The characterization of a speech signal using nonlinear dynamical features has been focus of intense research lately. In this work, the results obtained with time-dependent largest Lyapunov exponents (TDLEs) in a text-dependent speaker verification task are reported. The baseline system used 10 cepstral coefficients and 10 delta cepstral coefficients, and it is shown how the addition of TDLEs can improve the system’s accuracy. Cepstral mean subtraction (CMS) was applied to all features in the tests, as well as silence removal. The telephone speech corpus used, obtained from a subset of CSLU Speaker Recognition corpus, was composed by 91 different speakers, speaking the same sentence.