User perception of European network tones
Tones are used in telephone networks to signal various information to callers, both during call setup (dial tone, ringing, engaged, number nobtainable, etc) and during the talk phase of a call (e.g. call waiting and pay tones). These tones vary from one country or network to another, which can cause confusion. The European Telecommunications Standards Institute (ETSI) has been reviewing the functions and acoustic characteristics of existing and proposed tones, so as to devise guidelines for new tones and for possible future harmonisation across networks.
As part of the Dialogues 2000 project at CCIR, UK users' perceptions of a number of network tones were measured on semantic differential scales such as "pleasant - unpleasant" and "interesting - boring". A two-dimensional characterisation of tones emerged, in which the first dimension could be identified with the "OK - not OK" dimension in the model previously proposed by ETSI to relate tones' functions to their frequencies and cadences, and the second dimension corresponded closely to the "interesting - boring" scale.
Confirmation of spoken details in automated telephone dialogues
In many emerging applications of spoken telephone dialogue technology, the caller has to supply several details to define the desired transaction - such as the departure point, destination and date of a journey. The automated service should read back the recognised details to the user for confirmation before proceeding with the transaction. This should be done in a way which ensures detection and correction of any errors, while keeping the interaction as fluent and natural as possible in the case where the input has been recognised correctly.
A recent experiment at CCIR compared three styles of confirmation messages, differing in their wording and intonation, in the context of a simulated flight information service, in the cases with and without speech recognition errors. It was found to be important to use question rather than statement intonation to ensure a response from the user where an error had been made. It was also found that most users responded explicitly (with "yes" or a synonym), rather than remaining silent, when the recognised details were correct. Implications of the results for automated service design will be discussed.