NIN-Align-v1
Connect each word to its moment.
Discuss this model ↗Upcoming model. Capabilities describe our plans; weights and inference are not available.
Connect each word to its moment.
Align transcript words with recording timestamps for captions, playback navigation and searchable meeting notes.
Pair a recording with its transcript to locate words in time.
Prepare word timestamps for captions and playback navigation.
Connect searchable notes to the relevant moment in a recording.
See the intended workflow.
Illustrative example, authored by NIN. This is not a live model prediction.
Recording + transcript scenario
A recording begins with the words: “Hello everyone.”
Illustrative word timestamps
[
{ "word": "Hello", "start": 0.20, "end": 0.55 },
{ "word": "everyone", "start": 0.60, "end": 1.10 }
]Times here are invented to show the intended format. Alignment quality and supported formats will need evaluation.
On-device / private-hosted
Built around your workflow.
On-device execution in NIN products and private hosting for organisational workflows are intended deployment options.
Details after evaluation.
Model size, hardware requirements, supported formats, licensing and measured performance will be documented when available. No release date is announced.
Where could NIN-Align-v1 help?
Tell us about your workflow, the data it uses and what a useful result would look like. Discuss an evaluation partnership with NIN.
Discuss this model ↗