A first step into speaker-based indexing