Descript Overdub

Generates natural-sounding voiceovers by cloning a speaker's voice characteristics from existing audio samples, using deep learning models trained on prosody, tone, and speech patterns. The system analyzes source audio to extract voice embeddings, then synthesizes new speech matching those characteristics while accepting text input for the desired content. Integration with Descript's audio timeline allows direct placement of generated audio into projects without external rendering.

Maps synthesized speech back to the original transcript timeline, automatically calculating phoneme-level timing and adjusting playback speed to match original pacing or target duration. The system uses forced alignment algorithms to sync generated audio with transcript segments, enabling precise placement of voiceovers at specific transcript positions without manual time-shifting.

Generates multiple voiceover variations from the same script with different synthesis parameters (tone, speed, emphasis) and displays them as parallel tracks or switchable layers in the timeline. Users can audition variations in real-time, compare side-by-side, and select the best take without leaving the editor or managing separate audio files.

Allows editing of transcript text directly in the editor, with real-time synthesis and preview of how changes sound when spoken. Changes to transcript segments trigger immediate re-synthesis of affected voiceover sections, and the preview updates in the timeline without requiring manual re-generation or export steps.

Stores voice cloning profiles (speaker embeddings and synthesis parameters) as reusable assets that can be applied to new scripts across multiple projects. Once a speaker is cloned in one project, their voice profile is saved and can be instantly applied to new text in other projects without re-sampling or re-training.

Exposes synthesis parameters (tone, energy, emphasis, pacing) as adjustable sliders or presets that modify how the cloned voice delivers text. The system applies these parameters to the synthesis model to shift prosody, pitch variation, and speech rate without changing the underlying voice identity, enabling fine-grained control over delivery style.

Processes multiple transcript segments or script sections in a single operation, generating voiceovers for all segments with consistent speaker profile and synthesis parameters. The system queues synthesis jobs, manages API rate limits, and places all generated audio into the timeline with automatic timing synchronization, reducing manual per-segment generation overhead.

Overdub operates natively within Descript's non-linear audio/video editor, accessing transcripts, timelines, and media assets directly without export/import steps. Voiceovers are placed as native timeline tracks, inherit project settings (sample rate, bit depth), and can be edited alongside original audio using Descript's standard editing tools (trim, fade, effects).