Recording concatenation costs of most common acoustic unit sequential pairs to a concatenation cost database for speech synthesis