Text this: Statistic model of speech-like signal generation