Modelos supervisados asistidos por circuitos cuánticos variacionales para la caracterización conductual de interacciones orales en centros de contacto

Abstract

This project proposes an innovative approach for the behavioral characterization of oral interactions in contact centers, through supervised models assisted by variational quantum circuits. The objective is to classify attributes such as emotions, veracity, stress level, and speaker identity based on acoustic signals. The hypothesis posits that the use of hybrid quantum-classical models improves classification accuracy compared to traditional models. The justification lies in the need for more efficient systems for emotional and behavioral analysis in spoken language. An experimental prototype was developed using a simulated dataset of contact center conversations, processed through phonetic techniques such as MFCC. Dimensionality reduction methods (PCA) were applied, along with encoding classical data into parameterized quantum circuits. The architecture is based on supervised training of VQC models using parameter-shift optimization. Libraries such as PennyLane and openSMILE were used for signal processing and model training. As a preliminary result, the quantum model achieved accuracy comparable to or higher than classical models such as SVM and Random Forest in multiclass classification tasks. These results are encouraging, as they position variational quantum circuits as an important alternative for solving complex problems related to speech processing. It is concluded that quantum computing applied to machine learning offers promising advantages in voice analysis contexts and represents a high-potential research line for the future of spoken language processing. Keywords: supervised learning, variational quantum circuits, speech analysis, contact centers, emotions, quantum artificial intelligence.