With the help of deep learning algorithms and neural networks, voice clones have come a long way in emulating emotions. Although early voice cloning systems produced rather robot-like speech, the current technology can simulate emotions nearly indistinguishable by natural speech(-happy.sad.ang ), Both of these use neural models like the ones employed in next-gen Google Assistant, such as Tacotron 2 (used to determine pitch and speaking style) and WaveNet (used for natural sounding human-like voices), which enable them not just to judge the pitch and cadence of speech but also identify present in human conversation.
But the issue arises from the intricacies of human feelings. Voice synthesis platforms such as DeepMind by Google have indeed worked on models that can mimic emotion to a degree, and algorithms are now capable of recreating human emotions with 95% accuracy according to Purdue University, but even the most progressive systems tend not to be able fully emulate the human emotional spectrum. Its creators have said that at present, while mimicking emotional tones it achieves around 85% to 90% of the quality perceived by human listeners, but more subtle emotions like sarcasm or empathy remain difficult for voice cloning AI to capture correctly. That's a big step up from earlier models, which had an error rate of roughly 20%-25% when reproducing emotional speech.
There are some industries too like entertainment and customer service, who have been using voice clones to prepare customized and emotionally responsive virtual assistants. For example, Amazon's Alexa and Google Assistant have introduced simple emotional experiences to get people more involved. In fact, companies have seen an up to 15% increase in user satisfaction with virtual assistants that can respond with the correct emotional cues–for instance, by expressing cheerfulness when cheerful input is given and empathy when calibration output arises.
Still, sentiment mimicry within voice cloning movesthe bounds of ethics equally as much. It is used in mimicking public figures (local or international) and than can easily be scripted to make false claims. A high-profile case in 2019 saw fraudsters use a fake CEO voice clone to deceive staff into sending $240,000. If Elon Musk is to be believed, then “AI is far more dangerous than nukes” and it highlights the serious issue of voice cloning being misused.
For researchers who want to investigate how well it can manipulate the emotions, there are also more advanced "emotional speech synthesis" platforms like DupDub. These are platforms that allow easy creation of voice clone in high-quality that provide different emotional tones suitable for content creators, educators and businesses to use.