Empathic Voice: Enabling Emotional Intelligence in Virtual Assistants

Situated at the intersection of Natural Language Processing (NLP), Understanding (NLU) and Generation (NLG), Virtual Assistants have revolutionised the way we interact with technology, providing us with convenient and personalised assistance in a variety of domains. With advances in artificial intelligence, the rise of large language models has significantly enhanced the capabilities of virtual assistants, differentiating them from simple chatbots and creating a new type of VA defined Intelligent Virtual Assistant (IVA). This dissertation analyzes the holistic process of developing an emotional intelligence empowered voice virtual assistant from scratch, capable of engaging in natural language conversations with the user, exploring techniques to enhance user experiences by leveraging the power of large language models and emotion recognition systems. A comprehensive framework is proposed, incorporating natural language understanding, emotion detection, dialogue management and voice synthesis. The effectiveness of the virtual assistant is then evaluated through users evaluations and performance qualitative metrics over a specific use case to demonstrate the model's capabilities, namely an empathic conversation with film characters associated with specific tasks to be performed by the user. The findings of this work help inspire the development of a new generation of voice virtual assistants that harness the potential of large language models to deliver empathic and expressive conversational experiences.