Incorporating Duration Information into I-Vector-Based Speaker Recognition Systems

Most of the existing literature on i-vector-based speaker r ecognition focuses on recognition problems, where i-vectors ar e extracted from speech recordings of sufficient length. The majority of modeling/recognition techniques therefore simp ly ignores the fact that the i-vectors are most likely estimated u nreliably when short recordings are used for their computati on. Only recently, were a number of solutions proposed in the lit erature to address the problem of duration variability, all tr ea ing the i-vector as a random variable whose posterior distribut ion can be parameterized by the posterior mean and the posterior covariance. In this setting the covariance matrix serves as a measure of uncertainty that is related to the length of the av ailable recording. In contract to these solutions, we address t he problem of duration variability through weighted statisti c . We demonstrate in the paper how established feature transform ation techniques regularly used in the area of speaker recogn ition, such as PCA or WCCN, can be modified to take duration into account. We evaluate our weighting scheme in the scope o f the i-vector challenge organized as part of the Odyssey, Speaker and Language Recognition Workshop 2014 and achieve a minimal DCF of 0.280, which at the time of writing puts our approach in third place among all the participating instituti ons.