next up previous contents index
Next: Error measure presentation Up: Speech recognition systems Previous: Speech input and speech

Cut-through versus voice-stop (anticipation)

 

A major drawback of current speech technologies is the fact that when the system is prompting  messages (outgoing) it is unable to listen. That is, speech recognition and speech prompting  do not occur simultaneously. In most applications the caller has to wait for a system's instruction to be completed before speaking. Usually the users are instructed to ``please wait for the beep before speaking''.

Some technologies offer cut-through capabilities; these allow users to interrupt the system and anticipate on its messages. Speech recognition takes place even during the system prompts . This capability is highly appreciated by experienced users who may interrupt the system's outgoing speech any time. Moreover they may use a sequence of words instead of a single one. Of course it is necessary to carefully handle the dialogue structure in order to avoid interrupting the system because of noise  or unexpected words.

This capability has to be distinguished from the so called voice stop  where the system detects some speech activity  (energy level), stops prompting  its present message, and switches to the recognition mode expecting the user to say something. Usually the system instructs the user to say ``STOP'' whenever he is ready to input his choice and then wait for the beep before speaking. In that case the beep is still there and speech recognition does not occur simultaneously to the speech playback .

The application developer may use these capabilities or may not (when he wants the users to listen to a complete message for security reasons). So the technology provider should indicate whether this is optional, mandatory, or not implemented. The distinction may be defined as follows:

Voice stop:
  In a first step the system hears, it needs a second step to understand.
Cut-through:
The system hears and understands simultaneously (single step).

 



next up previous contents index
Next: Error measure presentation Up: Speech recognition systems Previous: Speech input and speech

EAGLES SWLG SoftEdition, May 1997. Get the book...