AI-selected break points grounded in the transcript, not just silence. Every candidate comes with a confidence score and a one-sentence rationale. Emit SCTE-35, HLS, and DASH sidecars ready for your dynamic ad server.
Someone on your team scrubs the timeline, guesses where a topic ends, drops a marker, and hopes. It's slow, inconsistent across episodes, and usually lands mid-thought. The alternative — pure silence-based placement — is worse: it cuts the host off mid-breath and ignores the story entirely.
You need break candidates that understand the content, not just the audio waveform.
Yes. The transcript-based approach is format-agnostic. Video-specific options (e.g. embedding SCTE-35 into the output m3u8) are available on video runs.
Any number. The model will tell you honestly if your target doesn't fit the content (e.g. "only 2 clean break points in a 10-minute episode") rather than forcing suboptimal picks.
Reject it, move it, or replace it. Your manual decision overrides the AI pick and is preserved across re-runs of the pipeline.
On pipelines with delivery_mode = auto_accept_on_confidence. If every marker clears your configured threshold, the run finalizes and ships to destinations without human review. If any marker falls below, the run holds for review.
Yes. SCTE-35 JSON, markers JSON, markers CSV, and an HLS patch file are all downloadable artifacts. Media output is optional and separately gated.