Exploiting contextual information for improved phoneme recognition