A Recurrence Network Approach for Characterization and Detection of Dynamical Transitions During Human Speech Production