playhead-cli 0.1.1 → 0.1.3

package/dist/mcp/bin.js

@@ -1503,7 +1503,10 @@ var init_schema = __esm({
          */
         audio: z.object({
           narration: z.enum(["none", "tts"]).default("none"),
-          provider: z.enum(["say"]).default("say"),
+          // 'say' — macOS built-in (zero deps, Mac-only, robotic).
+          // 'kokoro' — natural neural voice, offline, cross-platform; needs the optional
+          //   kokoro-js package (npm i kokoro-js). Voices: af_heart (default), am_michael, …
+          provider: z.enum(["say", "kokoro"]).default("say"),
           voice: z.string().optional(),
           rate: z.number().int().min(80).max(400).optional(),
           /** Click sound effects at click/select beats. Defaults on when this block is present. */
@@ -1513,7 +1516,10 @@ var init_schema = __esm({
             file: z.string(),
             gainDb: z.number().min(-40).max(6).default(-19)
           }).strict().optional()
-        }).strict().optional(),
+        }).strict().refine(
+          (a) => a.provider !== "kokoro" || a.voice === void 0 || ["heart", "af_heart", "michael", "am_michael"].includes(a.voice),
+          { message: "kokoro voice must be 'heart' (default) or 'michael'", path: ["voice"] }
+        ).optional(),
         /** Closing card. When set, the video ends on a title-card-styled end card. */
         endCard: z.object({
           title: z.string().min(1),
@@ -4601,12 +4607,59 @@ var SayProvider = class {
     await exec("say", ["-v", this.voice, "-r", String(this.rate), "-f", txt, "-o", outFile]);
   }
 };
+var KokoroProvider = class {
+  constructor(voice, speed) {
+    this.voice = voice;
+    this.speed = speed;
+  }
+  voice;
+  speed;
+  name = "kokoro";
+  tts = null;
+  load() {
+    if (this.tts) return this.tts;
+    this.tts = (async () => {
+      let mod;
+      try {
+        mod = await import("kokoro-js");
+      } catch {
+        throw new Error(
+          "the 'kokoro' voice needs the optional kokoro-js package \u2014 install it with `npm i -g kokoro-js` (or add it to your project). It bundles the neural model runtime (~400MB); the base playhead-cli stays lean."
+        );
+      }
+      return await mod.KokoroTTS.from_pretrained("onnx-community/Kokoro-82M-v1.0-ONNX", {
+        dtype: "q8",
+        device: "cpu"
+      });
+    })();
+    return this.tts;
+  }
+  async synth(text, outFile) {
+    const tts = await this.load();
+    const audio = await tts.generate(text, { voice: this.voice, speed: this.speed });
+    await audio.save(outFile);
+  }
+};
 function providerFor(opts) {
-  const voice = opts.voice ?? "Samantha";
-  const rate = opts.rate ?? 178;
   switch (opts.provider) {
-    case "say":
+    case "say": {
+      const voice = opts.voice ?? "Samantha";
+      const rate = opts.rate ?? 178;
       return { provider: new SayProvider(voice, rate), voice, rate };
+    }
+    case "kokoro": {
+      const KOKORO_VOICES = {
+        heart: "af_heart",
+        af_heart: "af_heart",
+        michael: "am_michael",
+        am_michael: "am_michael"
+      };
+      const requested = opts.voice ?? "heart";
+      const voice = KOKORO_VOICES[requested];
+      if (!voice) throw new Error(`kokoro voice must be 'heart' (default) or 'michael', got '${requested}'`);
+      const rate = opts.rate ?? 100;
+      return { provider: new KokoroProvider(voice, rate / 100), voice, rate };
+    }
     default:
       throw new Error(`unknown TTS provider: ${opts.provider}`);
   }
@@ -4626,10 +4679,11 @@ async function probeDurationMs(file) {
 async function synthesizeNarration(lines, opts, cacheDir) {
   await mkdir2(cacheDir, { recursive: true });
   const { provider, voice, rate } = providerFor(opts);
+  const ext = provider.name === "kokoro" ? "wav" : "aiff";
   const out = /* @__PURE__ */ new Map();
   for (const { stepRef, text } of lines) {
     const key = createHash2("sha256").update(`${provider.name}|${voice}|${rate}|${text}`).digest("hex").slice(0, 16);
-    const file = join5(cacheDir, `${key}.aiff`);
+    const file = join5(cacheDir, `${key}.${ext}`);
     if (!existsSync2(file)) await provider.synth(text, file);
     out.set(stepRef, { stepRef, file, durationMs: await probeDurationMs(file) });
   }