Skip to Content
kindui

Transcript

Show a recording's words by speaker and time, follow playback, and search what was said.

Use a transcript when the agent works with a call, a meeting or a voice note, so readers can follow along as it plays, jump to a moment, and find what someone said. Each segment shows its speaker, its start time and its words, and the segment and word playing at currentTime are marked.

Priya

Morning, everyone. Let's start with the release checklist for Thursday.

Marco

The migration ran cleanly on staging last night, so that box is ticked.

Priya

Great. Is the billing page still waiting on review?

Lena

It is. I pushed the copy changes an hour ago, and the screenshots are in the thread.

Marco

I can take that review after lunch.

Press Play to follow the recording, or press a segment to jump to it. Type billing in the search field to show only the segments that mention it.

Installation

npx shadcn@latest add @kind-ui/transcript

Composition

Transcript
├── TranscriptSearch
│   └── TranscriptCount
└── TranscriptViewport
    └── TranscriptList
        └── TranscriptSegment
            ├── div[data-transcript-header]
            │   ├── TranscriptSpeaker
            │   └── TranscriptSeek
            │       └── TranscriptTime
            └── TranscriptText
                └── TranscriptWord

TranscriptSearch renders a UI SDK Input with a clear button, and TranscriptCount beside it says how many segments match. TranscriptViewport scrolls the segments, and TranscriptList renders a TranscriptSegment for each one in view. A segment's header holds TranscriptSpeaker and its start in TranscriptTime, inside a TranscriptSeek button while onSeek is set. TranscriptText holds a TranscriptWord for each word, or one for the whole text when the segment has no words.

Usage

Segments

Pass segments in time order. Each has an id, a speaker, a start and end in seconds, and its text. Give it words, each with its text, start and end, to mark the word playing; without them, the transcript marks the whole segment. Only the segments in view are rendered, so a long recording costs the same as a short one. The view is 24rem tall, whatever the search leaves in it; set --kind-transcript-height to change it.

<Transcript
  segments={[{ id: "1", speaker: "Priya", start: 0, end: 4.5, text: "Let's start with the release checklist." }]}
/>

A segment added later fades in once; the ones already there stay where they are.

Playback

Pass currentTime as the recording plays. The segment at that time gets aria-current and its word data-current, and the words still to come in it are dimmed; in a gap between segments, nothing is marked. While focus is outside the transcript, a new segment scrolls to the middle of the view. Once someone focuses a time or types a search, the view stays where they put it. Pass follow={false} to never scroll.

Seeking

Pass onSeek to make each time a Seek to 0:05 button. A press on the time or anywhere on its segment calls onSeek with the segment's start in seconds, unless the press selected some of its text. Set your player's position there and pass the new currentTime.

Type in the search field to show only the segments whose text or speaker matches, with what matched marked. The count, like "2 matches", is announced. Clearing the search shows every segment again, and the segment that held focus keeps it. Pass defaultQuery to start with a search, or control it with query and onQueryChange.

Keyboard and Motion

The search field, the scrolling view and the times are a Tab stop each. Among the times, Up Arrow and Down Arrow move between segments and Home and End go to the first and last, bringing segments that aren't rendered into view; Enter or Space seeks. Times are written in locale, else the page's language. Following playback scrolls smoothly; under prefers-reduced-motion, or with motion="reduced", it jumps and new segments fade in over 120 ms, and with motion="none" nothing moves.

API Reference

Prop

Type

Agent

An agent connected with createAgent finds the transcript by its label and reads query. It can search with any text, which filters the segments as typing does, and while onSeek is set it can seek to a segment by its id; the segments are listed by speaker and time, and an id that isn't there is refused.

Prop

Type

Runs in React and Web Components. On <kind-transcript>, set segments as a property; the other props are kebab-case attributes. Add can-seek for the times to seek: onSeek is the seek event, and onQueryChange is querychange.