Real-time voice affect recognition for call center agents

Call centers depend depend on their agents to effectively serve their clients. But often times, arguments between agents and clients occur. Inability to deal with this problem may lead client discontent, business loss or moving of client to other competitors. Thus, it is important for call centers to handle this disputes. This study aims to build a real-time affect recognition system with the use of dimensional labels to determine the agents\u27 human affective states. This will allow agents to evaluate themselves, improve their performance while talking to their clients and at the same time avoid arguments with clients. A Call Center Affect Recognition System (CARSys) was built in this study. There were two applications developed, namely, CARSys Stand Alone and CARSys Logs. CARSys Stand Alone is the application which predicts affective state in real-time. The representation of affective states is dimensional and the dimensions used was valence and arousal. CARSys Logs, on the other hand, is the application responsible for viewing the logs stored in a server. The results showed that the CARSys needs a lot of improvement in terms of the affect recognition and log generation. The models used by CARSys achieved a mean absolute error (MAE) of 0.2312 and 0.2028 for the valence and arousal, respectively. Testing using the SEMAINE database obtained an MAE of 0.2980 and o.3893 for the valence and arousal, respectively. Testing using the gathered data from CARSys testing, it gained an MAE of 0.2405 for valence and 0.7827 for arousal