The following abstract is from a paper presented at the World Congress of Performance Analysis of Sport, August 2006, Hungary. It provides an example of using speech recognition with Focus, a technology which has now been fully integrated into the software to give the Focus Voice Control product.
It is 18 years since the seminal work of Taylor & Hughes (1988) first indicated voice activated data entry could be of major benefit to the sport science community. Due to a paucity of empirical research to expand upon these findings the performance analysis community still considers the introduction of voice entry methods as being the next big step forward. The specific aim of this study was to evaluate the accuracy and reliability of voice activated data entry methods for coding Coca Cola Championship Football League Matches. Voice recognition software integrated with computerised digital video sport analysis software (Focus X2, Elite Sports Analysis, Scotland), was used to conduct live and post match analysis of 12 football matches. For each individual match analysed a single category set containing 39 codes was created. To activate the codes and therefore generate a time log of events pre-trained commands were spoken into a head worn microphone and processed using a laptop computer. To determine the accuracy of the voice activated data entry method commands spoken throughout matches were simultaneously recorded using a digital voice recorder. For live and post match conditions the absolute and relative accuracy of voice activated coding was examined by sequentially comparing items in the event lists exported from Focus X2, against each spoken command recorded on the digital voice recorder.
Event lists from the live and post match conditions ranged in length between 1146-1522 and 1159-1577 items, and the number of errors ranged between 0-4 and 0-5 respectively. Relative accuracy scores for both conditions ranged between 99.7%-100% (median 99.8%). Differences in the magnitude of error between the live and post match conditions were not found (z = 0.23, p>.05, Mann Whitney U). Following inspection of the event lists to correct errors coded during each condition percentage difference calculations were used to provide an indication of intra-operator reliability. Empirically acceptable levels of reliability could readily be obtained with percentage difference values ranging between 4.6% and 1.9% (median 3.2%). Combined with inspection of the original video footage a sequential item by item analysis of the live and post match event lists was undertaken. This enabled the generation of a true data set from which the level of error incurred during each individual live and post match analysis could be determined. Results indicated that although the level of operator error was acceptable for both conditions there was a significantly greater level of error in the live as opposed to the post match condition (z = 2.67, p<.01 Wilcoxon Signed Pairs).
To conclude, it has been shown that voice activated data entry is an accurate method of coding events using computerised digital video sport analysis software. In consideration of the low levels of operator error incurred during both live and post match conditions voice activated methods of data entry can no longer be neglected by the performance analysis community. The current data set support the adoption of a methodology of data entry that enables:
Future research is required to determine further applications of this technology both within performance analysis and the wider disciplines of sport science.