Good things to do in voice applications

Correct wording of your prompts is particularly critical when you are using speech recognition. Your prompts must tell the caller what to say, when to say it, and how to say it. Here are several considerations for wording prompts:

• Voice output can be speeded up with no loss of intelligibility. This could be under the caller’s control.
• Subdivide the recording of voice output, to allow random access to it. Callers can skip and scan.
• Make sure that callers know they’re using a machine; don’t try to make them think they’re talking to a human agent.
• Ask the caller whether they are using a phone that generates tones, by asking them to press a specific key.
• Allow the caller to interrupt prompts (key ahead) wherever possible (that is, don’t force play the prompts).
• Refer to the # key as “pound” in the U.S. or “hash” in the U.K.
• Refer to the * key as “star”.
• Use the star key for the control menu.
• Refer to the 0 key as “zero”.
• Use the zero key to provide access to the operator.
• Always phrase your prompts so that the goal precedes the means of achieving it (for example, “to contact the operator, press 0”); if you mention the key first, the caller may forget which one it was before realizing that it was they one they wanted.
• Use questions and commands rather than statements, to encourage callers to take immediate action. For example:
  • To check your account balance, press 1.
  • Is this correct? Press 1 for yes, 2 for no.
• Ask closed questions such as “Do you want a large, medium, or small pizza?” rather than “What size pizza would you like?”.
• Add pauses to encourage the caller to take immediate action.
• Always try to allocate similar functions to the same keys.
• Use directional metaphors where appropriate (use the relative position of the keys on the keypad to indicate some logical direction associated with the command, for example, 7 for back, 8 for pause, and 9 for forward).
• Try to limit menus to about four options, with a fifth option (“for more choices”) if necessary; this adheres to the “magic number 7 plus or minus 2” items that people can hold in short-term memory; also it won’t take too long to play.
• The order in which you play the options depends on the application: you might choose to start with most frequently used, or least-frequently used, or play them in ascending numerical sequence.
• Use simple, explicit language, for example: “Press” for single key entry (no delimiter) and “Enter” for multiple keys (which need a delimiter).
• Give feedback: repeat long data entries back to the caller.
• If something’s going to take a long time, tell the caller and then play background music; repeat the message saying it’s going to take a long time at intervals.
• If the caller does not make a selection, repeat the menu.
• If the caller makes an error, explain valid choices.
• Employ a professional to record the final version of the prompts.
• Don’t mix prompts recorded in more than one voice, unless you do this consistently to convey information (for example, voice A for menus and voice B for data retrieved for playback to the caller). For this reason, you should get the system voice segments recorded by the same professional as your other voice segments.