Jorge Prendes, Marc S. Ressl, and Roxana Saint-Nom
Speaker verification, platform architecture, database, data selection
A new speaker verification platform architecture is introduced. Its design is focused on flexibility and ease of use, allowing for rapid generation of diverse experimental configurations. It uses a relational database and a modular structure that give the platform versatility. An archetypal experiment using gaussian mixture models with expectation maximization and maximum a posteriori is proposed and carried out. The results are within the range of other platforms' performance. Future work will be oriented towards support vector machines and combinations of support vector machines with gaussian mixture models.
Important Links:
Go Back
