How project processing works
Transcription, speaker selection, moment extraction, and post generation.
When you submit a project, SparkVox runs an async pipeline. Processing time is deducted once at the start (per minute of content, rounded up). Post generation and on-demand post images are included in that fee.
Project statuses
pending → transcribing → extracting → generating → ready
(or awaiting_speaker between transcribing and extracting)
(or failed at any stage)What you can upload
- URL: YouTube or direct audio/video links.
- YouTube: Sparky tries native captions first (fast); if unavailable or speaker diarization is needed, it falls back to audio download and Gladia transcription.
- File upload: MP3, WAV, MP4, and other common formats from the New Project form.
- Transcript file: .txt or .srt - skips transcription and goes straight to moment extraction.
For YouTube projects, title, channel, and tags from the video help Sparky spell names and technical terms correctly in excerpts and posts.
Project perspective
When creating a project you choose a source kind and perspective. They control how the transcript is processed:
Podcast / Interview
| Perspective | Best for | What happens |
|---|---|---|
| Host | Hosts building a personal brand | Speaker selection - only your lines are used. |
| Guest | Guest appearances on someone else's show | Speaker selection - only your lines are used. |
| Full Conversation | Highlight reels of the whole episode | Full cleaned transcript is used. |
Knowledge & Advisory
| Perspective | Best for | What happens |
|---|---|---|
| Expert | Keynotes, solo trainings, thought leadership | Extracts frameworks and masterclass lessons from your content. |
| Advisor | Client calls and strategy sessions | Speaker selection - isolates your strategic advice from collaborative calls. |
| Trainer | Workshops, demos, walkthroughs | Full transcript used for step-by-step playbook posts. |
Transcript file uploads (.txt / .srt) use Full Conversation or Trainer perspective only (Host, Guest, and Advisor are disabled for transcript files).
Moments and posts
Sparky scores every candidate moment from your transcript (relevance 70+), ranks them by insight density, and keeps the strongest subset for generation. Expect roughly eight curated posts per 30 minutes of content - fewer for short recordings, more for long ones, with a hard cap of 15. When processing finishes, the project ready screen shows how many moments were found versus how many posts were generated (for example 14 found, 8 curated). Open your project to review drafts, edit, and approve each draft. Moments that fail generation may not show a post card.
Before posts are written, Sparky may lower moment scores for topics or post styles you often edit heavily or discard on past projects, and can drop demoted moments from the curated set. An optional positioning claim on Settings → Brief Sparky (see sparkvox.io/help/brief-sparky) can boost moments that support that claim.
How each post is drafted
- Sparky plans three funnel-aware content pillars (Awareness, Authority, Conversion) for the project, with per-stage structure and first-comment guidance.
- For each curated moment, Sparky generates two draft candidates in parallel and keeps the one with the stronger hook and readiness score.
- Awareness and Authority drafts can pass a distinctiveness check so generic lines are less likely to reach your review queue.
- Generation uses Brief Sparky, project presets, your voice profile, top-performing published posts when analytics exist, and optional community context from Engage.
Every recording you process is a content asset. Advisor and Expert perspectives are built for calls and trainings where your strategic thinking would otherwise stay in Otter and never reach LinkedIn.