A speech fragment approach to localising multiple speakers in reverberant environments