No matter how much the performance of speech recognisers improves, it is unlikely that perfect recognition will be possible in all circumstances as environmental sounds interfere with recognition. In such circumstances it is necessary to provide feedback so that errors may be detected and corrected. In some situations, such as over the telephone, the feedback must be provided auditorily. The question arises as to whether this feedback should be provided after each word or after a group of words. It is shown that in the case of spoken digits this depends on the accuracy of the recogniser and on the times required for recognising the digits and for changing from recognition to synthesis mode.