Modern TTS systems are capable of creating highly realistic and natural-sounding speech, making them an important tool when designing virtual agents. While sounding highly realistic, the process of customizing such TTS voices remains a complex task, mostly requiring the expertise of specialists within the field. One reason for this is the utilization of deep learning models, which are characterized by their expansive, non-interpretable parameter spaces, restricting the feasibility of manual voice customization. In this paper, we present a novel human-in-the-loop paradigm based on an evolutionary algorithm for directly interacting with the parameter space of a neural TTS model. We integrated our approach into a user-friendly graphical user interface that allows users to efficiently create original voices. Those voices can then be used to equip virtual agents with highly customized TTS capabilities by using an open-source programming interface provided by us. Further, in a first pilot study, we show that VoiceX is an appropriate tool for creating individual, custom voices.
Evaluating the quality of generated nonverbal behavior remains a major challenge in the development of generative models. While human evaluations are reliable, they are costly and impractical for large-scale or iterative optimization. In this work, we propose an objective evaluation framework based on aggregated ranks across multiple fidelity and diversity metrics, computed from both raw features and learned latent representations. Compared to existing works, our framework emphasizes consistency across multiple metrics, aiming to provide a more holistic assessment.
AI companions (AICs; persistent digital personas driven by large language models) are increasingly popular, yet there is limited understanding of how non-experts think about the notion of companionship with machine agents. This descriptive work builds on recent scholarship by exploring topics salient to users as they discuss AI companionship in online forums. From a corpus of curated entries (N = 3,621) across 14 subreddits, LLM-assisted analysis induced four high-level themes associated with socioemotionality, control, limitations, and imaginaries. The topical hierarchy suggests this form of machine companionship both converges with and diverges from human relations.
This paper presents the creation and perceptual validation of the EVE corpus, a resource, in English and French, for speech emotion recognition of audio and audiovisual content and for the generation of verbal and non-verbal behaviour. For each language, ten actors performed 10 linguistically and semantically neutral sentences with different emotions (fear, anger, happiness, sadness, disgust, surprise, confidence, confusion, contempt, empathy) and a neutral condition. Each was expressed at two arousal levels, with two trials per level. The emotional content of the corpus was perceptually validated by 600 participants per language. The corpus is accessible through this link: https://doi.org/10.58119/ULG/VREIOB.
This paper presents an exploratory study of users’ first impressions of personality traits and preference for virtual receptionist agents with different body types and genders. The results show that gender and body type interact, since a female muscular agent was rated high on many traits while a male muscular agent was rated low. Another finding is that even though agents with large bodies were rated higher or as high on important traits, e.g. helpful and knowledgeable, the muscular female agent was still preferred over them. This implies that further research on body types in agent design is needed and may be important to support representation and norm critical design.
