fix(captions): fix segment grouping, display linger, and export sync

Improve mergeShortSegments to merge bidirectionally with a 16-word cap,
add 1.5s caption linger so text stays visible during speech gaps, and
fix export compositor to apply trimStartSeconds offset for correct
caption timing in trimmed exports.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: jkuri

Reframed/Compositor/CameraVideoCompositor+Captions.swift

@@ -11,11 +11,12 @@ extension CameraVideoCompositor {
   ) {
     guard instruction.captionsEnabled, !instruction.captionSegments.isEmpty else { return }
 
-    let time = CMTimeGetSeconds(compositionTime)
+    let time = CMTimeGetSeconds(compositionTime) + instruction.trimStartSeconds
     guard
-      let segment = instruction.captionSegments.first(where: {
-        time >= $0.startSeconds && time < $0.endSeconds
-      })
+      let segment = captionSegmentAt(
+        time: time,
+        in: instruction.captionSegments
+      )
     else { return }
 
     let displayText = visibleText(
@@ -133,6 +134,30 @@ extension CameraVideoCompositor {
     context.restoreGState()
   }
 
+  static func captionSegmentAt(
+    time: Double,
+    in segments: [CaptionSegment]
+  ) -> CaptionSegment? {
+    if let segment = segments.first(where: {
+      time >= $0.startSeconds && time < $0.endSeconds
+    }) {
+      return segment
+    }
+
+    let maxLinger = 1.5
+    guard
+      let previous = segments.last(where: { $0.endSeconds <= time }),
+      time - previous.endSeconds < maxLinger
+    else { return nil }
+
+    let nextStart = segments.first(where: { $0.startSeconds > time })?.startSeconds
+    if let nextStart, time >= nextStart {
+      return nil
+    }
+
+    return previous
+  }
+
   private static func visibleText(
     for segment: CaptionSegment,
     at time: Double,

Reframed/Editor/EditorState+Captions.swift

@@ -62,7 +62,7 @@ extension EditorState {
   }
 
   func captionAtTime(_ time: Double) -> CaptionSegment? {
-    captionSegments.first { time >= $0.startSeconds && time < $0.endSeconds }
+    CameraVideoCompositor.captionSegmentAt(time: time, in: captionSegments)
   }
 
   func visibleCaptionText(at time: Double) -> String? {

Reframed/Utilities/TranscriptionService.swift

@@ -112,31 +112,43 @@ enum TranscriptionService {
     guard segments.count > 1 else { return segments }
 
     let minWordCount = 4
-    let maxGap = 2.0
+    let maxMergedWordCount = 16
+    let maxGap = 1.5
 
     var merged: [CaptionSegment] = []
     for segment in segments {
-      let wordCount = segment.text.split(separator: " ").count
-      if wordCount < minWordCount,
-        let last = merged.last,
-        segment.startSeconds - last.endSeconds < maxGap
-      {
-        let combinedText = last.text + " " + segment.text
-        let combinedWords: [CaptionWord]? = {
-          guard let lw = last.words, let sw = segment.words else { return last.words ?? segment.words }
-          return lw + sw
-        }()
-        merged[merged.count - 1] = CaptionSegment(
-          id: last.id,
-          startSeconds: last.startSeconds,
-          endSeconds: segment.endSeconds,
-          text: combinedText,
-          words: combinedWords
-        )
+      let currentWordCount = segment.text.split(separator: " ").count
+
+      if let lastIdx = merged.indices.last {
+        let last = merged[lastIdx]
+        let lastWordCount = last.text.split(separator: " ").count
+        let gap = segment.startSeconds - last.endSeconds
+        let canMerge =
+          gap < maxGap && (lastWordCount + currentWordCount) <= maxMergedWordCount
+
+        if canMerge && (currentWordCount < minWordCount || lastWordCount < minWordCount) {
+          let combinedText = last.text + " " + segment.text
+          let combinedWords: [CaptionWord]? = {
+            guard let lw = last.words, let sw = segment.words else {
+              return last.words ?? segment.words
+            }
+            return lw + sw
+          }()
+          merged[lastIdx] = CaptionSegment(
+            id: last.id,
+            startSeconds: last.startSeconds,
+            endSeconds: segment.endSeconds,
+            text: combinedText,
+            words: combinedWords
+          )
+        } else {
+          merged.append(segment)
+        }
       } else {
         merged.append(segment)
       }
     }
+
     return merged
   }