Today, call centers are one of the most important means of providing service to consumers and end users. Call centers provide a cheap and effective way to interact with consumers in a more personal manner. However, the managers of such centers find it difficult to grade the performance of their employees due to the abstract nature of the conversation. Many factors, such as consumer satisfaction, duration of call, and improving the company’s image from the consumer’s perspective, determine the effectiveness of the call center employee.

In this presentation, Sathvik proposes a framework that can monitor, analyze, and classify conversations between agents and customers. In a two-step process, we first diarize the calls using vector quantization, recognize turns, and annotate acts. Robert Bosch Engineering uses hidden Markov models with Mel-frequency cepstral coefficients in an approach to identify emotions. The engine has been tested using a Berlin database of German emotional speech, yielding 83% recognition rate. Real-world data of call center recordings is used to verify results.