Skavio Meeting: create a first sample with fictional participants
Skavio Meeting turns a meeting recording into a speaker-aware transcript and a structured report. For a first run, use a short fictional discussion whose outcomes you already know. Include one agreed decision, one accepted task, one tentative suggestion and one unanswered question—then check whether the generated draft preserves those distinctions.
1. Write the reference before you record
Prepare a three-to-five-minute discussion about an invented design project. Use three fictional roles: Alex, the coordinator; Sam, the designer; and Priya, the reviewer. Label both the script and recording “Fictional sample—not a real business meeting.” The following dialogue provides your review anchors. It is an illustrative script, not an excerpt from a processed Skavio result. Add substantive opening remarks about the fictional project to give each reader enough speaking time, but keep these decisions and commitments unchanged:
- Alex: “I propose that the next step is an internal review, not a public launch.”
- Sam: “I agree: an internal review, not a public launch. I will prepare two mock-ups by 12 November 2026.”
- Priya: “I agree with holding an internal review. I might investigate another colour scheme, but I am not committing to that task.”
- Alex: “Who will approve the final design?” Leave this question unanswered and end the discussion.
2. Use three readers, not just three names
Ask three consenting readers to perform the roles. Obtain their permission for recording and upload, and separately for any publication. Invented names and project details do not make the readers’ actual voices anonymous. Record in a quiet setting with clear turn-taking. Give each role a substantive opening turn: AssemblyAI’s speaker-diarization documentation recommends at least 30 seconds of uninterrupted speech per speaker for best results. That is recording guidance, not a Skavio upload requirement or an accuracy guarantee. Reading all three roles yourself can demonstrate the report workflow, but it is not a meaningful test of separating three distinct voices. Avoid background conversation, music and overlapping speech in this first exercise.
3. Submit the recording through the current interface
Open Skavio Meeting at https://getskavio.com/meeting/. Check the current upload requirements and any displayed processing quote or eligibility conditions before submitting the recording. Follow the formats and limits shown there rather than assuming that another tool’s requirements apply. Keep the script and original recording available for review. Meeting’s supported outputs include a speaker transcript, summary, key points, decisions, action items, open questions and a follow-up draft. Check the transcript against the audio. For the report, focus on whether it preserves the discussion’s meaning rather than reproducing the script word for word.
4. Check attribution before checking commitments
Start with the speaker transcript and listen to the recording alongside it. Speaker diarization labels different speakers; it does not, by itself, establish their names. AssemblyAI documents speaker identification as a separate capability. Do not assume Skavio automatically knows which speaker is Alex, Sam or Priya. Check attribution against the recording before attaching those fictional names to speaker labels. Then compare the report with your reference. The following are intended review targets, not observed results. Look for omissions as well as invented details:
- Decision: the group agreed to an internal review. A public launch was explicitly not the agreed next step.
- Accepted task: Sam committed to preparing two mock-ups by 12 November 2026. Check the owner, quantity and deadline.
- Tentative suggestion: Priya might investigate another colour scheme, but explicitly did not accept that task. It must not become a firm assignment.
- Open question: final design approval remains unresolved. The report should not invent an approver.
- Follow-up draft: check that it does not turn Priya’s suggestion into a promise or describe the project as ready for launch.
5. Keep a reviewed copy, not an assumed approval
Download a supported TXT, DOCX or PDF output and check the exported document too. Keep any corrections in a separate reviewed copy unless the current interface confirms an editing feature. Clearly distinguish the original generated draft from your corrected version. Before sharing a follow-up, verify names, commitments and unresolved questions again. Keep sending under human control; a generated draft is not authorization to make promises on someone’s behalf. If you later test overlapping speech or a noisier setting, treat that as a separate exercise. A clean scripted sample does not establish performance across accents, similar voices or real-world meeting conditions.