ABSTRACT
Depending on the company, error calculation may be carried out by respeakers or by trainers. Once the spoken and the respoken text have been automatically aligned, a human operator goes through the text and decides whether or not the deletions have caused loss of information. Although a good starting point for this discussion is the so-called word error rate traditionally used in speech recognition, its use in respeaking calls for an adaptation to the particular reality of this technique, where for instance an omission that does not lose any meaning is usually not regarded as an error. Source and respoken text are included for each programme, along with the formula and a brief discussion of the results. In the respoken text, the elements highlighted in bold are those which have undergone some sort of change or modification.
