The Sound Engine API.
Send a finished short-form video. Get back a timed sound map — classified moments, recommended sound effects and ambience with timecodes, levels and per-scene voice treatment. Your editor mixes it; the system decides everything.
Your keys
Create a key, keep it secret, and pass it as a bearer token. Usage is metered per second of video analysed.
Loading…
Request
One call. Pass a videoUrl (≤ 120s) to analyse a video end to end, or pass your own classified moments to skip analysis.
curl -X POST https://peakcut.ai/api/v1/sound-map \
-H "Authorization: Bearer pk_live_..." \
-H "content-type: application/json" \
-d '{ "videoUrl": "https://.../clip.mp4" }'Response
A structured analysis plus a cue map your editor places on the timeline.
{
"analysis": {
"durationSec": 42, "sceneCount": 5, "overallMood": "energetic",
"summary": "...", "scenes": [ ... ],
"moments": [ { "atSec": 3.2, "moment": "reveal", "intensity": 0.8 } ]
},
"cueMap": {
"cues": [
{ "atSec": 3.2, "durationSec": 1, "soundKey": "...", "moment": "reveal",
"gainDb": -2, "voiceTreatment": null }
],
"voiceRegions": [
{ "fromSec": 0, "toSec": 8, "ambienceKey": "...",
"treatment": { "space": "small_room", "reverb": 0.25, "tone": "muffled", "bed_level_db": -18 } }
]
}
}Moment taxonomy
Moments are classified into a fixed vocabulary — the same codes you can pass in directly.
How it decides
The engine analyses the whole video — scene cuts, on-screen action and emotion, sudden movement, and the existing audio — highlights the moments worth scoring, and picks sounds, counts and levels automatically. Louder beds for loud environments (a plane, an excavator), and a voice-treatment profile so a flat AI voiceover sits in each space. There are no intensity knobs; you give a thumbs up or down in the editor and the library learns.