Quality management
The evaluation of a speech application is an important component in the process of development. During the evaluation and in agreement with our costumers, dialog design, and implementation are constantly improved.
Efficient testing criter
Structure tests
During the structure tests, dialog syntax as a formal verification as well as the structure of the menu and the dialog flow are checked. For this purpose emulators were developed by atip, which allow confronting the dialog with all possible events and recognitions. Thereby all dialog conditions and all imaginable paths through the dialog can be simulated and tested under real conditions.
Load test
In the context of load tests, it will be endures that the dialog as well as the connected backend-system and databases work as expected even under extreme conditions. A great number of parallel calls can be simulated by a software-system which is unlikely to occur in reality. Especially the functionality of critical components such as backend-systems is warranted even for unlikely call-loads.
Offline recognition tests
The performance of the speech-recognition-software plays the most important role in dialog systems. Through offline recognition tests, which can already be done before the friendly user tests, the correct interaction between recognition and grammars can be assured. The reliable recognition of speech is of great importance, for example with dialectal pronunciation.
Friendly user test
Apart from recognition tests for the isolated revision of recognition performance friendly user tests evaluate the handling and acceptance with selected end-users under laboratory conditions. Since a user friendly dialog-flow and an efficient dialog-process are the main design criteria, the method of the friendly user tests is of great importance.
Pilot study / post tuning
The completion of the evaluation constitutes of a pilot study in which an adequate number of users is asked to test the application under real conditions. In the live service this pilot study is undertaken shortly after launching the application. Important criteria for the handling, like actual use and motivation of the caller, can only be raised by live-data and not only under laboratory conditions. | 