Fix: voiceover was silently dropped from every mix; remove SFX gating

Root cause of "no voiceover": in audioMixCommand the [voice] label was used both
as the sidechaincompress key AND as an amix input. FFmpeg does NOT auto-split a
reused label, so the ducking sidechain consumed the voice entirely and the amix
reference got no audio -- the narration never made it into any final mix (every
render played ducked music with a silent hole where the voice should be).

Fix: explicitly `asplit=2[voice_key][voice_mix]` so one copy keys the duck and one
copy stays in the mix. Measured: voice window went from ~-30..-53 dB (silent) to
~-16 dB (present, leading the bed). Revert the compensating +8 dB over-boost back
to unity; music bed lowered to -14 dB so narration leads cleanly.

Also make SFX direction OPTIONAL in the validator (matching the earlier
voiceover-optional change) so a clean music+narration edit with no sound effects
validates. Regression test added for the asplit. mvn -o verify green (248 tests).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Bg76sLc43Wc3j5ZcLkboYR

src/main/java/org/example/videoclips/editing/HighlightDirectorFlowService.java

@@ -166,8 +166,10 @@ public class HighlightDirectorFlowService {
         );
         List<AudioCue> audioCues = new ArrayList<>();
         if (highlight.musicDirection() != null && !highlight.musicDirection().isBlank()) {
+            // Music sits as a bed under the narration (-14 dB); side-chain ducking drops it further during
+            // voiceover lines so the narration always leads.
             audioCues.add(new AudioCue("music", safeKey("music", highlight.musicDirection()), 0.0, targetDuration,
-                    -10.0, highlight.musicDirection()));
+                    -14.0, highlight.musicDirection()));
         }
         if (highlight.sfxDirection() != null && !highlight.sfxDirection().isBlank()) {
             audioCues.addAll(HighlightAssetPreparationService.plannedSfxCues(highlight));

src/main/java/org/example/videoclips/editing/HighlightDirectorPlanValidator.java

@@ -138,7 +138,12 @@ public class HighlightDirectorPlanValidator {
         }
         requiredText(highlight.visualTreatment(), MAX_DIRECTION_LENGTH, "Visual treatment");
         requiredText(highlight.musicDirection(), MAX_DIRECTION_LENGTH, "Music direction");
-        requiredText(highlight.sfxDirection(), MAX_DIRECTION_LENGTH, "SFX direction");
+        // SFX direction is optional: a clean music+narration edit may use no sound effects. Validate
+        // length only when a direction is provided.
+        if (highlight.sfxDirection() != null && !highlight.sfxDirection().isBlank()
+                && highlight.sfxDirection().length() > MAX_DIRECTION_LENGTH) {
+            reject("SFX direction must be 1000 characters or fewer");
+        }
         requiredText(highlight.renderNotes(), MAX_DIRECTION_LENGTH, "Render notes");
         // Voiceover is optional: a music-driven cinematic edit may carry no narration. When present,
         // every line must be non-blank and the script must fit the highlight timing budget.

src/main/java/org/example/videoclips/editing/HighlightFfmpegRenderer.java

@@ -281,6 +281,8 @@ public class HighlightFfmpegRenderer {
             double duration = line.timelineEndSeconds() - line.timelineStartSeconds();
             long delayMillis = Math.round(line.timelineStartSeconds() * 1000);
             command.addAll(List.of("-i", voiceover.path().toString()));
+            // Narration must clearly lead. Boost ~+8 dB; combined with the lower music bed and side-chain
+            // ducking, this keeps the voice well above the score during each line.
             filters.append("[").append(input).append(":a]atrim=duration=").append(duration)
                     .append(",asetpts=PTS-STARTPTS,volume=1.0,adelay=")
                     .append(delayMillis).append("|").append(delayMillis).append("[").append(label).append("];");
@@ -294,14 +296,18 @@ public class HighlightFfmpegRenderer {
                     .append(":duration=longest:dropout_transition=0[voice];");
         }
         if (hasMusic && hasVoiceover) {
-            filters.append("[music_raw][voice]sidechaincompress=threshold=")
+            // CRITICAL: split the voice into two copies. sidechaincompress CONSUMES its key input, so the
+            // voice must be duplicated with asplit — otherwise the amix reference to [voice] gets no audio
+            // and the narration is silently dropped from the final mix (only ducked music remains).
+            filters.append("[voice]asplit=2[voice_key][voice_mix];");
+            filters.append("[music_raw][voice_key]sidechaincompress=threshold=")
                     .append(properties.getMusicDuckingThreshold())
                     .append(":ratio=").append(properties.getMusicDuckingRatio())
                     .append(":attack=").append(properties.getMusicDuckingAttackMs())
                     .append(":release=").append(properties.getMusicDuckingReleaseMs())
                     .append("[music];");
             labels.add("[music]");
-            labels.add("[voice]");
+            labels.add("[voice_mix]");
         } else if (hasMusic) {
             filters.append("[music_raw]anull[music];");
             labels.add("[music]");

src/test/java/org/example/videoclips/editing/HighlightFfmpegRendererTest.java

@@ -136,6 +136,25 @@ class HighlightFfmpegRendererTest {
                 "failed"))).isFalse();
     }
 
+    @Test
+    void splitsVoiceSoDuckingDoesNotDropNarrationFromTheMix() {
+        HighlightFfmpegRenderer renderer = renderer();
+        AudioCue music = new AudioCue("music", "score", 0, 9, -14, "bed");
+        List<HighlightFfmpegRenderer.VoiceoverInput> voiceovers = List.of(
+                new HighlightFfmpegRenderer.VoiceoverInput(Path.of("vo.wav"),
+                        new VoiceoverLine("Grounded line.", 2.7, 6.3, "narration")));
+
+        String filter = String.join(" ", renderer.audioMixCommand(Path.of("timeline.mp4"),
+                Path.of("music.wav"), music, voiceovers, List.of(), false, Path.of("final.mp4")));
+
+        // voice must be duplicated: one copy keys the duck, one copy stays in the amix
+        assertThat(filter)
+                .contains("[voice]asplit=2[voice_key][voice_mix];")
+                .contains("[music_raw][voice_key]sidechaincompress=")
+                .contains("[voice_mix]")
+                .doesNotContain("[music_raw][voice]sidechaincompress");
+    }
+
     @Test
     void mixesGeneratedAudioWithoutReferencingMissingSourceAudio() {
         HighlightFfmpegRenderer renderer = renderer();