Use cases
None of these are impossible without word timings. All of them currently mean generating audio and then paying something else to tell you where the words went.
Burned-in captions are the format's whole grammar, and getting them right normally means running the audio you just generated back through a transcriber to find out where the words fell. The timings come back with the audio, so that second pass — and its cost, and its errors — disappears.
This is the exact loop Ounie's own ask-video pipeline runs, which is why it exists.
Every word is a start and an end in milliseconds, which is what a timeline is made of. Trim on a word boundary, land a graphic on a phrase, hold a shot until a sentence finishes — without scrubbing, and without a waveform to squint at.
Feed them straight into Remotion, ffmpeg or an NLE's marker track.
A lip-sync driven by word boundaries beats one driven by a fixed cadence, and it is the cheapest possible input: no phoneme model, no viseme pipeline, just the boundaries you already have.
Punctuation stays attached to its word, so a sentence end is visible.
An agent that speaks needs to know when its own turn ends — for barge-in, for a timer, for the next thing in the queue. The last word's end is that number, returned before playback rather than measured after it.
The same call gives you the total duration, taken from the alignment itself.
Twenty-nine languages on the rich engine, thirty-two on the fast ones — and the timings tell you whether the German runs eleven seconds over the shot before anyone opens an edit.
A language the chosen engine does not speak is refused up front, not mispronounced.
Store the timings beside the file and every word becomes a seek point. Search a phrase, jump to the second it was said — for generated audio, this is free where transcription would be a running cost.
Runs are kept, and re-reading one costs nothing.
The cast
Adam
socialDominant, Firm
male · american · middle aged
Alice
explainerClear, Engaging Educator
female · british · middle aged
Bella
explainerProfessional, Bright, Warm
female · american · middle aged
Bill
adsWise, Mature, Balanced
male · american · old
Brian
socialDeep, Resonant and Comforting
male · american · middle aged
Callum
characterHusky Trickster
male · american · middle aged
Charlie
conversationDeep, Confident, Energetic
male · australian · young
Chris
conversationCharming, Down-to-Earth
male · american · middle aged
Daniel
explainerSteady Broadcaster
male · british · middle aged
Eric
conversationSmooth, Trustworthy
male · american · middle aged
George
narrationWarm, Captivating Storyteller
male · british · middle aged
Harry
characterFierce Warrior
male · american · young
Jessica
conversationPlayful, Bright, Warm
female · american · young
Laura
socialEnthusiast, Quirky Attitude
female · american · young
Liam
socialEnergetic, Social Media Creator
male · american · young
Lily
explainerVelvety Actress
female · british · middle aged
Matilda
explainerKnowledgable, Professional
female · american · middle aged
River
conversationRelaxed, Neutral, Informative
neutral · american · middle aged
Roger
conversationLaid-Back, Casual, Resonant
male · american · middle aged
Sarah
broadcastMature, Reassuring, Confident
female · american · young
Will
conversationRelaxed Optimist
male · american · young
21 voices in the catalogue. GET /api/voices returns all of them, with the models and every language each one speaks.
Paste it in, pick a voice, and read the timings back before you build anything around them.