fix(tts): make stepfun preset pick case-stable and per-character

- Hash the lowercased description (matching the case-insensitive scoring)
  so the same archetype text picks the same preset regardless of case.
- Thread the character name through provisionVoice -> stepfunProvision as
  the hash salt, so two characters that share archetype keywords spread
  across the top-N candidate presets instead of collapsing on one voice.

Xiaomi path is unaffected (voicedesign mints a unique clip per call).

app/play/page.tsx

@@ -525,7 +525,7 @@ async function resolveByoVoice(
     return ready;
   }
   if (!speaker.voiceDescription) return null;
-  const p = provisionVoice(cfg, speaker.voiceDescription);
+  const p = provisionVoice(cfg, speaker.voiceDescription, speaker.name);
   cache.set(speaker.name, p);
   try {
     return await p;

lib/engine/agents/characterDesigner.ts

@@ -103,7 +103,7 @@ export async function provisionCharacterVoice(
 ): Promise<CharacterVoice | undefined> {
   if (!config.tts) return undefined;
   try {
-    return await provisionVoice(config.tts, voiceDescription);
+    return await provisionVoice(config.tts, voiceDescription, charName);
   } catch (err) {
     const msg = err instanceof Error ? err.message : String(err);
     console.error(`[characterDesigner] voice provision failed for ${charName}: ${msg}`);

lib/tts-client/index.ts

@@ -13,9 +13,14 @@ function isStepfun(cfg: TtsConfig): boolean {
 export async function provisionVoice(
   cfg: TtsConfig,
   description: string,
+  // Optional per-character salt (typically the character name). Only
+  // StepFun's preset-picker uses it — Xiaomi voicedesign mints a unique
+  // clip per call regardless. Threading it through keeps the API uniform
+  // and prevents archetype collisions on the StepFun path.
+  salt?: string,
 ): Promise<CharacterVoice> {
   return isStepfun(cfg)
-    ? stepfunProvision(cfg, description)
+    ? stepfunProvision(cfg, description, salt)
     : xiaomiProvision(cfg, description);
 }
 

lib/tts-client/stepfun.ts

@@ -115,19 +115,24 @@ export function pickStepfunVoiceId(description: string, salt = ""): string {
 
   // Pick from the top 3 (or fewer) deterministically by hashing the
   // description + an optional salt (charName) so two characters that share
-  // archetype keywords don't collapse onto the identical preset.
+  // archetype keywords don't collapse onto the identical preset. Hash the
+  // lowercased desc so case differences in the same description don't pick
+  // different presets (scoring above is already case-insensitive).
   const top = scored.slice(0, Math.min(3, scored.length));
-  const idx = hashStr(description + "|" + salt) % top.length;
+  const idx = hashStr(desc + "|" + salt.toLowerCase()) % top.length;
   return top[idx]!.v.id;
 }
 
 // Provision is synchronous / no network — StepFun has no voicedesign equivalent.
 // We mirror xiaomiProvision's async signature so the router stays uniform.
+// The optional `salt` (character name) spreads two characters that share
+// archetype keywords across the top-N candidate presets.
 export async function stepfunProvision(
   cfg: TtsConfig,
   description: string,
+  salt?: string,
 ): Promise<CharacterVoice> {
-  const voiceId = pickStepfunVoiceId(description);
+  const voiceId = pickStepfunVoiceId(description, salt);
   return {
     provider: "stepfun",
     voiceId,