The L 2 F Language Verification Systems for Albayzin-2010 Evaluation

This paper presents a description of INESC-ID’s Spoken Language Systems Laboratory (L 2 F) Language Verification systems submitted to the ALBAYZIN-2010 evaluation. The primary submission consists of the fusion of six individual subsystems: one Gaussian supervector approach with support vector machines that relies on the acoustic characteristics extracted by a front-end of shifted deltas, and five individual Phone Recognition and Language Modeling detectors based on five different phone tokenizers. Additionally, two contrastive systems have been developed. Language detection results have been submitted for all the evaluation conditions for every system. The main particularity of the systems developed for this evaluation is that individual language models for clean and noisy conditions have been trained for each target language. Results for the different systems and evaluation conditions are reported.