Design of Efficient Speech Emotion Recognition Based on Multi Task Learning

Speech emotion recognition technology includes feature extraction and classifier construction. However, the efficiency is reduced due to noise interference gender differences. To solve this problem, paper used two multi-task learning models based on adversarial learning(ASP-MTL). The first model took as main task auxiliary task, removed part identified by task. After identifying non-noise part, second was constructed. classification These can not only use shared information learn relationship between different tasks, but also identify specific tasks. This Audio/Visual Emotion Challenge (AVEC) database AFEW6.0 database,which were recorded in field environment. Considering problem of data imbalance datasets, balance operation carried out sets process preprocessing. shows an increase around 10% terms accuracy F1 score with recent works using AVEC which proved that has made a great progress SER.