What is speaker diarization and why does it matter for LinkedIn repurposing?
Direct answer
Speaker diarization is the process of identifying and separating different speakers in an audio or video recording - labeling who said what. In SparkVox, diarization is used when you upload a podcast interview or multi-speaker call with Guest or Advisor perspective. It separates your voice from the host's or other participants', so LinkedIn post drafts are written from what you said, not from what everyone said.
Why it matters for podcast guest repurposing
Without speaker diarization, a repurposing tool treats a podcast transcript as one voice. On a typical interview, roughly half the transcript is the host asking questions, reacting, or adding commentary. If the tool cannot tell who said what, it drafts posts that open with the host's setup or quote their questions as if you said them. The result is content that does not sound like you - or worse, content that misattributes ideas.
How SparkVox handles it
When you create a project with Guest (Podcast / Interview) or Advisor (Knowledge & Advisory) perspective using an audio or video source, SparkVox runs diarization using Gladia after captions when needed. The project pauses at awaiting_speaker status and shows 2-3 sample utterances from each detected speaker. You pick the voice that matches you on the mic. Only your utterances feed moment extraction and post generation.
When diarization does not run
Transcript-only uploads (.txt or .srt files) do not support speaker diarization - SparkVox has no audio to analyze. For guest appearances or advisory calls where you want speaker isolation, use the audio or video file, or a YouTube URL, so diarization can run. Transcript projects are limited to Full Conversation or Trainer perspective.
Related questions
Diarization accuracy varies with audio quality and the number of speakers. If sample lines shown in the speaker selection step do not clearly match the participants, you can re-upload with higher-quality audio or switch to a Full Conversation perspective and manually edit any misattributed lines in the draft review step.
Yes. SparkVox detects multiple speakers on panel discussions and group calls. You select your voice from the samples shown. Only your utterances feed the drafts.