Architecture of speech synthesisers