Response time for a vxml application

Question

I'm developing a voicexml application, and I want to define a part of the performance by defining the maximum response time for an answer by the application.

The application is on a server realized by two virtual boxes and you can call it due to an ISDN connection.

I'm looking for some things:

-Are there any "scientific sources" that descripe the length of a "normal" response time in a dialog between two people or a voice application and a human?

-->Which response time can I assume? (for example when the user choose "option1" after two seconds the application says "you chose option1" - but what time is realistic, are for example two seconds too long? so the user "feels" that this is not a "real" communication)

(- Are there any speacial delays by calling an application from the telephone network? )

thanks in advance

Answer 1

I haven't seen any studies that show this information, but I would expect it to be highly contextual. Simple responses to simple questions would be quick, but complex and long answers would be wrong. And, from a human perspective, there's all sorts of emotional responses and sociological behaviors that affect the speed of a person's response.

If I understand your goal correctly, you want to specify a maximum response time for the IVR (VoiceXML system) to respond back to a caller.

Even this can be contextual. If the machine needs to look up some data, it might take longer. In systems I've built, if it's more than 2 seconds, we've played a please wait or other transitional message.

In practice, systems are fairly responsive. If they don't seem reasonably responsive, you have a problem or some other artifact in place. And, with machines, people expect a little bit more of a delay than a normal human operator (with humans, there is also a lot of non-verbal noise that lets the user know their input is being accepted, like keyboard sounds).

As for delays on the phone network, not so much any more. But, some international calls or weird routing can still introduce some unnatural delays.

To be more specific, 2 seconds is too long. If you know you have server delays add in some sort of audio queue to let people know data has been taken. I've seen a few a few shops add cute (for the first 50 times) processing sound to let users know their speech was recognized.