This work generalizes the duration model, the time elapsed between two consecutive transactions, such as financial transactions data; and explores the consequences of durations in the instantaneous volatility. The approach have been motivated by Engel and Russel[3], that proposed an autoregressive conditional duration (ACD) model to explain the variation of volatility, where the innovations proposes were Exponential andWeibull distributions. Besides they used the GARCH-t to modeling the instantaneous volatility in high frequency data. This work uses the Generalized Gamma to the innovations in order to generalize the ACD model, this distribution has been first suggested by Zhang, Russel and Tsay[9], in the threshold ACD (TACD) framework. We justify the generalized Gamma specification in order to allow for more flexibility than the ACD model of Engel and Russel[3]. We find evidences that the ACD model with this specification was better to captur the behavior such as sub- dispersion.