Submitted to: Applied Statistics In Agriculture Conference Proceedings
Publication Type: Proceedings
Publication Acceptance Date: 6/29/1997
Publication Date: N/A
Interpretive Summary: In this article, we discussed the development of COTTONTALK as an example of how speech recognition capabilities, achievable within the foreseeable future, can be used to obtain agricultural data that involve eye and hands busy' tasks. Also discussed are the use of traditional statistical methods for helping to design the software that performs the speech recognition task. The use of several experiments have been found necessary to identify strengths and weaknesses of sample prototypes designed for speech recognition tasks relevant to cotton plant mapping. It is expected that other applications involving the acquisition of agricultural information will have design problems similar to those described for plant mapping. Therefore, it is necessary to continue the use of prototypes and to evaluate their performance with clients familiar with specific tasks. Initially, these simple prototypes can be easily evaluated by watching how they perform, or by asking simple questions of the user. The information learned from a simple example can be employed to develop a more sophisticated and robust prototype. However, as the prototypes develop into a finished product, it becomes necessary to apply more rigorous criteria for assessing performance. Speech interface software application can profit from employing traditional statistical methods during their development.
Technical Abstract: It will be argued that customary software design strategies, by themselves, fall short when designing speech recognition applications. Concepts of experimental design and analysis are also necessary for developing speech interface software. This study demonstrates that these tools can be advantageous to the software developer, especially if the prototype methodology model of software development is applied. A case study for the problem of developing a speech interface for collecting, or mapping, information on cotton plant growth is presented. The acquisition of cotton plant map data is a 'hands and eyes' busy task that requires considerable investment to record and convert hand-written data sheets into computer data files. The project goal is to develop software that converts spoken key words and phrases describing a cotton plant into text 'strings' that are subsequently manipulated into a computer ready data file.