Selecting the winner

Selecting the best track can (and maybe should) involve more information than just what is contained in the analysis itself. For example, information about the speaker, the language, and the linguistic content of the token can all help guide an analysis.

At the moment, FastTrack focuses on selecting the smoothest tracks by trying to find the formants that can be predicted with the least error:

1. The difference between observed and predicted formant track is found, resulting in residuals.

2. The median absolute residual is found for each formant, and then the sum of these values is found.

The analysis with the minimum error is selected as the winner.

In the future, a wider variety of selection algorithms will be introduced.