File: GGIA Welcome (Cut Flower Symposium 2026, segment 01)
Date Corrected: August 8, 2026
Webinar Date: July 23, 2026
Series: Cut Flower Symposium
Topic: Trade association welcome: GGIA scope, history, membership
Speakers: Lanie Riner, Executive Director, Georgia Green Industry Association
Recorded introduction: Bethany Harris (narrator)
Moderator: Bethany Harris
Track: Light (Stages 1, 2, 5)
Output: GTBOP_Transcript_2026-07-23_RinerGGIAWelcome.srt
SOURCE VERIFICATION
This segment did not come from a single transcript. Three files were merged by role, and each was read end to end before any correction was made.
| Input | Blocks | Role in the merge |
|---|---|---|
_SourceBaselines/Lanie_parakeet.srt | 50 | Block shape, timing, coverage. Supplies the skeleton |
_SourceBaselines/Lanie_whisperkit.srt | 11 | Text and terminology |
Ingest/Cut Flower Symposium 2026/Lanie Riner.srt | 11 | Speaker names only |
- Skeleton blocks: 50 (Parakeet)
- Corrected blocks: 50 ✓ MATCH CONFIRMED
- Time range: 00:00:00,001 to 00:04:06,300
- Media:
Lanie.mp4, 246.911 s (4:06.911), measured withffprobe - Runtime: ~4 minutes 7 seconds
- File reading: COMPLETE ✓ (all three inputs read in full: 50, 11, and 11 blocks; no truncation)
- Coverage proof:
- Early [~0:27]: Riner opens, “Good morning everyone. Thank you for being here. I’m really excited about this second Cut Flower Symposium.”
- Middle [~2:04]: In 2005 the Georgia Flower Growers Association merged into GGIA to become the floriculture division, and many cut flower growers went out of business during the 2007 drought
- Late [~3:19]: Riner asks Rich to put the GGIA website link and her email address in the chat box
Media Coverage Check
Run against Lanie.mp4 before any correction work, per the standing rule that a transcript with a hole in it looks exactly like a transcript.
| Measure | Value |
|---|---|
| Media duration | 246.911 s |
| Merged SRT covered time | 240.400 s |
| Coverage | 97.37 % |
| Untranscribed gaps over 3 s | none |
| Untranscribed gaps over 1 s | 2 |
| Head before first block | 0.001 s |
| Tail after last block | 0.611 s |
| Block starts out of clock order | none |
Both gaps over one second were checked by reading the text on either side, and both were then cross-checked against the WhisperKit pass, which covers each span inside a single continuous block.
| Gap | Text before | Text after | Verdict |
|---|---|---|---|
| 00:00:57,379 to 00:00:58,380 (1.00 s) | “We have a broad scope of" | "membership for this sector including wholesale growers” | Hesitation. WhisperKit renders the span continuously as “a broad scope of membership for this sector,” so no speech is missing |
| 00:01:15,180 to 00:01:16,280 (1.10 s) | “And we’ve been" | "1990.” | This verdict was wrong. Recorded at Stage 1 as a pause after an abandoned clause, on the strength of WhisperKit rendering “And we’ve been 1990” inside one continuous block. The audio sweep of 2026-08-08 found two spoken words in that span, “around since.” Both engines dropped them, so cross-checking one against the other could not detect it |
No block ends mid-word anywhere in the file. Coverage of 97.37 percent with a 0.6-second tail is normal for a four-minute talk with natural pauses.
One gap verdict did not survive the audio sweep, and the method is worth adjusting. The 1.10-second gap at 00:01:15,180 was cleared as a natural pause because the other engine covered the same span inside one continuous block with no apparent loss. The recording has two words in it. Engine agreement across a gap is weaker evidence than it looks: when both passes drop the same low-energy syllables, the second pass confirms the first pass’s hole rather than filling it. Treat a gap that sits inside an incomplete sentence as unresolved regardless of what the other engine shows, and send it to the audio sweep.
Merge Method
Parakeet supplied the skeleton. All 50 sequence numbers and all 100 timestamp values were copied from Lanie_parakeet.srt and never touched. A programmatic diff confirms every index line and every timestamp line in the corrected file is byte-identical to the Parakeet source.
WhisperKit supplied the text. Its 11 coarse blocks were mapped onto the Parakeet block boundaries. Where the two engines produced different words, WhisperKit won unless the floriculture correction table or the surrounding text proved it wrong.
The GUI export supplied only the two names. Both CLI runs emit Speaker 1 through Speaker 3. Lanie Riner.srt is the only file carrying real names, because the CLI does not read the People voice library.
Word accounting after the merge: 633 words in the corrected file, against 641 in the Parakeet source and 630 in WhisperKit. Every difference from either source is itemized below. Nothing else changed.
Speaker Mapping and Boundary Verification
Parakeet’s three diarization labels were mapped to real names, then every boundary was checked against content rather than accepted from either engine.
| Parakeet label | Blocks | Mapped to | Basis |
|---|---|---|---|
| Speaker 1 | 1 to 3 | Bethany Harris | Recorded bio narration. The GUI export labels the same span Bethany Harris, and the text is a third-person introduction of Riner |
| Speaker 2 | 4 to 45 | Lanie Riner | First person throughout, “my name is Lanie Riner,” “our website and my email address” |
| Speaker 3 | 46 to 50 | split, see below | Parakeet’s boundary is wrong |
Parakeet put the handoff two blocks early. Its Speaker 3 begins at block 46, but blocks 46 and 47 are still Riner: “We hope you reach out and thank you for being here today. Bethany, I’ll give it back to you.” A speaker does not hand the floor to herself. The real change is at block 48, where the reply “Perfect” is followed by “the website that she shared,” a third-person reference to Riner. Blocks 46 and 47 are labeled Lanie Riner; blocks 48 through 50 are labeled Bethany Harris.
The GUI export gets this boundary wrong in the other direction. Its block 10 runs from 00:03:38,620 to 00:04:06,140 under the single label “Lanie Riner” and swallows Harris’s entire closing. That is the label-lag pattern already logged on five of the seven symposium files. On the finer Parakeet skeleton the two turns fall in separate blocks, so no block in the corrected file mixes speakers, and no Stage 5 split is needed here.
One attribution was flagged rather than reassigned, and the flag has since closed in its favour. Block 38 as transcribed opened “Richie, did you have that link that you could maybe put in the chat box for everyone?”, a line addressed to Rich. The content that follows in blocks 38 and 39 is first-person Riner (“our website and my email address”), so the label stood. The flag went into the SRT for a listen, not as a proposed change. The audio sweep of 2026-08-08 confirmed the speaker as Lanie Riner and corrected the name to “Rich” (sweep item L2). The block now reads “Rich, did you have that link…” and the label is unchanged.
The Block Past the End of the Recording
Lanie Riner.srt block 11 and Lanie_whisperkit.srt block 11 both read “Okay.” and both run 00:04:08,450 to 00:04:08,580. The media ends at 246.911 s, so that block starts 1.539 seconds after the recording stops. It cannot be real.
Decision: excluded, and the exclusion is structural rather than an edit. The corrected file is built on the Parakeet skeleton, and Parakeet produced no such block; its last block ends at 246.300 s, inside the media. No block was deleted from the file being corrected. Both engines’ final substantive content ends at the same place, and the only thing lost is a 130-millisecond “Okay” that WhisperKit alone reports.
The finding is worth carrying forward. A block that starts after the recording ends is a hallucination, not a transcription, and the end-of-media test is cheap: compare the last block’s start against the ffprobe duration. It runs in every coverage pass on this series.
Corrections Applied
33 individual corrections across 22 of 50 blocks in the Stage 1 merge, plus 2 post-sweep edits at blocks 15 and 38, for a running total of 35 across 23 blocks. Every one is a recognition, capitalization, or sentence-boundary fix. No content was added, removed, or reworded for style.
Proper Nouns: Speaker Names
- “Laney Reiner” → “Lanie Riner” (Blocks 1, 9)
Both CLI engines and the GUI export give three different spellings across the file: Laney Reiner (Parakeet), Lani Reiner and Lainey Reiner (WhisperKit). The GUI export’s speaker label field carries the correct spelling, Lanie Riner, and it is used throughout.
Institutions and Programs
- “GTIA” → “GGIA” (Blocks 7, 10)
- “But as as GTIA” → “But as GGIA” (Block 18)
- “the Florida Culture Division of GGIA” → “the floriculture division of GGIA” (Block 26)
- “the Center for Urban Act” → “the Center for Urban Agriculture” (Block 42)
- “Southeast Green Conference and trade show” → “Southeast Green Conference and Trade Show” (Block 37)
Parakeet renders the association as GTIA in three places and GGIA in five. WhisperKit has GGIA in all eight of its instances, and the recorded bio in block 1 spells out Georgia Green Industry Association, which settles it.
“Center for Urban Act” is the one correction where both engines produced the identical wrong token. Normally that is evidence the odd word is what was said. It is overridden here because the phrase is not a name, because the recorded bio two minutes earlier says “the UGA Center for Urban Agriculture” in full, and because the floriculture correction table already carries this pattern at high confidence from other segments. Noted rather than flagged.
Capitalization
- “Forestry Program, she’s thrilled” → “Forestry program, she’s thrilled” (Block 3; descriptive noun, not part of the program name)
- “I am the executive director” → “I am the Executive Director” (Block 10; her title, per WhisperKit and the recorded bio)
- “our State Trade Association for Ornamental and Environmental Horticulture” → “our state trade association for ornamental and environmental horticulture” (Block 11; a description of what GGIA is, not a proper name)
- “A lot of y’all may not know” → “a lot of y’all may not know” (Block 25; continues the sentence begun in block 24)
- “So excited about the reemergence” → “so excited about the reemergence” (Block 29; continues “But we are” in block 28)
- “Business growth, all of those things” → “business growth, all of those things” (Block 31; continues the list begun in block 30)
Stutters and Repetitions
WhisperKit’s cleaner reading was taken in each case. Parakeet captured the repetition; the coarser engine did not hear it as two words.
- “membership for for this sector” → “membership for this sector” (Block 13)
- “But as as GTIA” → “But as GGIA” (Block 18)
- “did did you have that link” → “did you have that link” (Block 38)
- “put in the in the chat box” → “put in the chat box” (Block 38)
Sentence Boundaries and Punctuation
Parakeet ends most blocks with a period whether or not the sentence ends there. Where the sentence continues into the next block, the period was replaced with the mark the sentence actually needs.
- “greenhouse growers, retail.” → “greenhouse growers, retail,” (Block 13; the list continues into block 14)
- “federal levels to regulatory efforts to helping” → “federal levels, to regulatory efforts, to helping” (Block 19)
- “actually merged into GGIA.” → “actually merged into GGIA” (Block 25; the clause continues into block 26)
- “And then of course, as many of you may know,” → “And then, of course, as many of you may know,” (Block 27)
- “in this area, and we are excited to partner” → “in this area. And we are excited to partner” (Block 29)
- “plant health, liability, weather” → “plant health, liability, weather,” (Block 30; the list continues into block 31)
- “We would love to have you” → “We would love to have you.” (Block 36)
- “membership and of course education” → “membership and, of course, education” (Block 37)
- “help you with, I would hope you please reach out.” → “help you with. I would hope you please reach out.” (Block 39)
- “before we get started” → “before we get started.” (Block 10)
- “pulling this together extension.” → “pulling this together. Extension,” (Block 42; the sentence continues into block 43)
Technical Terms
- “4H or FFA” → “4-H or FFA” (Block 19)
- “the re-emergence of this industry” → “the reemergence of this industry” (Block 29)
Where Parakeet Was Right and WhisperKit Was Wrong
- Block 17, “a merger between landscape professionals.” WhisperKit heard “escape professionals,” which is already in the floriculture correction table as a known failure. Parakeet’s reading was kept unchanged.
- Blocks 18 and 19. WhisperKit’s blocks 4 and 5 overlap by 140 milliseconds and duplicate the word “to” across the seam (“federal levels, to” then “to regulatory efforts”). Parakeet has it once. The duplicate was not carried into the merge.
- Block 41, “I think it’s a great opportunity.” WhisperKit splits this sentence across its blocks 9 and 10, which capitalizes “Great” mid-sentence. Parakeet keeps it whole.
Merge Decisions Where the Engines Disagreed
Five disagreements were not stutters or terminology, and each was decided on the record.
| Block | Parakeet | WhisperKit | Kept | Reason |
|---|---|---|---|---|
| 24 | ”I know a lot of the” | (smoothed away) | Parakeet | A discrete 1.2-second block sits at 00:02:03,080 with audio in it. WhisperKit’s coarse block reads the passage as one fluent clause and loses the self-correction. Emptying the block is not an option |
| 37 | ”So many of those regulatory efforts" | "Many of those regulatory efforts” | WhisperKit | A single discourse marker at a Parakeet block start, with no other evidence either way. The role split gives WhisperKit the text |
| 40 | ”thank y’all for investing your time" | "thank you all for investing your time” | WhisperKit | Same reason. Riner says “y’all” elsewhere in the file and both engines agree on those instances, so this one is not decided by her speech pattern |
| 48, 49 | ”Perfect.” twice, in two blocks 0.32 s apart | ”Perfect.” once | Parakeet | Two separately timed blocks with a gap between them. A coarse engine collapses an immediate repetition; a fine one does not |
| 50 | ”we had had her drop in the website" | "we have dropped in the website” | WhisperKit, flagged | The engines disagree on who did the dropping. WhisperKit’s reading is coherent with block 38, where Riner asks Rich to post the link. Flagged in the SRT with the Parakeet token. Confirmed against the recording 2026-08-08, sweep item L3; the flag is closed |
Flagged for Verification, All Resolved
Two flags were raised at Stage 1, both in the SRT, both carrying the original ASR token and neither carrying the expected answer. A third item was documented without a flag. All three were resolved against Lanie.mp4 with Rich on 2026-08-08, in the symposium-wide audio verification sweep recorded in GTBOP_AudioSweep_CutFlowerSymposium.md. Nothing was left inaudible, and the SRT now carries zero [VERIFY] markers.
| Sweep item | Blocks | Flag or item | Audio resolution | SRT effect |
|---|---|---|---|---|
| L1 | 15 to 17 | No flag. “And we’ve been” / “1990.” documented as an incomplete sentence | ”And we’ve been around since 1990.” Two words were missing from both engines | Text edit at block 15 |
| L2 | 38 | [VERIFY: speaker, "Richie, did did you have that link"] | ”Rich,” not “Richie.” The speaker is Lanie Riner, so the label was already right | Text edit on the name, flag removed, label unchanged |
| L3 | 50 | [VERIFY: "we had had her drop in"] | WhisperKit’s reading is correct as written | Flag removed, no text change |
On L1 and where the two words were placed. The recovered words sit in the 1.10-second untranscribed gap between block 15, which ends at 00:01:15,180, and block 16, which starts at 00:01:16,280. Bedrock forbids splitting a block or moving a timestamp, so the words had to be attached to one of the two existing blocks. They were appended to block 15, which now reads “And we’ve been around since,” because the gap opens immediately after block 15’s end and “we’ve been around since” is a single prosodic unit. Block 16 is unchanged at “1990.” The words are source-confirmed; only the placement is a judgment, and it is recorded here so a later audit can see it was a decision rather than a drift. The Stage 5 prose reads “And we’ve been around since 1990.” and the ellipsis that marked the hole has been retired.
On L2 and the earlier reasoning. The Stage 1 note read: both engines agree on “Richie,” and agreement between engines is evidence it is what was said. The recording says otherwise. This is the second place in this segment where joint engine agreement was wrong, alongside “Center for Urban Act” at block 42, which was overridden at Stage 1 on in-transcript evidence. Two for two on this file: joint agreement is not proof.
One further item is documented above and carries no SRT flag:
- “Center for Urban Act” (Block 42). Both engines produced the identical truncation. Corrected at Stage 1 on the strength of the full name appearing in block 1 and the correction table. Not raised in the audio sweep and not in doubt.
SRT Format Compliance
✅ All timestamps preserved exactly as the Parakeet source (verified by script diff, 100 of 100 values byte-identical)
✅ All sequence numbers maintained (1 to 50)
✅ Blank lines between segments preserved
✅ No segments merged or split
✅ Block count: 50 skeleton = 50 corrected ✓ (re-verified after the post-sweep edits of 2026-08-08: 100 of 100 index and timestamp lines still byte-identical to the pre-sweep file)
✅ Block starts in clock order throughout; no sort by index was needed on this file
✅ Maximum line length: 42 characters, every line
✅ Zero [VERIFY] markers remain: both closed by the audio sweep of 2026-08-08
⚠️ Line count: 27 of 50 blocks fit the 2-line rule; 23 need three or more lines
On the 2-line rule
The CLI accepts --max-chars-per-line 42 and does not honor it, so every block was rewrapped by hand to a hard 42-character ceiling. That part is met with no exceptions.
The 2-line ceiling is a different matter. Two lines of 42 characters hold about 84 characters, and 23 of these blocks carry more speech than that, up to 203 characters in block 37. Meeting the ceiling would mean splitting blocks, which Bedrock forbids without exception. The 42-character wrap was treated as the hard rule and the line count was allowed to follow the block.
Distribution: 11 blocks at one line, 16 at two, 12 at three, 6 at four, 4 at five, 1 at six.
The speaker label costs 13 to 15 characters of the first line on every block. Dropping the label to speaker changes only would bring the count of blocks over two lines from 23 to 21, which does not justify the loss of per-block attribution. Labels are on every block, as in all three source files.
This file is a working text source and the Stage 5 input. Before it is used as a caption track it needs a line-splitting pass in a caption editor, which can break a long block into two timed cues without touching the archive copy.
Total Corrections: 35 across 23 of 50 subtitle blocks. 33 in the Stage 1 merge of 2026-08-08, plus 2 post-sweep text edits on the same date
Processing: Complete merge of three source files (50 + 11 + 11 blocks), all read in full
New Patterns for the Cut Flower Correction Reference
| ASR output | Correct form | Engine | Category |
|---|---|---|---|
| Laney Reiner | Lanie Riner | Parakeet | Speaker names |
| GTIA | GGIA | Parakeet | Organizations |
| Florida Culture Division | floriculture division | Parakeet | Organizations |
| Center for Urban Act | Center for Urban Agriculture | both | Institutions |
| escape professionals | landscape professionals | WhisperKit | Industry terms |
Confirmed on this segment and added to the hub README, section 4.
Two engine-shape patterns worth carrying to the next segment. Parakeet ends nearly every block with a period regardless of whether the sentence ends there, so sentence-boundary repair is a per-block job on the whole file rather than an occasional fix. WhisperKit’s overlapping block seams duplicate the word that straddles them, which is what produced the doubled “to” here; check every WhisperKit seam for a repeated word before trusting its text across a boundary.
New Speakers for Roster
- Lanie Riner, Executive Director, Georgia Green Industry Association; co-owner, Thunderwood Farms, a commercial greenhouse in Woodbury, Georgia. Graduate of UGA’s Advancing Georgia’s Leaders in Agriculture and Forestry program. Credentials taken from the recorded introduction narrated by Bethany Harris.
Bethany Harris is already in the project roster and appears here as both moderator and the narrator of the recorded introduction.
Correction summary produced for UGA Center for Urban Agriculture / Cut Flower Symposium Archives
Sources: Lanie_parakeet.srt (50 blocks, skeleton), Lanie_whisperkit.srt (11 blocks, text), Lanie Riner.srt (11 blocks, speaker names)
Media: Lanie.mp4, 246.911 s