An international team of researchers has released a new emotion-aware dataset called MELD-ST. It includes labels that capture the emotional tone of speech, allowing models to understand and translate not just the words, but also the emotions that people convey in a dialogue. Using speech data from the TV show “Friends,” the dataset includes timestamps of each dialogue, enabling models to accurately translate the speech into different languages and detect emotions to create more natural translations.
Image credit: Romain Vignes