Analysis of Lombard effect under diffe with application to In-set Sp

This paper presents an analysis of Lombard speech produced under different types and levels of noise. The speech used for the analysis forms a part of the UT-SCOPE database and consists of sentences from the well-known TIMIT corpus, spoken in the presence of highway, large crowd and pink noise. Differences are shown to exist in the speech characteristics under these varying noise types. The deterioration of the EER of an in-set speaker identification system trained on neutral and tested with Lombard speech is also illustrated. A clear demarcation between the effect of noise and Lombard effect on noise is also given by testing with noisy Lombard speech. The effect of test-token duration on system performance under the Lombard condition is addressed. It is seen that test duration has no effect on the EER under Lombard effect. The average EER for 3s test duration is 14.7, 28.3, 48.2, 51.3 for neutral clean, clean Lombard, noisy neutral and noisy Lombard respectively, and 7.2, 26.4, 45.8, 50.8 respectively for 12s test duration.