MODEL AND DEVICE FOR COUPLING LANGUAGE FEATURE AND EMOTION FEATURE OF VOICE AND ESTIMATING EMOTION, AND GENERATION METHOD OF THE MODEL

摘要

To provide an emotion estimation model capable of estimating emotion of a vocalization subject from voice with high accuracy while a risk of determining a language type related to the voice first is avoided.SOLUTION: An emotion estimation model includes: a language identification model portion having a language feature extraction part for extracting a feature related to a language type from inputted voice data; an emotion recognition model portion having an emotion feature extraction part for extracting a feature related to emotion from the voice data; a coupling part for coupling output of the language identification model portion and output of the emotion recognition model portion; and an emotion estimation part for estimating an emotion type related to a vocalization subject of the voice data from output of the coupling part. The emotion recognition model portions are the plurality of emotion recognition model portions related to the languages of different types, and it is preferable that the emotion feature extraction parts of the respective emotion recognition model portions extract the feature related to the emotion with voice data as voice data of the language related to itself.SELECTED DRAWING: Figure 1