Assuming that most word pairs have min(word_len) = 5, there are ~5x2x2 = 20 db called per word pair
Could drop this to 2 calls (one to GraphemeProbs, one to PhonemeProbs) by querying before the loop starts, and caching the results. Expect this to yield substantial speedup overall, 2-4x of total time
Assuming that most word pairs have min(word_len) = 5, there are ~5x2x2 = 20 db called per word pair
Could drop this to 2 calls (one to GraphemeProbs, one to PhonemeProbs) by querying before the loop starts, and caching the results. Expect this to yield substantial speedup overall, 2-4x of total time