feat(daemon): add language boost support for Minimax TTS (#773)

* feat(daemon): add language boost support for Minimax TTS

Add --language CLI flag to support language boost parameter for Minimax TTS.
This enables better pronunciation for specific languages like Cantonese (Yue).

* docs(media): add --language flag to media generation contract

Document the language boost parameter for Minimax TTS, enabling
better pronunciation for specific languages like Cantonese (Yue).

* fix(media): correct Cantonese language_boost value and add input validation

- Use correct MiniMax value 'Chinese,Yue' for Cantonese (no space)
- Add type guard in server.ts to reject non-string language values
- Trim language string before sending to MiniMax API

---------

Co-authored-by: root <root@DELLN40.asiacredit.org>

apps/daemon/src/cli.ts

@@ -44,6 +44,7 @@ const MEDIA_GENERATE_STRING_FLAGS = new Set([
   'composition-dir',
   'image',
   'daemon-url',
+  'language',
 ]);
 const MEDIA_GENERATE_BOOLEAN_FLAGS = new Set([
   'help',
@@ -333,6 +334,7 @@ async function runMediaGenerate(rawArgs) {
     audioKind: flags['audio-kind'],
     compositionDir: flags['composition-dir'],
     image: flags.image,
+    language: flags.language,
   };
   if (flags.length != null) body.length = Number(flags.length);
   if (flags.duration != null) body.duration = Number(flags.duration);
@@ -567,6 +569,7 @@ Common options:
   --length <seconds>        Video length.
   --duration <seconds>      Audio duration.
   --voice <voice-id>        Speech / TTS voice.
+  --language <lang>         Language boost for TTS (e.g. Chinese,Yue for Cantonese).
   --audio-kind music|speech|sfx
   --composition-dir <path>  hyperframes-html only — project-relative path
                             to the dir containing hyperframes.json /

apps/daemon/src/media.ts

@@ -211,6 +211,7 @@ function clampWithWarning(value, allowed, flagName) {
  * @param {number} [args.duration]
  * @param {string} [args.voice]
  * @param {string} [args.audioKind]
+ * @param {string} [args.language]
  * @returns {Promise<{ name: string, size: number, mtime: number, kind: string, mime: string, model: string, surface: string, providerNote: string, providerId: string }>}
  */
 export async function generateMedia(args) {
@@ -227,6 +228,7 @@ export async function generateMedia(args) {
     duration,
     voice,
     audioKind,
+    language,
     compositionDir,
     image,
   } = args;
@@ -310,6 +312,7 @@ export async function generateMedia(args) {
     duration: clampedDuration,
     voice: voice || '',
     audioKind: resolvedAudioKind,
+    language: language || '',
     // Project-relative path to the directory the agent scaffolded with
     // hyperframes.json / meta.json / index.html. Only consumed by the
     // hyperframes renderer; null/empty for every other provider.
@@ -1361,10 +1364,13 @@ async function renderMinimaxTTS(ctx, credentials) {
   // platform.minimaxi.com under voice management.
   const voiceId = (ctx.voice && ctx.voice.trim()) || 'male-qn-qingse';
 
+  const languageBoost = typeof ctx.language === 'string' ? ctx.language.trim() : '';
+
   const body = {
     model: wireModel,
     text,
     stream: false,
+    ...(languageBoost ? { language_boost: languageBoost } : {}),
     voice_setting: {
       voice_id: voiceId,
       speed: 1.0,

apps/daemon/src/prompts/media-contract.ts

@@ -87,6 +87,7 @@ Run via your shell tool (Bash on Claude Code, exec on Codex/Gemini, etc.):
   [--duration <seconds>]            # audio only
   [--audio-kind music|speech|sfx]   # audio only
   [--voice <provider-voice-id>]     # audio:speech only; omit to use provider default
+  [--language <lang>]               # audio:speech only; language boost (e.g. Chinese,Yue for Cantonese)
 \`\`\`
 
 Always quote the prompt value. Use \`--prompt "<full prompt>"\` (or the
@@ -255,13 +256,15 @@ substitution. Do not silently fall back.
 ### Workflow rules
 
 1. **Read project metadata first.** The "Project metadata" block above
-   tells you the user's pre-selected model, aspect, length, voice, audio
-   kind, etc. Treat those as authoritative defaults — only override if
-   the user's chat message explicitly contradicts them.
-   For \`minimax-tts\`, \`voice\` must be a valid MiniMax \`voice_id\`
-   (example: \`male-qn-qingse\`). Do not pass natural-language voice
-   descriptions like "warm Mandarin narrator" as \`--voice\`; omit the
-   flag instead unless you have a real id.
+    tells you the user's pre-selected model, aspect, length, voice, audio
+    kind, etc. Treat those as authoritative defaults — only override if
+    the user's chat message explicitly contradicts them.
+    For \`minimax-tts\`, \`voice\` must be a valid MiniMax \`voice_id\`
+    (example: \`male-qn-qingse\`). Do not pass natural-language voice
+    descriptions like "warm Mandarin narrator" as \`--voice\`; omit the
+    flag instead unless you have a real id.
+    \`language\` enables pronunciation boost for specific languages
+    (e.g. \`Chinese,Yue\` for Cantonese, \`Chinese\` for Mandarin).
 2. **One discovery turn before generating.** Even with metadata defaults
    present, restate what you're about to make and ask one targeted
    question if anything is ambiguous (subject, mood, brand, voice). The

apps/daemon/src/server.ts

@@ -3825,6 +3825,7 @@ export async function startServer({ port = 7456, host = process.env.OD_BIND_HOST
             : undefined,
         voice: req.body?.voice,
         audioKind: req.body?.audioKind,
+        language: typeof req.body?.language === 'string' ? req.body.language : undefined,
         compositionDir: req.body?.compositionDir,
         image: req.body?.image,
         onProgress: (line) => appendTaskProgress(task, line),

packages/contracts/src/prompts/media-contract.ts

@@ -30,7 +30,8 @@ Run media generation through the dispatcher:
   [--length <seconds>] \\
   [--duration <seconds>] \\
   [--audio-kind music|speech|sfx] \\
-  [--voice <provider-voice-id>]
+  [--voice <provider-voice-id>] \\
+  [--language <lang>]
 \`\`\`
 
 Always quote the prompt value. Never splice unquoted user text into the