6. Conclusions

A new method for word synthesis is proposed that simply concatenates syllabic components selected from a database. Selection is based on requested mora position and mora length, with alignment of pre- and post- phonemes. An experiment with a small word set confirms that this new synthesis method gives sufficient voice quality while drastically reducing the recording cost by decreasing the number of words that must be recorded by 84%. This is especially effective when a huge number of words are needed to fill the slots of the templates. Our next goal is to reduce the word set even further by grouping syllabic components according to their prior and posterior phonemes. For example, two kind of components and are treated as the same component as both are plosives. This should the number of words to be recorded by a further 64%.