used most successfully is image deduplication. In that case

I don't think the best use case for BK-trees is spell-checking and words. The area in which they are used most successfully is image deduplication. In that case the metric you're going to use is some form of perceptual hashing.