Speaker Adaptation of Intimacy Factors based on F_0 Histograms

For realizing the dialog agent which gets along with human beings, it is necessary to detect the user’s mental state to the agent, e. g. the intimacy factor with it. In our previous work, we should that speaker’s intimacy factors reflected to the histograms of the fundamental frequency (F0). Because the frequently-used F0 range strongly depends on the speaker, the shape of F0 histogram related to the intimacy factor may differ from that of another speaker. In this paper, speaker adaptation of the intimacy factors based on F0 histograms is proposed. Roughly speaking, the histogram feature for any persons can exact appropriately according to EM alogorithm whose initial value is embraced F0 histogram’s mode value in the early time of utterances. The intimacy extraction accuracy for 7 persons is about 88 % based on the proposed approach.