Adobe's ‘Corrective AI’ Can Change the Emotions of a Voice-Over (www.wired.com)

At Adobe’s MAX Sneaks showcase, the company demoed “Corrective AI,” a set of audio tools that let editors alter the emotion and performance of existing voice-overs and surgically repair audio tracks without re-recording. Users can highlight transcript text and apply preset emotion tags (confident, whisper, etc.) to change inflection in seconds, or split a single mix into up to five separable stems—voices, ambient noise, music, effects—so offending sounds (a bell, loud background music) can be removed or individually rebalanced and even swapped for similar tracks from Adobe Stock with matched reverb and ambience. Beyond corrective dubbing, Adobe showed generative capabilities that auto-analyze scenes and add AI-generated sound effects (alarm clocks, car ambience, door closes) and a conversational interface for scene-specific edits. The company claims these sounds are commercially safe to use. Technically, the demos extend Firefly’s emotion-tagged speech to practical touch-up workflows and use source separation and scene-understanding models to place and synthesize audio assets. Significance: these tools could dramatically speed post-production and reduce costly ADR, but they also accelerate debates over consent and voice-rights—coming months after game voice actors secured AI-use protections—and foreshadow broader industry shifts as such Sneaks prototypes typically reach Adobe’s suite within a year or two.