Robin Mukherjee
The aim of this project was design and implement a system to classify incoming email using non-speech audio, in this case abstract sounds, known as earcons. Such a system was intended for use by any regular email user, to allow them to more efficiently manage their communication.
The earcons were designed to classify an email under the categories of 'Sender', 'Length' and 'Urgency'. They were implemented on E Z Vision for the Apple Macintosh.
An algorithm was designed and implemented in C to perform the classification of an email in terms of three chosen categories. The program used a combination keyword searchers and frequency analysis and comparisons of the email with user-defined data files, for classification.
The earcons were tested in two stages with participants of varying musical abilities. Results indicated that the earcons had a real life application, could be remembered and that musical ability had a significant effect for the particular design and tests carried out.
An automated system, which generates the earcons from the results of the classification program could be implemented by programming a Soundblaster card. Future work could extend the testing of the design to a larger sample to obtain more significant results. Additionally, the design itself could be extended to classify more categories of an email, possibly using parallel earcons.
Expansion of the classification algorithm to include term distribution within a document as well as consideration of word/phrase context and/or natural language usage, also warrants further study.