Modified Least-to-Most Greedy Algorithm to Search a Minimum Sentence Set

Searching minimum sentence sets is one of the key issues in building continuous speech corpora for speech study. The minimum sentence set should contain smallest number of phonemes but cover all desired phonetic units. This paper introduces two modification procedures to improve performance of least-to-most (LTM) greedy algorithm in searching a minimum sentence set. The first procedure groups sentences containing the to-be-covered units having the same frequencies into a subset, and then selects a best sentence from that subset. The second procedure optimizes the resulting set by removing redundant sentences. Testing to an Indonesian mother sentence set showed that modified LTM greedy algorithm produced slightly smaller sentence set than that generated by LTM greedy