Adobe's Corrective AI: Revolutionizing Voice-Over Emotions
Adobe has unveiled a groundbreaking feature called Corrective AI, which has the potential to transform the way voice-overs are produced. In a recent demonstration, Adobe showcased how this AI technology can dramatically alter the emotional tone of a voice-over in mere seconds. By highlighting specific text and selecting from a range of preset emotions, the software can seamlessly transition from a flat delivery to a confident tone, and even whisper, all without altering the original recording.
This innovative feature is part of Adobe's MAX Sneaks showcase, a platform where the company unveils cutting-edge technologies and prototypes. The Sneaks event provides a glimpse into Adobe's future creative tools, many of which are expected to be integrated into the company's full creative suite within months. During the MAX conference, Adobe also introduced generative speech for Firefly, allowing users to manipulate the inflection of voices. However, Corrective AI takes this a step further by enabling users to enhance existing voice performances rather than relying solely on AI-generated voices.
One of the most impressive demonstrations involved Adobe's Project Clean Take, a prototype that can separate different elements of an audio track. It successfully differentiated between voices, ambient noise, and sound effects, even in complex scenarios. For instance, it effectively removed the sound of a drawbridge bell that had drowned out the host's voice, and it could restore the bell's sound by adjusting the levels of the split-up tracks.
Another practical application was demonstrated when Adobe's AI model separated licensed music from a video, replaced it with a similar track from Adobe Stock, and applied effects to mimic the original ambiance. This feature not only saves time and effort for video editors but also helps creators avoid copyright strikes on platforms like YouTube.
Adobe's Oriol Nieto showcased another AI-powered feature, where the model analyzed a video and broke it down into scenes, applying emotional tags and descriptions. It then automatically added sound effects, such as an alarm clock sound, based on the scene's context. While not perfect, the system can be improved using a conversational interface, similar to ChatGPT, to guide the AI in making more accurate choices.
Looking ahead, Adobe's experimental features are expected to find their way into the company's creative suite. For instance, Harmonize, a feature that automatically adjusts assets for accurate color and lighting, was first demonstrated at the Sneaks event last year and is now available in Photoshop. These advancements come at a time when the creative industry is grappling with the impact of AI, particularly in the context of voice actors' recent strike to secure protections around AI usage.
