@nodaro/shared 1.20.0 → 1.21.0

package/dist/index.cjs

@@ -12189,7 +12189,28 @@ var entitySlotSchema = zod.z.object({
 var audioLayerSchema = zod.z.object({
   mode: zod.z.enum(["speech", "music", "sfx"]),
   content: zod.z.string().min(1),
-  voice: zod.z.string().optional()
+  voice: zod.z.string().optional(),
+  /**
+   * SPEECH ONLY — `slotId` of the on-screen speaker saying these words.
+   *
+   * `voice` casts a voice ("male, proud triumphant shouting"); this says WHO it
+   * belongs to, so a recreation can route the line to the right character
+   * instead of guessing. Usually one person speaks per scene and the guess is
+   * right, which is exactly why the cases with two speakers over one cut fail
+   * silently without this field.
+   *
+   * Optional by design and deliberately NOT refined against `mode` here: the
+   * window schema is the enforced decode grammar, and rejecting a whole roll
+   * because the model tagged a music layer would be a hair-trigger failure. A
+   * speaker on a non-speech layer, or one naming a slot that no longer exists,
+   * is stripped structurally by `dropUnknownSpeakers` — the same
+   * unwrap/drop/sweep philosophy the slot-token and binding channels use.
+   *
+   * An unseen narrator gets NO speaker: a voice with no body is never a slot
+   * (doctrine §5), so attribution here would resurrect the phantom-entity
+   * defect that `stripOrphanSlots` exists to kill.
+   */
+  speakerSlot: zod.z.string().optional()
 });
 var windowSceneBase = zod.z.object({
   startSec: zod.z.number().min(0),
@@ -12278,6 +12299,24 @@ function dropUnknownBindings(sv, validBySlot) {
   }
   return { kept: Object.keys(kept).length > 0 ? kept : void 0, dropped };
 }
+function rewriteSpeakerSlots(audio, slotRenames) {
+  if (!audio.some((a) => a.speakerSlot !== void 0 && slotRenames[a.speakerSlot])) return audio;
+  return audio.map((a) => {
+    const to = a.speakerSlot !== void 0 ? slotRenames[a.speakerSlot] : void 0;
+    return to ? { ...a, speakerSlot: to } : a;
+  });
+}
+function dropUnknownSpeakers(audio, validSlotIds) {
+  const dropped = [];
+  const out = audio.map((a) => {
+    if (a.speakerSlot === void 0) return a;
+    if (a.mode === "speech" && validSlotIds.has(a.speakerSlot)) return a;
+    dropped.push(a.speakerSlot);
+    const { speakerSlot: _drop, ...rest } = a;
+    return rest;
+  });
+  return { audio: dropped.length > 0 ? out : audio, dropped };
+}
 function renderAnalyzedScene(scene, slots, castMap) {
   const byId = new Map(slots.map((s) => [s.slotId, s]));
   return scene.visual.replace(SLOT_TOKEN_RE, (_whole, id) => castMap?.[id] ?? byId.get(id)?.description ?? id);
@@ -12839,6 +12878,7 @@ exports.describeEdgeBehavior = describeEdgeBehavior;
 exports.describeMaskRegion = describeMaskRegion;
 exports.describeSlotControl = describeSlotControl;
 exports.dropUnknownBindings = dropUnknownBindings;
+exports.dropUnknownSpeakers = dropUnknownSpeakers;
 exports.durationsByMode = durationsByMode;
 exports.effectiveReasoningEffort = effectiveReasoningEffort;
 exports.encodeProviderItem = encodeProviderItem;
@@ -13022,6 +13062,7 @@ exports.resolveVideoProviderForMode = resolveVideoProviderForMode;
 exports.resolveXfadeName = resolveXfadeName;
 exports.rewriteSceneBindings = rewriteSceneBindings;
 exports.rewriteSlotTokens = rewriteSlotTokens;
+exports.rewriteSpeakerSlots = rewriteSpeakerSlots;
 exports.rgbaArrayToHex = rgbaArrayToHex;
 exports.roleToPhrase = roleToPhrase;
 exports.runSelector = runSelector;