This paper presents results on gender identification (GI) for children's speech, using the OGI Kids corpus and GMM-UBM and GMM-SVM systems. Regions of the spectrum containing important gender information for children are identified by conducting GI experiments over 21 frequency sub-bands. Results show that the frequencies below 1.8 kHz and above 3.8 kHz are most useful for GI for older children, while the frequencies above 1.4 kHz are most useful for the youngest children. The effect of using age-independent and age-dependent gender modelling (including the effects of puberty on boys voices) is explored. The application of intersession variability compensation is explored but experiments showed only little improvement. Experiments on human GI were also conducted and the results show that the humans do not achieve the performance of the machine.
Bibliographic reference. Safavi, Saeid / Jančovič, Peter / Russell, Martin / Carey, Michael (2013): "Identification of gender from children's speech by computers and humans", In INTERSPEECH-2013, 2440-2444.