We considered age-related disguise as the intentional modification of the speaker's voice to sound like a younger or older person. Corpus 60 speakers, native Finnish Speech: read text in Finnish and English 26 segments in each voice type: modal, old voice, young voice. Odyssey 2016, pp. 277-282.
Performance evaluation Feature: Mel-frequency cepstral coefficients (18 MFCCs + Δ + ΔΔ). ASV systems: GMM-UBM and i-vector. Test setup For each of the conditions (natural and disguise), the number of trials are: Female: 403 genuine, 24180 impostor. Male: 377 genuine, 21112 impostor. Channels: Close-talk microphone, 2 smartphones. Odyssey 2016, pp. 277-282.
Recommend
More recommend