Odyssey 2012

The Speaker and Language Recognition Workshop

Memory and Computation Effective Approaches for i-Vector Extraction

Presented by:
Sandro Cumani
Sandro Cumani, Pietro Laface and Vasileios Vasilakakis

This paper focuses on the extraction of i-vectors, a compact representation of spoken utterances that is used by most of the state-of-the-art speaker recognition systems. This work was mainly motivated by the need of reducing the memory demand of the huge data structures that are usually precomputed for fast computation of the i-vectors. We propose a set of new approaches allowing accurate i-vector extraction but requiring less memory, showing their relations with the standard computation method introduced for eigenvoices. We analyze the time and memory resources required by these solutions, which are suited to different fields of application, and we show that it is possible to get accurate results with solutions that reduce both computation time and memory demand compared with the standard solution.