Artificial intelligence is advancing rapidly, and industry transformation is undeniable, whether in terms of task automation, data analysis or content creation. One of the areas that generates most doubts is that of emotion: can artificial intelligence recognise emotions, and emulate them? Could it actually feel them? In this article, based on a literature review by Tan, Q., & Huang, Y. (2025), we will see how emotions are identified and recreated, as well as the current and potential applications of this technology, ranging from customer service, to education, health or entertainment.
How Does AI Emotion Recognition Work?
Emotion recognition systems analyse facial, voice and text data to identify feelings such as joy, sadness or anger. There are three main approaches:
- For facial expresion recognition, algorithms are used to detect and classify expressions in static (images) or dynamic (videos) format. It is based on deep learning techniques that identify key micro-expressions and muscle movements.
- In emotion recognition in speech acoustic characteristics such as pitch, intensity and speed of speech are analysed to interpret emotional states.
- Text sentiment recognition uses natural language processing (NLP) models to identify emotions in written messages. It is trained with large volumes of text labelled with specific emotions.
These systems are already in use in a variety of contexts. For example, in e-commerce they allow customer satisfaction assessment through the analysis of comments and facial expressions in video calls. In healthcare, they are used to monitor the emotional state of patients with neurodegenerative diseases.
AI-Made Emotion: Generating Realistic Expressions
In addition to recognising emotions with AI, they can now also be generated in different formats:
- Generating Facial Expresions: Models such as Generaetd Adversarial Networks (GANs) can create face animations showing realistic expressions based on text or audio.
- Generating Speech Emotion: Using voice conversion techniques, a system can transform a neutral recording into a version that expresses specific emotions, while maintaining the identity of the speaker.
- Generating In-Text Emotion: There are already several language models that can produce texts with different emotional tones, adapting their style according to the context.
Such technologies are being implemented in virtual assistants, video games and personalised advertising. For example, there are meditation apps that generate audios with a soothing voice tone and customer service platforms that adjust their responses to convey empathy.
Ethical Challenges and Dilemmas
Despite its potential, emotion recognition and generation present significant challenges:
- Precision and Bias: AI may misinterpret emotions due to cultural differences or atypical expressions. It should be borne in mind that many of the models use data labelled with one emotion or another, so the results they provide depend to a large extent on the biases and interferences that may have been included in their training.
- Privacy: The analysis of emotions in personal interactions raises questions about the collection and use of sensitive data.
- Manipulation and misinformation: The generation of realistic content can be used to create deepfakes or influence public opinion in unethical ways.
It is crucial that companies adopting these technologies do so responsibly, ensuring transparency in their use and respecting users’ privacy.
The Future of Emotional AI
Emotion recognition and generation continues to evolve and integrate into increasingly sophisticated applications. Multimodal systems, which combine face, voice and text information, are expected to improve the accuracy of emotional assessment. In addition, the development of more ethical and transparent models will help mitigate risks and build public confidence in these tools.
For all those interested in innovation in the field of communication, such as influencers, advertising and marketing professionals, and even artists, these technologies open up a range of opportunities to improve interaction with audiences, take personalised experiences to another level and generate engaging and original content. However, it is also essential to be informed about its implications and to contribute to the ethical use of AI.
References
Tan, Q., & Huang, Y. (2025). Emotion Recognition and Generation: A Comprehensive Review of Face, Speech, and Text Modalities. Journal of Theory and Practice in Humanities and Social Sciences, 2(1), 7-17. https://doi.org/10.5281/zenodo.14636258
