Editing AI Results - Getting the Details Right
Chapter 29: Editing AI Results — Getting the Details Right
Think of it like proofreading after a transcription service delivers the draft—the heavy lifting is done; you're just polishing the last few rough edges. The AI has done the work of capturing and structuring everything. Now you decide whether "Speaker 2" should actually say "Alice", and whether today's extraction is better than last week's version.
This chapter covers two kinds of editing:
- Part A: Speaker Labels — rename, merge, and reassign speaker labels in live transcripts
- Part B: Version History — browse, compare, and navigate previous AI extraction runs
Chapter Objectives
After reading this chapter, you will be able to:
- Rename a speaker label and have it update everywhere instantly
- Merge two speaker labels that belong to the same person
- Reassign an individual transcript segment to the correct speaker
- Understand where the version badge appears and how to use it
- Browse previous extraction runs without losing the latest version
- Identify when you're viewing a historical version via the stale-version banner
Part A: Speaker Labels
Why Labels Matter
Automatic speaker detection assigns placeholder names like "Speaker 1", "Speaker 2". These are functional but not readable. "Speaker 1 said we need to ship by Friday" is far less useful—and far harder to share—than "Alice said we need to ship by Friday."
Renaming takes 30 seconds and transforms the transcript from a technical output into a professional document.
Speaker label editing is available for live transcripts (recordings where live transcription was active, as described in Chapter 27). Transcripts generated by AI Extraction use the same labels and can also be renamed from the Transcript tab.
Renaming a Speaker
Renaming one speaker label changes every occurrence of that speaker across the entire recording — the transcript, summary, action items, decisions, and chapters all update simultaneously. If the AI wrote "Speaker 1 proposed a 15% budget cut", renaming "Speaker 1" to "Alice" makes that line read "Alice proposed a 15% budget cut" everywhere.
Step-by-step:
-
Open the recording in your Recording Library
-
Click AI Insights in the detail panel, then select the Transcript tab
- What you see: The transcript with all segments and speaker labels. On the left edge (or right side, depending on layout) is a Speakers panel listing all detected speaker labels.
-
In the Speakers panel, click the pencil icon ✏️ next to the speaker you want to rename
- What you see: The label becomes an editable text field, pre-filled with the current name ("Speaker 1").
-
Type the new name (e.g., "Alice")
- What you see: Characters replace the placeholder as you type.
-
Press Enter (or click away from the field)
- What you see: Every occurrence of "Speaker 1" across all transcript segments updates simultaneously to "Alice". The Speakers panel now lists "Alice" in place of "Speaker 1".
Tip: Rename speakers immediately after the recording while you remember who was speaking. Context fades quickly.
Merging Two Speakers
Sometimes the AI creates two labels for one person—this happens when a voice sounds different at different points in the recording (e.g., someone moved further from the microphone, or laughed and then resumed a normal tone). Merging combines both labels into one.
Step-by-step:
-
In the Speakers panel, find the speaker you want to absorb into another (the "source" speaker)
-
Click the merge icon (or right-click the speaker label and choose Merge into…)
- What you see: A dropdown or sub-panel appears listing all other detected speakers.
-
Select the target speaker — the one you want the source to be merged into
- Example: You want "Speaker 3" to become "Alice". Select "Alice" as the target.
-
Click Apply (or confirm in the popup)
- What you see: All segments previously attributed to "Speaker 3" are now attributed to "Alice". "Speaker 3" disappears from the Speakers panel. The transcript updates everywhere instantly.
Note: Merging cannot be undone. If you merge incorrectly, you will need to rename individual segments manually (see next section). Double-check before applying.
Reassigning a Single Segment
If most segments are correctly attributed but a specific segment was misidentified, you can correct that individual segment without touching any others.
Step-by-step:
-
In the Transcript tab, find the segment attributed to the wrong speaker
-
Click the speaker name displayed on that segment (e.g., "Speaker 2")
- What you see: A small popup or inline dropdown appears listing all detected speakers.
-
Select the correct speaker from the list
- What you see: The segment's speaker label updates to the selected speaker. Other segments are unaffected.
This is useful for isolated misattributions—for example, when someone else briefly answered a question and the AI attributed it to the wrong person.
Limitations of Speaker Editing
| Limitation | Detail |
|---|---|
| Live transcripts only | Speaker editing is available for transcripts generated during recording or by AI Extraction. It is not available for imported transcript files. |
| No rename-to-existing | You cannot rename "Speaker 2" to "Alice" if "Alice" already exists as a label—use Merge instead. |
| No undo | Speaker changes save immediately and cannot be undone. Use Merge with care. |
| Names are local | Speaker names are stored with this recording only. They do not affect other recordings. |
Part B: Version History
Why Versions Exist
Every time you click ⟳ to regenerate an AI extraction, SeaMeet quietly saves the previous results as a numbered version snapshot before overwriting them. Nothing is ever destroyed.
This means you can:
- Regenerate freely without worrying about losing a good result
- Compare the current extraction to an older one
- Return to a previous version if a regeneration produced worse results
The Version Badge
The version badge appears in the AI Insights tab bar, between the content tabs and the export buttons:
[Transcript] [Summary] [Actions] [Decisions] [Chapters] · v3 ▾ [MD] [SRT] [JSON] [⟳]
- "v3 ▾" — you're viewing version 3 (the latest), the ▾ indicates a dropdown
- The badge only appears after at least one regeneration — first-time users see no change to the interface
If you have run AI Extraction only once (never regenerated), the badge will not appear. This is by design — there is nothing to compare against yet.
Browsing Versions
Click the version badge to open the version picker:
┌─────────────────────────────────────────┐
│ ✓ v3 — Latest │
│ Mar 1 2026, 9:15 AM │
├─────────────────────────────────────────┤
│ v2 │
│ Mar 1 2026, 9:00 AM │
├─────────────────────────────────────────┤
│ v1 │
│ Feb 27 2026, 10:30 AM │
└─────────────────────────────────────────┘
- Checkmark (✓) marks the version you are currently viewing
- Versions are listed newest-first
- Each entry shows the date and time the extraction completed
- Latest is always the most recent run
To view an older version:
- Click the version badge ("v3 ▾")
- Click the version you want (e.g., "v1")
- What you see: All five tabs reload with the historical data from that run. The stale-version banner appears at the top.
The Stale-Version Banner
Whenever you're viewing any version other than the latest, a slim amber banner appears at the top of the content area:
┌──────────────────────────────────────────────────────────────┐
│ ⚠ Viewing v2 of 3 · Generated Mar 1, 9:00 AM [View latest →]│
└──────────────────────────────────────────────────────────────┘
- "Viewing v2 of 3" — which historical version and how many versions exist
- "Generated Mar 1, 9:00 AM" — when this version was created
- "View latest →" — one click returns you to the most recent extraction
The stale-version banner is a safety signal. It ensures you always know when you're not looking at the current results, preventing confusion when sharing or acting on the content.
Exporting while viewing a historical version exports that version's data, not the latest. The export buttons work normally — they just operate on whatever is currently displayed.
The Live Baseline
If live transcription was active during the recording (Chapter 27), the original real-time transcript is always available as a special entry at the bottom of the version picker, regardless of how many regenerations you've done:
┌─────────────────────────────────────────────┐
│ ✓ v3 — Latest │
│ Mar 1 2026, 9:15 AM │
├─────────────────────────────────────────────┤
│ v2 │
│ Mar 1 2026, 9:00 AM │
├─────────────────────────────────── ──────────┤
│ v1 │
│ Feb 27 2026, 10:30 AM │
├─────────────────────────────────────────────┤
│ Live (original) │
│ AI-powered · recorded Feb 27 │
└─────────────────────────────────────────────┘
When "Live (original)" is selected:
- Only the Transcript tab is populated — this is the word-for-word text captured in real time during the recording
- The Summary, Actions, Decisions, and Chapters tabs show "not available for this version" — the live AI engine produced only a transcript, not the full set of extraction outputs
- The stale-version banner appears, since this is not the latest extraction
The Live baseline is useful for comparing:
- What the real-time AI heard (Live) vs what the post-processing AI transcribed (any vN)
- Whether speaker detection improved or changed between the live pass and subsequent extractions
Limitations of Version History
| Limitation | Detail |
|---|---|
| Read-only | Historical versions are snapshots. You cannot edit them directly. To edit, switch to Latest and make changes there. |
| No delete | Individual versions cannot be deleted (the entire recording can be deleted, which removes everything). |
| No rename | Versions cannot be named or labelled (e.g., "good version"). They are identified by number and timestamp only. |
| No diff view | Versions must be viewed one at a time. Side-by-side comparison is not available in this release. |
| Badge requires 1+ regen | The badge only appears after at least one regeneration. A recording with a single extraction run shows no badge. |
Troubleshooting
"Version badge not showing"
Symptom: You've run AI Extraction but no version badge appears in the tab bar.
What's happening: The badge only appears after at least one regeneration (clicking ⟳). A single initial extraction run does not create a version — there is nothing to compare against.
What to do: Click ⟳ to regenerate. After the new extraction completes, the badge will appear showing "v2 ▾".
"Speaker changes not saving"
Symptom: You rename a speaker and click away, but the old name reappears on next load.
Try these in order:
- Check your internet connection — some speaker edits require a sync step
- Wait 5 seconds before closing the recording — changes are saved with a brief delay
- Reload the recording from the library and check again
If the issue persists, try restarting SeaMeet and applying the rename again.
"The merge option is greyed out"
Symptom: Right-clicking a speaker label or clicking the merge icon produces no response or a greyed-out option.
What's happening: Merge requires at least two distinct speaker labels. If only one speaker was detected, there is nothing to merge into.
What to do: The transcript may have only identified one speaker — this is normal for solo recordings or interviews where one person dominates.
Best Practices
Rename speakers right after the recording Memory of who said what fades quickly. Open AI Insights within an hour of the recording and do the renaming before context is lost.
Keep Latest as your working version Historical versions are for reference and comparison. Do not stay on an old version as your default view — the latest is always the most complete and most recently processed.
Export before regenerating — optionally Version history automatically preserves past results, so exporting before ⟳ is not strictly necessary. But if you want a standalone file (to email or attach), export first.
Use Merge before Rename for duplicates If you notice the same person appearing as two speakers (e.g., "Speaker 1" and "Speaker 3"), use Merge rather than renaming both manually. Merge is faster and ensures all future references are consolidated.
Review the Live baseline for accuracy If you ran live transcription, check the "Live (original)" view against the latest extraction. Real-time transcription sometimes catches nuances that post-processing misses, and vice versa.
Quick Reference
┌────────────────────────────────────────────────────────────┐
│ EDITING AI RESULTS │
│ Quick Reference │
├────────────────────────────────────────────────────────────┤
│ SPEAKER LABELS │
├────────────────────────────────────────────────────────────┤
│ Rename │ Speakers panel → ✏️ → type → Enter │
│ Merge │ Speaker → Merge into… → target → Apply │
│ Single segment │ Click speaker name on segment → pick │
│ Applies to │ All tabs: Transcript, Summary, Actions, │
│ │ Decisions, Chapters (everywhere) │
│ Undo │ Not available — changes save instantly │
├────────────────────────────────────────────────────────────┤
│ VERSION HISTORY │
├────────────────────────────────────────────────────────────┤
│ Badge location │ Tab bar, between tabs and export btns │
│ Badge appears │ Only after ≥1 regeneration │
│ Open picker │ Click "v3 ▾" │
│ View old ver. │ Click any version in the picker │
│ Return latest │ "View latest →" in amber banner │
│ Live baseline │ "Live (original)" — Transcript tab only │
│ Versions are │ Read-only snapshots │
└────────────────────────────────────────────────────────────┘
Last updated: 2026-03-20
Published: