@linto-ai/transcript-ui-core

Part of @linto-ai/transcript-ui — see there for the full package list and quick start.

The main piece of @linto-ai/transcript-ui — a Vue component for displaying and interacting with a transcript. It bundles the document state (createCore), the editor layout (Layout), and adapters that convert LinTO Studio API or WhisperX transcripts into a format it understands.

Usage

<script setup lang="ts">
import { TranscriptUI, mapApiDocument } from "@linto-ai/transcript-ui-core"
</script>

<template>
  <TranscriptUI ref="editor" locale="en" />
</template>

TranscriptUI handles everything on its own: it creates the document state, shows a loading spinner, then the transcript once a document is loaded.

Building your own layout instead

If you don't want TranscriptUI's default screen, createCore/provideCore and Layout are the two pieces it uses internally — use them directly to build your own.

Turning a transcript into something this package understands

• mapApiDocument — for a transcript from the LinTO Studio API.
• mapWhisperXDocument — for raw WhisperX output.

Data model

A document (EditorDocument) has a flat list of speakers and one or more channels.

• Channel — one audio/recording track (e.g. one microphone, one room). Has one or more Translations.
• Translation — one language track within a channel: the source language, or an auto-translated version. Exactly one per channel is the source (isSource: true). Holds an ordered list of Turns and, for the source, the AudioSource to play.
• Turn — one speech segment: a speakerId (pointing into the document's shared speakers map, or null if unassigned), a language, and either text (a plain string, e.g. from a live text-only source) or words (per-word timing, e.g. from ASR) — never both.
• Word — one word inside a turn: its text, character offsets into the turn's plain text, and optional start/end time and confidence.
• Speaker — id, display name, color. Stored once per document, referenced by id from every Turn — renaming a speaker updates every turn that points at them for free.

EditorDocument
├── speakers: Map<id, Speaker>
└── channels: Channel[]
    └── translations: Translation[]   (one is the source; others are translations)
        └── turns: Turn[]
            ├── speakerId → looked up in EditorDocument.speakers
            └── text (string) OR words (Word[]) — never both