better prediction
of prosody:
bigger databases give more data for training
better labels give finer `cuts' on the data
of voice quality:
acoustic measures of `type of voice'
adding a new tier of labelling information
