Member Junction
    Preparing search index...

    Segments audio/video into chapters using its timed transcript.

    A 60-minute session recording embedded as a single vector is a mush vector — multimodal embedders only see a short window, and no single vector can represent an hour of distinct topics. This segmenter finds real boundaries first (silence gaps, speaker changes, a duration ceiling) and emits time-windowed chapters that each carry both a media reference (with StartMs/EndMs) and the transcript text for that window.

    That dual payload is deliberate: the media reference is what a multimodal model embeds for retrieval, while the transcript is what an agent can actually read, what a cross-encoder can rerank, and what keyword search can match.

    Cues come from any ASR source, including MemberJunction's realtime session capture, which already records per-turn text with speaker labels and timings.

    Hierarchy (View Summary)

    Index

    Constructors

    Accessors

    • get Key(): string

      The registration key for this segmenter. Must match the key passed to @RegisterClass so metadata-driven resolution round-trips.

      Returns string

    Methods

    • Resolve a registered segmenter by key via the MJ class factory. Returns null when no segmenter is registered under that key.

      Uses TryCreateInstance rather than CreateInstance deliberately: the latter never returns null for an unregistered key — it falls back to new BaseSegmenter(), a hollow object whose abstract Key/SegmentCore are undefined. That failure stays invisible until something calls it, so an unresolvable key must be reported as such here.

      Parameters

      • key: string

      Returns BaseSegmenter