We have presented a small-footprint speech synthesizer for Egyptian Arabic that runs on a PDA with a size of 9MB. The synthesizer has been evaluated with standard word-level and sentence-level tests with results comparable to those achieved for English voices of similar size and level of development (a few months to build a new-language voice from scratch).
We have discussed some of the challenges encountered, describing our solutions to them. Challenges include vowelization of undiacritized text, synthesis of spoken dialect, and engine optimization.
Promising future work includes experimenting with vowel processing, possibly treating short vowels as unit features, or automatic diacritization optimized for speech synthesis.