use-voice-control 0.1.95 → 0.1.97

package/dist/client.js

@@ -1,276 +1,3 @@
-var k = Object.defineProperty;
-var U = (r, t, e) => t in r ? k(r, t, { enumerable: !0, configurable: !0, writable: !0, value: e }) : r[t] = e;
-var a = (r, t, e) => U(r, typeof t != "symbol" ? t + "" : t, e);
-import { looksLikeMarkdown as E, markdownToSpeech as L } from "./markdown.js";
-import { markdownToSpeechSegments as D, stripInlineMarkdown as N } from "./markdown.js";
-import { s as C } from "./semantic-split-CXhk-k1F.js";
-const T = "/api/speech/tts", v = "af_heart", R = 240;
-function w(r) {
-  return r.aborted ? Promise.resolve() : new Promise((t) => {
-    r.addEventListener("abort", () => t(), { once: !0 });
-  });
-}
-class O {
-  constructor(t = {}) {
-    a(this, "options");
-    a(this, "state", "idle");
-    a(this, "abortController", null);
-    a(this, "audio", null);
-    a(this, "objectUrl", null);
-    /** Set once the endpoint has proved unreachable, so later chunks skip the retry. */
-    a(this, "endpointUnavailable", !1);
-    this.options = t;
-  }
-  /** Replace the options (voice, callbacks, …) without discarding playback state. */
-  setOptions(t) {
-    this.options = t;
-  }
-  getState() {
-    return this.state;
-  }
-  isActive() {
-    return this.state === "speaking" || this.state === "paused" || this.state === "loading";
-  }
-  /**
-   * Speak `text`, cancelling anything already playing. Resolves when playback
-   * finishes or is stopped — it never rejects; failures go to `onError`.
-   */
-  async speak(t) {
-    var o, h, c, f, u, g, d, y;
-    this.stop();
-    const e = this.options.maxChunkLength ?? R, i = C(this.toSpeakableText(t ?? ""), e).map((l) => l.trim()).filter((l) => l.length > 0);
-    if (i.length === 0) return;
-    const s = new AbortController();
-    this.abortController = s;
-    const { signal: n } = s;
-    this.setState("loading");
-    try {
-      let l = this.synthesize(i[0], n);
-      for (let p = 0; p < i.length; p += 1) {
-        const S = l;
-        l = p + 1 < i.length ? this.synthesize(i[p + 1], n) : Promise.resolve(null);
-        const b = await S;
-        if (n.aborted || (this.setState("speaking"), (h = (o = this.options).onChunk) == null || h.call(o, { text: i[p], index: p, total: i.length }), n.aborted) || (await this.playChunk(b, i[p], n), n.aborted)) return;
-      }
-      this.cleanup(), this.setState("idle"), (f = (c = this.options).onEnd) == null || f.call(c, "finished");
-    } catch (l) {
-      if (n.aborted) return;
-      this.cleanup(), this.setState("idle"), (g = (u = this.options).onError) == null || g.call(
-        u,
-        l instanceof Error ? l : new Error(String(l))
-      ), (y = (d = this.options).onEnd) == null || y.call(d, "stopped");
-    } finally {
-      this.abortController === s && (this.abortController = null);
-    }
-  }
-  pause() {
-    this.state === "speaking" && (this.audio ? this.audio.pause() : typeof speechSynthesis < "u" && speechSynthesis.pause(), this.setState("paused"));
-  }
-  resume() {
-    this.state === "paused" && (this.audio ? this.audio.play().catch(() => {
-    }) : typeof speechSynthesis < "u" && speechSynthesis.resume(), this.setState("speaking"));
-  }
-  /** Stop playback and drop any queued chunks. Safe to call when already idle. */
-  stop() {
-    var e, i, s;
-    const t = this.isActive();
-    (e = this.abortController) == null || e.abort(), this.abortController = null, this.cleanup(), t && (this.setState("idle"), (s = (i = this.options).onEnd) == null || s.call(i, "stopped"));
-  }
-  /**
-   * Converts Markdown to spoken words before chunking, so a document read out
-   * of an editor does not have its syntax read back to the listener.
-   */
-  toSpeakableText(t) {
-    const e = this.options.format ?? "auto";
-    return e === "text" ? t : e === "markdown" || E(t) ? L(t, this.options.markdown) : t;
-  }
-  setState(t) {
-    var e, i;
-    this.state !== t && (this.state = t, (i = (e = this.options).onStateChange) == null || i.call(e, t));
-  }
-  cleanup() {
-    this.audio && (this.audio.pause(), this.audio.src = "", this.audio = null), this.objectUrl && (URL.revokeObjectURL(this.objectUrl), this.objectUrl = null), typeof speechSynthesis < "u" && speechSynthesis.cancel();
-  }
-  synthesize(t, e) {
-    const i = this.options.synthesize ? this.options.synthesize(t, e) : this.fetchAudio(t, e);
-    return i.catch(() => {
-    }), i;
-  }
-  async fetchAudio(t, e) {
-    if (this.endpointUnavailable) return null;
-    const i = this.options.endpoint ?? T;
-    try {
-      const s = await fetch(i, {
-        method: "POST",
-        headers: { "Content-Type": "application/json" },
-        body: JSON.stringify({
-          text: t,
-          provider: this.options.provider ?? "kokoro",
-          voice: this.options.voice ?? v
-        }),
-        signal: e
-      });
-      if (!s.ok) throw new Error(`TTS failed: ${s.status}`);
-      const n = s.headers.get("Content-Type") || "audio/wav";
-      return new Blob([await s.arrayBuffer()], { type: n });
-    } catch (s) {
-      if (e.aborted) throw s;
-      return this.endpointUnavailable = !0, null;
-    }
-  }
-  playChunk(t, e, i) {
-    return t ? this.playAudioBlob(t, i) : this.playWithSpeechSynthesis(e, i);
-  }
-  async playAudioBlob(t, e) {
-    const i = URL.createObjectURL(t), s = new Audio(i);
-    this.audio = s, this.objectUrl = i;
-    const n = new Promise((o, h) => {
-      s.addEventListener("ended", () => o(), { once: !0 }), s.addEventListener(
-        "error",
-        () => h(new Error("Audio playback failed")),
-        { once: !0 }
-      );
-    });
-    try {
-      await s.play(), await Promise.race([n, w(e)]);
-    } finally {
-      this.audio === s && (this.audio = null), this.objectUrl === i && (this.objectUrl = null), s.pause(), URL.revokeObjectURL(i);
-    }
-  }
-  async playWithSpeechSynthesis(t, e) {
-    if (typeof speechSynthesis > "u" || typeof SpeechSynthesisUtterance > "u")
-      throw new Error("No speech synthesis available in this browser");
-    const i = new SpeechSynthesisUtterance(t), s = new Promise((n, o) => {
-      i.onend = () => n(), i.onerror = () => e.aborted ? n() : o(new Error("Speech synthesis failed"));
-    });
-    speechSynthesis.speak(i), await Promise.race([s, w(e)]);
-  }
-}
-function m() {
-  return typeof window > "u" ? null : window.SpeechRecognition || window.webkitSpeechRecognition || null;
-}
-function x() {
-  var r;
-  return typeof window > "u" ? !1 : m() ? !0 : !!((r = navigator.mediaDevices) != null && r.getUserMedia);
-}
-class M {
-  constructor(t = {}) {
-    a(this, "options");
-    a(this, "listening", !1);
-    /** Distinguishes a deliberate `stop()` from Chromium's idle auto-stop. */
-    a(this, "stopRequested", !1);
-    a(this, "recognition", null);
-    a(this, "moonshine", null);
-    this.options = t;
-  }
-  setOptions(t) {
-    this.options = t;
-  }
-  isListening() {
-    return this.listening;
-  }
-  async start() {
-    var i, s;
-    if (this.listening) return;
-    this.stopRequested = !1;
-    const t = this.options.engine ?? "auto", e = m();
-    try {
-      if (t !== "moonshine" && e) {
-        this.startWebSpeech(e);
-        return;
-      }
-      if (t === "webspeech")
-        throw new Error("Speech recognition is not available in this browser");
-      await this.startMoonshine();
-    } catch (n) {
-      this.setListening(!1), (s = (i = this.options).onError) == null || s.call(
-        i,
-        n instanceof Error ? n : new Error(String(n))
-      );
-    }
-  }
-  async stop() {
-    var t, e;
-    if (this.stopRequested = !0, this.recognition) {
-      try {
-        this.recognition.stop();
-      } catch {
-      }
-      this.recognition = null;
-    }
-    if (this.moonshine) {
-      try {
-        await ((e = (t = this.moonshine).stop) == null ? void 0 : e.call(t));
-      } catch {
-      }
-      this.moonshine = null;
-    }
-    this.setListening(!1);
-  }
-  async toggle() {
-    this.listening ? await this.stop() : await this.start();
-  }
-  setListening(t) {
-    var e, i;
-    this.listening !== t && (this.listening = t, (i = (e = this.options).onStateChange) == null || i.call(e, t));
-  }
-  startWebSpeech(t) {
-    const e = new t();
-    e.continuous = !0, e.interimResults = !0, e.lang = this.options.language ?? "en-US", e.onstart = () => this.setListening(!0), e.onresult = (i) => {
-      var n, o, h, c, f;
-      let s = "";
-      for (let u = i.resultIndex; u < i.results.length; u += 1) {
-        const g = i.results[u], d = String(((n = g[0]) == null ? void 0 : n.transcript) ?? "").trim();
-        d && (g.isFinal ? (h = (o = this.options).onCommit) == null || h.call(o, d) : s += `${s ? " " : ""}${d}`);
-      }
-      s && ((f = (c = this.options).onPartial) == null || f.call(c, s));
-    }, e.onerror = (i) => {
-      var n, o;
-      const s = i == null ? void 0 : i.error;
-      s === "no-speech" || s === "aborted" || (this.stopRequested = !0, (o = (n = this.options).onError) == null || o.call(n, new Error(`Speech recognition error: ${s}`)));
-    }, e.onend = () => {
-      if (this.stopRequested || this.recognition !== e) {
-        this.recognition = null, this.setListening(!1);
-        return;
-      }
-      try {
-        e.start();
-      } catch {
-        this.recognition = null, this.setListening(!1);
-      }
-    }, this.recognition = e, e.start();
-  }
-  async startMoonshine() {
-    const t = await import("@moonshine-ai/moonshine-js"), e = new t.MicrophoneTranscriber(
-      this.options.model ?? "model/small",
-      {
-        onTranscriptionUpdated: (i) => {
-          var s, n;
-          (n = (s = this.options).onPartial) == null || n.call(s, String(i ?? "").trim());
-        },
-        onTranscriptionCommitted: (i) => {
-          var n, o, h, c;
-          const s = String(i ?? "").trim();
-          s ? (o = (n = this.options).onCommit) == null || o.call(n, s) : (c = (h = this.options).onPartial) == null || c.call(h, "");
-        }
-      },
-      !1
-      // streaming mode
-    );
-    if (this.moonshine = e, await e.start(), this.stopRequested) {
-      await this.stop();
-      return;
-    }
-    this.setListening(!0);
-  }
-}
-export {
-  M as LiveTranscriber,
-  O as ReadAloudController,
-  x as isTranscriptionSupported,
-  E as looksLikeMarkdown,
-  L as markdownToSpeech,
-  D as markdownToSpeechSegments,
-  N as stripInlineMarkdown
-};
-//# sourceMappingURL=client.js.map
+import { looksLikeMarkdown as e, markdownToSpeech as t, markdownToSpeechSegments as n, stripInlineMarkdown as r } from "./markdown.js";
+import { n as i, r as a, t as o } from "./client-CwHmea9T.js";
+export { o as LiveTranscriber, a as ReadAloudController, i as isTranscriptionSupported, e as looksLikeMarkdown, t as markdownToSpeech, n as markdownToSpeechSegments, r as stripInlineMarkdown };

package/dist/index.js

@@ -1,59 +1,49 @@
-import { K as t, s as a, D as c } from "./kokoro-node-DJ_Rxp_N.js";
-import { c as h, d as k, e as w, w as m } from "./kokoro-node-DJ_Rxp_N.js";
-import { looksLikeMarkdown as v, markdownToSpeech as _, markdownToSpeechSegments as y, stripInlineMarkdown as A } from "./markdown.js";
-async function s(r, e = "af_heart", n = {}) {
-  const o = t.includes(e) ? e : "af_heart";
-  return a(r, { ...n, voice: o });
+import { a as e, c as t, i as n, l as r, o as i, s as a, u as o } from "./kokoro-node-CDhCeD6u.js";
+import { looksLikeMarkdown as s, markdownToSpeech as c, markdownToSpeechSegments as l, stripInlineMarkdown as u } from "./markdown.js";
+//#region speech/core/kokoro.ts
+async function d(e, t = "af_heart", i = {}) {
+	let a = r.includes(t) ? t : "af_heart";
+	return n(e, {
+		...i,
+		voice: a
+	});
 }
-function i() {
-  var e, n;
-  const r = globalThis;
-  try {
-    if (typeof r.getCloudflareContext == "function") {
-      const o = r.getCloudflareContext();
-      if ((e = o == null ? void 0 : o.env) != null && e.AI) return o.env.AI;
-    }
-  } catch {
-  }
-  return ((n = r.__env__) == null ? void 0 : n.AI) ?? r.AI;
+//#endregion
+//#region speech/core/deepgram.ts
+function f() {
+	let e = globalThis;
+	try {
+		if (typeof e.getCloudflareContext == "function") {
+			let t = e.getCloudflareContext();
+			if (t?.env?.AI) return t.env.AI;
+		}
+	} catch {}
+	return e.__env__?.AI ?? e.AI;
 }
-async function u(r, e = "angus") {
-  const n = c.includes(e) ? e : "angus", o = i();
-  if (!o)
-    throw new Error("Cloudflare AI binding not available");
-  return {
-    audio: await o.run("@cf/deepgram/aura-1", {
-      text: r.slice(0, 5e3),
-      speaker: n,
-      encoding: "mp3"
-    }),
-    contentType: "audio/mpeg"
-  };
+async function p(e, n = "angus") {
+	let r = t.includes(n) ? n : "angus", i = f();
+	if (!i) throw Error("Cloudflare AI binding not available");
+	return {
+		audio: await i.run("@cf/deepgram/aura-1", {
+			text: e.slice(0, 5e3),
+			speaker: r,
+			encoding: "mp3"
+		}),
+		contentType: "audio/mpeg"
+	};
 }
-async function l(r) {
-  const { text: e, provider: n = "kokoro", voice: o = "af_heart" } = r;
-  if (!e || typeof e != "string" || e.trim().length === 0)
-    throw new Error("Text is required");
-  switch (n) {
-    case "kokoro":
-      return s(e, o);
-    case "deepgram":
-      return u(e, o);
-    default:
-      throw new Error(`Unknown TTS provider: ${n}`);
-  }
+//#endregion
+//#region speech/index.ts
+async function m(e) {
+	let { text: t, provider: n = "kokoro", voice: r = "af_heart" } = e;
+	if (!t || typeof t != "string" || t.trim().length === 0) throw Error("Text is required");
+	switch (n) {
+		case "kokoro": return d(t, r);
+		case "deepgram": return p(t, r);
+		default: throw Error(`Unknown TTS provider: ${n}`);
+	}
 }
-export {
-  c as DEEPGRAM_SPEAKERS,
-  t as KOKORO_VOICES,
-  h as concatSamples,
-  k as describeKokoroVoice,
-  w as encodeWav,
-  l as generateSpeech,
-  v as looksLikeMarkdown,
-  _ as markdownToSpeech,
-  y as markdownToSpeechSegments,
-  A as stripInlineMarkdown,
-  m as wavDurationSeconds
-};
-//# sourceMappingURL=index.js.map
+//#endregion
+export { t as DEEPGRAM_SPEAKERS, r as KOKORO_VOICES, e as concatSamples, o as describeKokoroVoice, i as encodeWav, m as generateSpeech, s as looksLikeMarkdown, c as markdownToSpeech, l as markdownToSpeechSegments, u as stripInlineMarkdown, a as wavDurationSeconds };
+
+//# sourceMappingURL=index.js.map

package/dist/index.js.map

@@ -1 +1 @@
-{"version":3,"file":"index.js","sources":["../speech/core/kokoro.ts","../speech/core/deepgram.ts","../speech/index.ts"],"sourcesContent":["/**\n * @fileoverview Kokoro TTS provider for `generateSpeech`.\n *\n * The work happens in `kokoro-node.ts`, which runs the Kokoro model locally\n * through `kokoro-js`; this file is the thin provider adapter that validates the\n * voice and returns the package's `TTSResult` shape.\n */\nimport type { TTSResult } from \"../types/types\";\nimport { KOKORO_VOICES, type KokoroVoice } from \"../types/types\";\nimport { synthesizeWav, type KokoroNodeOptions } from \"./kokoro-node\";\n\n/**\n * Generate speech from text using Kokoro.\n *\n * Long text is chunked and joined automatically. An unrecognised voice falls\n * back to `af_heart` rather than failing the request.\n */\nexport async function generateKokoroSpeech(\n  text: string,\n  voice: string = \"af_heart\",\n  options: Omit<KokoroNodeOptions, \"voice\"> = {}\n): Promise<TTSResult> {\n  const kokoroVoice = KOKORO_VOICES.includes(voice as KokoroVoice)\n    ? (voice as KokoroVoice)\n    : \"af_heart\";\n\n  return synthesizeWav(text, { ...options, voice: kokoroVoice });\n}\n","/**\n * @fileoverview Deepgram TTS provider implementation using Cloudflare Workers AI\n * Requires Cloudflare AI binding\n */\nimport type { TTSResult } from \"../types/types\";\nimport { DEEPGRAM_SPEAKERS, type DeepgramSpeaker } from \"../types/types\";\n\n/**\n * Resolve the Cloudflare Workers AI binding at runtime without a hard\n * dependency on the host application. Consumers running on Cloudflare can\n * expose the context by setting `globalThis.getCloudflareContext` (as the\n * `@opennextjs/cloudflare` / `@cloudflare/next-on-pages` helpers do) or by\n * placing the bound `env` on `globalThis.__env__`.\n */\nfunction resolveCloudflareAI(): any {\n  const g = globalThis as any;\n  try {\n    if (typeof g.getCloudflareContext === \"function\") {\n      const ctx = g.getCloudflareContext();\n      if (ctx?.env?.AI) return ctx.env.AI;\n    }\n  } catch {\n    // CF context helper threw (e.g. called outside a request scope)\n  }\n  return g.__env__?.AI ?? g.AI;\n}\n\n/**\n * Generate speech from text using Deepgram Aura via Cloudflare Workers AI\n */\nexport async function generateDeepgramSpeech(\n  text: string,\n  speaker: string = \"angus\"\n): Promise<TTSResult> {\n  // Validate speaker\n  const auraVoice = DEEPGRAM_SPEAKERS.includes(speaker as DeepgramSpeaker)\n    ? (speaker as DeepgramSpeaker)\n    : \"angus\";\n\n  const ai = resolveCloudflareAI();\n\n  if (!ai) {\n    throw new Error(\"Cloudflare AI binding not available\");\n  }\n\n  const result = await ai.run(\"@cf/deepgram/aura-1\", {\n    text: text.slice(0, 5000),\n    speaker: auraVoice,\n    encoding: \"mp3\",\n  });\n\n  return {\n    audio: result,\n    contentType: \"audio/mpeg\",\n  };\n}\n","/**\n * @fileoverview Unified text-to-speech API supporting Kokoro (default) and Deepgram\n *\n * Kokoro: Faster, more natural, runs on Node CPU\n * Deepgram: Requires Cloudflare AI binding, MP3 output\n */\nimport type { TTSOptions, TTSResult } from \"./types/types\";\nimport { generateKokoroSpeech } from \"./core/kokoro\";\nimport { generateDeepgramSpeech } from \"./core/deepgram\";\n\nexport * from \"./types/types\";\n\n// Markdown handling is exported from the root entry so a server route can turn a\n// document into speakable text with the same rules the CLI and the browser use.\nexport {\n  looksLikeMarkdown,\n  markdownToSpeech,\n  markdownToSpeechSegments,\n  stripInlineMarkdown,\n  type MarkdownToSpeechOptions,\n  type SpeechSegment,\n  type SpeechSegmentType,\n} from \"./utils/markdown-to-speech\";\n\nexport { encodeWav, concatSamples, wavDurationSeconds } from \"./utils/wav\";\n\n/**\n * Generate speech from text using the specified provider\n *\n * @param options - TTS configuration\n * @returns Audio buffer and content type\n *\n * @example\n * ```ts\n * // Use Kokoro (default)\n * const audio = await generateSpeech({\n *   text: \"Hello world\",\n *   voice: \"af_heart\"\n * });\n *\n * // Use Deepgram\n * const audio = await generateSpeech({\n *   text: \"Hello world\",\n *   provider: \"deepgram\",\n *   voice: \"angus\"\n * });\n * ```\n */\nexport async function generateSpeech(\n  options: TTSOptions\n): Promise<TTSResult> {\n  const { text, provider = \"kokoro\", voice = \"af_heart\" } = options;\n\n  if (!text || typeof text !== \"string\" || text.trim().length === 0) {\n    throw new Error(\"Text is required\");\n  }\n\n  switch (provider) {\n    case \"kokoro\":\n      return generateKokoroSpeech(text, voice);\n\n    case \"deepgram\":\n      return generateDeepgramSpeech(text, voice);\n\n    default:\n      throw new Error(`Unknown TTS provider: ${provider}`);\n  }\n}\n"],"names":["generateKokoroSpeech","text","voice","options","kokoroVoice","KOKORO_VOICES","synthesizeWav","resolveCloudflareAI","g","ctx","_a","_b","generateDeepgramSpeech","speaker","auraVoice","DEEPGRAM_SPEAKERS","ai","generateSpeech","provider"],"mappings":";;;AAiBA,eAAsBA,EACpBC,GACAC,IAAgB,YAChBC,IAA4C,CAAA,GACxB;AACpB,QAAMC,IAAcC,EAAc,SAASH,CAAoB,IAC1DA,IACD;AAEJ,SAAOI,EAAcL,GAAM,EAAE,GAAGE,GAAS,OAAOC,GAAa;AAC/D;ACbA,SAASG,IAA2B;;AAClC,QAAMC,IAAI;AACV,MAAI;AACF,QAAI,OAAOA,EAAE,wBAAyB,YAAY;AAChD,YAAMC,IAAMD,EAAE,qBAAA;AACd,WAAIE,IAAAD,KAAA,gBAAAA,EAAK,QAAL,QAAAC,EAAU,GAAI,QAAOD,EAAI,IAAI;AAAA,IACnC;AAAA,EACF,QAAQ;AAAA,EAER;AACA,WAAOE,IAAAH,EAAE,YAAF,gBAAAG,EAAW,OAAMH,EAAE;AAC5B;AAKA,eAAsBI,EACpBX,GACAY,IAAkB,SACE;AAEpB,QAAMC,IAAYC,EAAkB,SAASF,CAA0B,IAClEA,IACD,SAEEG,IAAKT,EAAA;AAEX,MAAI,CAACS;AACH,UAAM,IAAI,MAAM,qCAAqC;AASvD,SAAO;AAAA,IACL,OAPa,MAAMA,EAAG,IAAI,uBAAuB;AAAA,MACjD,MAAMf,EAAK,MAAM,GAAG,GAAI;AAAA,MACxB,SAASa;AAAA,MACT,UAAU;AAAA,IAAA,CACX;AAAA,IAIC,aAAa;AAAA,EAAA;AAEjB;ACPA,eAAsBG,EACpBd,GACoB;AACpB,QAAM,EAAE,MAAAF,GAAM,UAAAiB,IAAW,UAAU,OAAAhB,IAAQ,eAAeC;AAE1D,MAAI,CAACF,KAAQ,OAAOA,KAAS,YAAYA,EAAK,KAAA,EAAO,WAAW;AAC9D,UAAM,IAAI,MAAM,kBAAkB;AAGpC,UAAQiB,GAAA;AAAA,IACN,KAAK;AACH,aAAOlB,EAAqBC,GAAMC,CAAK;AAAA,IAEzC,KAAK;AACH,aAAOU,EAAuBX,GAAMC,CAAK;AAAA,IAE3C;AACE,YAAM,IAAI,MAAM,yBAAyBgB,CAAQ,EAAE;AAAA,EAAA;AAEzD;"}
+{"version":3,"file":"index.js","names":[],"sources":["../speech/core/kokoro.ts","../speech/core/deepgram.ts","../speech/index.ts"],"sourcesContent":["/**\n * @fileoverview Kokoro TTS provider for `generateSpeech`.\n *\n * The work happens in `kokoro-node.ts`, which runs the Kokoro model locally\n * through `kokoro-js`; this file is the thin provider adapter that validates the\n * voice and returns the package's `TTSResult` shape.\n */\nimport type { TTSResult } from \"../types/types\";\nimport { KOKORO_VOICES, type KokoroVoice } from \"../types/types\";\nimport { synthesizeWav, type KokoroNodeOptions } from \"./kokoro-node\";\n\n/**\n * Generate speech from text using Kokoro.\n *\n * Long text is chunked and joined automatically. An unrecognised voice falls\n * back to `af_heart` rather than failing the request.\n */\nexport async function generateKokoroSpeech(\n  text: string,\n  voice: string = \"af_heart\",\n  options: Omit<KokoroNodeOptions, \"voice\"> = {}\n): Promise<TTSResult> {\n  const kokoroVoice = KOKORO_VOICES.includes(voice as KokoroVoice)\n    ? (voice as KokoroVoice)\n    : \"af_heart\";\n\n  return synthesizeWav(text, { ...options, voice: kokoroVoice });\n}\n","/**\n * @fileoverview Deepgram TTS provider implementation using Cloudflare Workers AI\n * Requires Cloudflare AI binding\n */\nimport type { TTSResult } from \"../types/types\";\nimport { DEEPGRAM_SPEAKERS, type DeepgramSpeaker } from \"../types/types\";\n\n/**\n * Resolve the Cloudflare Workers AI binding at runtime without a hard\n * dependency on the host application. Consumers running on Cloudflare can\n * expose the context by setting `globalThis.getCloudflareContext` (as the\n * `@opennextjs/cloudflare` / `@cloudflare/next-on-pages` helpers do) or by\n * placing the bound `env` on `globalThis.__env__`.\n */\nfunction resolveCloudflareAI(): any {\n  const g = globalThis as any;\n  try {\n    if (typeof g.getCloudflareContext === \"function\") {\n      const ctx = g.getCloudflareContext();\n      if (ctx?.env?.AI) return ctx.env.AI;\n    }\n  } catch {\n    // CF context helper threw (e.g. called outside a request scope)\n  }\n  return g.__env__?.AI ?? g.AI;\n}\n\n/**\n * Generate speech from text using Deepgram Aura via Cloudflare Workers AI\n */\nexport async function generateDeepgramSpeech(\n  text: string,\n  speaker: string = \"angus\"\n): Promise<TTSResult> {\n  // Validate speaker\n  const auraVoice = DEEPGRAM_SPEAKERS.includes(speaker as DeepgramSpeaker)\n    ? (speaker as DeepgramSpeaker)\n    : \"angus\";\n\n  const ai = resolveCloudflareAI();\n\n  if (!ai) {\n    throw new Error(\"Cloudflare AI binding not available\");\n  }\n\n  const result = await ai.run(\"@cf/deepgram/aura-1\", {\n    text: text.slice(0, 5000),\n    speaker: auraVoice,\n    encoding: \"mp3\",\n  });\n\n  return {\n    audio: result,\n    contentType: \"audio/mpeg\",\n  };\n}\n","/**\n * @fileoverview Unified text-to-speech API supporting Kokoro (default) and Deepgram\n *\n * Kokoro: Faster, more natural, runs on Node CPU\n * Deepgram: Requires Cloudflare AI binding, MP3 output\n */\nimport type { TTSOptions, TTSResult } from \"./types/types\";\nimport { generateKokoroSpeech } from \"./core/kokoro\";\nimport { generateDeepgramSpeech } from \"./core/deepgram\";\n\nexport * from \"./types/types\";\n\n// Markdown handling is exported from the root entry so a server route can turn a\n// document into speakable text with the same rules the CLI and the browser use.\nexport {\n  looksLikeMarkdown,\n  markdownToSpeech,\n  markdownToSpeechSegments,\n  stripInlineMarkdown,\n  type MarkdownToSpeechOptions,\n  type SpeechSegment,\n  type SpeechSegmentType,\n} from \"./utils/markdown-to-speech\";\n\nexport { encodeWav, concatSamples, wavDurationSeconds } from \"./utils/wav\";\n\n/**\n * Generate speech from text using the specified provider\n *\n * @param options - TTS configuration\n * @returns Audio buffer and content type\n *\n * @example\n * ```ts\n * // Use Kokoro (default)\n * const audio = await generateSpeech({\n *   text: \"Hello world\",\n *   voice: \"af_heart\"\n * });\n *\n * // Use Deepgram\n * const audio = await generateSpeech({\n *   text: \"Hello world\",\n *   provider: \"deepgram\",\n *   voice: \"angus\"\n * });\n * ```\n */\nexport async function generateSpeech(\n  options: TTSOptions\n): Promise<TTSResult> {\n  const { text, provider = \"kokoro\", voice = \"af_heart\" } = options;\n\n  if (!text || typeof text !== \"string\" || text.trim().length === 0) {\n    throw new Error(\"Text is required\");\n  }\n\n  switch (provider) {\n    case \"kokoro\":\n      return generateKokoroSpeech(text, voice);\n\n    case \"deepgram\":\n      return generateDeepgramSpeech(text, voice);\n\n    default:\n      throw new Error(`Unknown TTS provider: ${provider}`);\n  }\n}\n"],"mappings":";;;AAiBA,eAAsB,EACpB,GACA,IAAgB,YAChB,IAA4C,CAAC,GACzB;CACpB,IAAM,IAAc,EAAc,SAAS,CAAoB,IAC1D,IACD;CAEJ,OAAO,EAAc,GAAM;EAAE,GAAG;EAAS,OAAO;CAAY,CAAC;AAC/D;;;ACbA,SAAS,IAA2B;CAClC,IAAM,IAAI;CACV,IAAI;EACF,IAAI,OAAO,EAAE,wBAAyB,YAAY;GAChD,IAAM,IAAM,EAAE,qBAAqB;GACnC,IAAI,GAAK,KAAK,IAAI,OAAO,EAAI,IAAI;EACnC;CACF,QAAQ,CAER;CACA,OAAO,EAAE,SAAS,MAAM,EAAE;AAC5B;AAKA,eAAsB,EACpB,GACA,IAAkB,SACE;CAEpB,IAAM,IAAY,EAAkB,SAAS,CAA0B,IAClE,IACD,SAEE,IAAK,EAAoB;CAE/B,IAAI,CAAC,GACH,MAAU,MAAM,qCAAqC;CASvD,OAAO;EACL,OAAO,MAPY,EAAG,IAAI,uBAAuB;GACjD,MAAM,EAAK,MAAM,GAAG,GAAI;GACxB,SAAS;GACT,UAAU;EACZ,CAAC;EAIC,aAAa;CACf;AACF;;;ACPA,eAAsB,EACpB,GACoB;CACpB,IAAM,EAAE,SAAM,cAAW,UAAU,WAAQ,eAAe;CAE1D,IAAI,CAAC,KAAQ,OAAO,KAAS,YAAY,EAAK,KAAK,CAAC,CAAC,WAAW,GAC9D,MAAU,MAAM,kBAAkB;CAGpC,QAAQ,GAAR;EACE,KAAK,UACH,OAAO,EAAqB,GAAM,CAAK;EAEzC,KAAK,YACH,OAAO,EAAuB,GAAM,CAAK;EAE3C,SACE,MAAU,MAAM,yBAAyB,GAAU;CACvD;AACF"}

package/dist/kokoro-node-CDhCeD6u.js

@@ -0,0 +1,115 @@
+import { t as e } from "./semantic-split-DF26uo-Z.js";
+//#region speech/types/types.ts
+var t = /* @__PURE__ */ "af_heart.af_alloy.af_aoede.af_bella.af_jessica.af_kore.af_nicole.af_nova.af_river.af_sarah.af_sky.am_adam.am_echo.am_eric.am_fenrir.am_liam.am_michael.am_onyx.am_puck.am_santa.bf_alice.bf_emma.bf_isabella.bf_lily.bm_daniel.bm_fable.bm_george.bm_lewis".split(".");
+function n(e) {
+	let [t, n] = e, r = e.slice(3) || e;
+	return {
+		id: e,
+		name: r.charAt(0).toUpperCase() + r.slice(1),
+		accent: t === "b" ? "British English" : "American English",
+		gender: n === "m" ? "Male" : "Female"
+	};
+}
+var r = [
+	"angus",
+	"asteria",
+	"arcas",
+	"orion",
+	"orpheus",
+	"athena",
+	"luna",
+	"zeus",
+	"perseus",
+	"helios",
+	"hera",
+	"stella"
+], i = 44;
+function a(e, t = 0) {
+	let n = Math.max(e.length - 1, 0) * Math.max(t, 0), r = e.reduce((e, t) => e + t.length, 0) + n, i = new Float32Array(r), a = 0;
+	return e.forEach((n, r) => {
+		i.set(n, a), a += n.length, r < e.length - 1 && (a += Math.max(t, 0));
+	}), i;
+}
+function o(e, t = 24e3, n = 1) {
+	let r = /* @__PURE__ */ new ArrayBuffer(i + e.length * 2), a = new DataView(r), o = (e, t) => {
+		for (let n = 0; n < t.length; n += 1) a.setUint8(e + n, t.charCodeAt(n));
+	}, s = t * n * 2;
+	o(0, "RIFF"), a.setUint32(4, 36 + e.length * 2, !0), o(8, "WAVE"), o(12, "fmt "), a.setUint32(16, 16, !0), a.setUint16(20, 1, !0), a.setUint16(22, n, !0), a.setUint32(24, t, !0), a.setUint32(28, s, !0), a.setUint16(32, n * 2, !0), a.setUint16(34, 16, !0), o(36, "data"), a.setUint32(40, e.length * 2, !0);
+	for (let t = 0; t < e.length; t += 1) {
+		let n = Math.max(-1, Math.min(1, e[t]));
+		a.setInt16(i + t * 2, n < 0 ? n * 32768 : n * 32767, !0);
+	}
+	return r;
+}
+function s(e, t) {
+	return t > 0 ? e.length / t : 0;
+}
+//#endregion
+//#region speech/core/kokoro-node.ts
+var c = "onnx-community/Kokoro-82M-v1.0-ONNX", l = "af_heart", u = "q8", d = "cpu", f = 400, p = 120, m = /* @__PURE__ */ new Map();
+function h() {
+	let e = globalThis;
+	return !!(e.process?.versions?.node || e.Bun || e.Deno);
+}
+async function g(e = {}) {
+	let t = e.model ?? c, n = e.dtype ?? u, r = e.device ?? (h() ? d : "wasm"), i = `${t}|${n}|${r}`, a = m.get(i);
+	if (a) return a;
+	let o = (async () => {
+		let i;
+		try {
+			({KokoroTTS: i} = await import(
+				/* @vite-ignore */
+				"kokoro-js"
+));
+		} catch (e) {
+			throw Error(`Local Kokoro speech needs the optional \`kokoro-js\` package. Install it with \`npm install kokoro-js\`. (import failed: ${e instanceof Error ? e.message : String(e)})`);
+		}
+		return i.from_pretrained(t, {
+			dtype: n,
+			device: r,
+			progress_callback: e.onModelProgress
+		});
+	})();
+	m.set(i, o);
+	try {
+		return await o;
+	} catch (e) {
+		throw m.delete(i), e;
+	}
+}
+function _() {
+	m.clear();
+}
+async function v(t, n = {}) {
+	let r = (t ?? "").trim();
+	if (!r) throw Error("Text is required");
+	let i = e(r, n.maxChunkLength ?? f).map((e) => e.trim()).filter((e) => e.length > 0), o = await g(n), s = n.voice ?? l, c = n.speed ?? 1, u = [], d = 24e3;
+	for (let e = 0; e < i.length; e += 1) {
+		n.onChunk?.({
+			index: e,
+			total: i.length,
+			text: i[e]
+		});
+		let t = await o.generate(i[e], {
+			voice: s,
+			speed: c
+		});
+		u.push(t.audio), d = t.sampling_rate ?? d;
+	}
+	let m = n.gapMs ?? p;
+	return {
+		samples: a(u, Math.round(m / 1e3 * d)),
+		sampleRate: d
+	};
+}
+async function y(e, t = {}) {
+	let { samples: n, sampleRate: r } = await v(e, t);
+	return {
+		audio: o(n, r),
+		contentType: "audio/wav"
+	};
+}
+//#endregion
+export { a, r as c, y as i, t as l, _ as n, o, v as r, s, g as t, n as u };
+
+//# sourceMappingURL=kokoro-node-CDhCeD6u.js.map

package/dist/kokoro-node-CDhCeD6u.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"kokoro-node-CDhCeD6u.js","names":[],"sources":["../speech/types/types.ts","../speech/utils/wav.ts","../speech/core/kokoro-node.ts"],"sourcesContent":["/**\n * @fileoverview Type definitions for text-to-speech providers\n */\n\nexport type TTSProvider = \"kokoro\" | \"deepgram\";\n\nexport interface TTSOptions {\n  text: string;\n  provider?: TTSProvider;\n  voice?: string;\n}\n\nexport interface TTSResult {\n  audio: ArrayBuffer;\n  contentType: string;\n}\n\n// Kokoro voices from the model. The `a`/`b` prefix is the accent (American /\n// British English) and the `f`/`m` that follows it is the voice's gender, which\n// is how `describeKokoroVoice` derives a listing without a second catalog.\nexport const KOKORO_VOICES = [\n  \"af_heart\", \"af_alloy\", \"af_aoede\", \"af_bella\", \"af_jessica\",\n  \"af_kore\", \"af_nicole\", \"af_nova\", \"af_river\", \"af_sarah\", \"af_sky\",\n  \"am_adam\", \"am_echo\", \"am_eric\", \"am_fenrir\", \"am_liam\",\n  \"am_michael\", \"am_onyx\", \"am_puck\", \"am_santa\",\n  \"bf_alice\", \"bf_emma\", \"bf_isabella\", \"bf_lily\",\n  \"bm_daniel\", \"bm_fable\", \"bm_george\", \"bm_lewis\",\n] as const;\n\nexport type KokoroVoice = (typeof KOKORO_VOICES)[number];\n\nexport interface KokoroVoiceDescription {\n  id: string;\n  /** Display name, e.g. `Heart` for `af_heart`. */\n  name: string;\n  /** `American English` or `British English`. */\n  accent: string;\n  gender: \"Female\" | \"Male\";\n}\n\n/**\n * Describes a voice from its id, so listings (`--list-voices`, a voice picker)\n * do not need a hand-maintained table that can drift from `KOKORO_VOICES`.\n */\nexport function describeKokoroVoice(id: string): KokoroVoiceDescription {\n  const [accentCode, genderCode] = id;\n  const suffix = id.slice(3) || id;\n\n  return {\n    id,\n    name: suffix.charAt(0).toUpperCase() + suffix.slice(1),\n    accent: accentCode === \"b\" ? \"British English\" : \"American English\",\n    gender: genderCode === \"m\" ? \"Male\" : \"Female\",\n  };\n}\n\n// Deepgram Aura speakers\nexport const DEEPGRAM_SPEAKERS = [\n  \"angus\", \"asteria\", \"arcas\", \"orion\", \"orpheus\", \"athena\",\n  \"luna\", \"zeus\", \"perseus\", \"helios\", \"hera\", \"stella\",\n] as const;\n\nexport type DeepgramSpeaker = (typeof DEEPGRAM_SPEAKERS)[number];\n","/**\n * @fileoverview Minimal WAV (RIFF) encoder for raw Float32 PCM.\n *\n * The browser path gets its WAV from `convertAudioBufferToWav`, which needs a Web\n * Audio `AudioBuffer`. Node has no such thing: the Kokoro model hands back a bare\n * `Float32Array`, and long documents are synthesized one chunk at a time and\n * joined before encoding. Both of those are pure buffer work, so they live here\n * with no platform dependency.\n */\n\n/** Number of bytes in the fixed RIFF/fmt/data header this encoder writes. */\nconst HEADER_BYTES = 44;\n\n/**\n * Joins per-chunk sample buffers into one contiguous track.\n *\n * @param chunks Sample buffers in playback order.\n * @param silenceSamples Samples of silence to insert between chunks, so the\n *   seams between separately synthesized chunks do not sound clipped.\n */\nexport function concatSamples(\n  chunks: Float32Array[],\n  silenceSamples = 0\n): Float32Array {\n  const gaps = Math.max(chunks.length - 1, 0) * Math.max(silenceSamples, 0);\n  const total = chunks.reduce((sum, chunk) => sum + chunk.length, 0) + gaps;\n  const out = new Float32Array(total);\n\n  let offset = 0;\n  chunks.forEach((chunk, i) => {\n    out.set(chunk, offset);\n    offset += chunk.length;\n    if (i < chunks.length - 1) offset += Math.max(silenceSamples, 0);\n  });\n\n  return out;\n}\n\n/**\n * Encodes Float32 samples (nominally -1..1) as a 16-bit PCM WAV file.\n *\n * @param samples Interleaved samples when `channels` is greater than 1.\n * @param sampleRate Sample rate of the audio, e.g. 24000 for Kokoro.\n * @param channels Channel count. Kokoro is mono.\n */\nexport function encodeWav(\n  samples: Float32Array,\n  sampleRate = 24000,\n  channels = 1\n): ArrayBuffer {\n  const buffer = new ArrayBuffer(HEADER_BYTES + samples.length * 2);\n  const view = new DataView(buffer);\n\n  const writeString = (offset: number, value: string): void => {\n    for (let i = 0; i < value.length; i += 1) {\n      view.setUint8(offset + i, value.charCodeAt(i));\n    }\n  };\n\n  const byteRate = sampleRate * channels * 2;\n\n  writeString(0, \"RIFF\");\n  view.setUint32(4, 36 + samples.length * 2, true);\n  writeString(8, \"WAVE\");\n  writeString(12, \"fmt \");\n  view.setUint32(16, 16, true); // PCM header size\n  view.setUint16(20, 1, true); // format: PCM\n  view.setUint16(22, channels, true);\n  view.setUint32(24, sampleRate, true);\n  view.setUint32(28, byteRate, true);\n  view.setUint16(32, channels * 2, true); // block align\n  view.setUint16(34, 16, true); // bits per sample\n  writeString(36, \"data\");\n  view.setUint32(40, samples.length * 2, true);\n\n  for (let i = 0; i < samples.length; i += 1) {\n    // Clamp before scaling: the model occasionally overshoots 1.0 and wrapping\n    // a 16-bit sample turns that into a loud click.\n    const sample = Math.max(-1, Math.min(1, samples[i]));\n    view.setInt16(HEADER_BYTES + i * 2, sample < 0 ? sample * 0x8000 : sample * 0x7fff, true);\n  }\n\n  return buffer;\n}\n\n/** Duration of a Float32 track in seconds, for progress and log lines. */\nexport function wavDurationSeconds(samples: Float32Array, sampleRate: number): number {\n  return sampleRate > 0 ? samples.length / sampleRate : 0;\n}\n","/**\n * @fileoverview Kokoro TTS running locally on Node (or Bun/Deno) via `kokoro-js`.\n *\n * This is the engine behind `generateSpeech({ provider: \"kokoro\" })` on a server\n * and behind the `use-voice-control` CLI. The model runs on the CPU through\n * onnxruntime, so nothing is sent to a third-party service; the first call\n * downloads the weights (~90 MB at the default `q8`) into the Hugging Face cache\n * and every later call reuses the instance held in this module.\n *\n * Kokoro's context is a few hundred phonemes, so anything longer than a\n * paragraph has to be synthesized in pieces. `synthesizeSamples` chunks on\n * sentence and paragraph boundaries with `splitTextSmart`, runs the chunks in\n * order, and joins the audio with a short silence at each seam.\n */\nimport type { TTSResult } from \"../types/types\";\nimport { splitTextSmart } from \"../utils/semantic-split.js\";\nimport { concatSamples, encodeWav } from \"../utils/wav\";\n\nexport type KokoroDtype = \"fp32\" | \"fp16\" | \"q8\" | \"q4\" | \"q4f16\";\nexport type KokoroDevice = \"wasm\" | \"webgpu\" | \"cpu\";\n\nexport interface KokoroNodeOptions {\n  /** Voice id, e.g. `af_heart`. Default `af_heart`. */\n  voice?: string;\n  /** Speaking rate; 1 is the model's natural pace. Default 1. */\n  speed?: number;\n  /** Hugging Face model id. Default `onnx-community/Kokoro-82M-v1.0-ONNX`. */\n  model?: string;\n  /** Weight quantization. `q8` (default) is the best size/quality trade on CPU. */\n  dtype?: KokoroDtype;\n  /** Execution device. Default `cpu`. */\n  device?: KokoroDevice;\n  /**\n   * Target chunk size in characters. Kokoro's context is limited, so long text\n   * is split before synthesis; 400 leaves headroom for phoneme expansion.\n   */\n  maxChunkLength?: number;\n  /** Silence inserted between chunks, in milliseconds. Default 120. */\n  gapMs?: number;\n  /** Model download/loading progress, forwarded from transformers.js. */\n  onModelProgress?: (progress: unknown) => void;\n  /** Called before each chunk is synthesized, for CLI progress output. */\n  onChunk?: (info: { index: number; total: number; text: string }) => void;\n}\n\nexport interface KokoroAudio {\n  samples: Float32Array;\n  sampleRate: number;\n}\n\nconst DEFAULT_MODEL = \"onnx-community/Kokoro-82M-v1.0-ONNX\";\nconst DEFAULT_VOICE = \"af_heart\";\nconst DEFAULT_DTYPE: KokoroDtype = \"q8\";\nconst DEFAULT_DEVICE: KokoroDevice = \"cpu\";\nconst DEFAULT_MAX_CHUNK = 400;\nconst DEFAULT_GAP_MS = 120;\n\n/** One loaded model per model/dtype/device combination, keyed by those three. */\nconst instances = new Map<string, Promise<any>>();\n\n/** True on Node, Bun and Deno — anywhere `kokoro-js` can load onnxruntime-node. */\nfunction isServerRuntime(): boolean {\n  const g = globalThis as any;\n  return Boolean(g.process?.versions?.node || g.Bun || g.Deno);\n}\n\n/**\n * Loads (and caches) a `kokoro-js` instance.\n *\n * `kokoro-js` is an optional dependency: a browser-only consumer should not have\n * to download onnxruntime-node, so a missing install is reported as an\n * actionable error rather than a module-resolution stack trace.\n */\nexport async function loadKokoroTTS(options: KokoroNodeOptions = {}): Promise<any> {\n  const model = options.model ?? DEFAULT_MODEL;\n  const dtype = options.dtype ?? DEFAULT_DTYPE;\n  const device = options.device ?? (isServerRuntime() ? DEFAULT_DEVICE : \"wasm\");\n  const key = `${model}|${dtype}|${device}`;\n\n  const cached = instances.get(key);\n  if (cached) return cached;\n\n  const loading = (async () => {\n    let KokoroTTS: any;\n    try {\n      // Kept in a variable so bundlers treat this as an optional runtime\n      // dependency rather than something to resolve at build time.\n      const specifier = \"kokoro-js\";\n      ({ KokoroTTS } = await import(/* @vite-ignore */ specifier));\n    } catch (error) {\n      throw new Error(\n        \"Local Kokoro speech needs the optional `kokoro-js` package. \" +\n          \"Install it with `npm install kokoro-js`. \" +\n          `(import failed: ${error instanceof Error ? error.message : String(error)})`\n      );\n    }\n\n    return KokoroTTS.from_pretrained(model, {\n      dtype,\n      device,\n      progress_callback: options.onModelProgress,\n    });\n  })();\n\n  instances.set(key, loading);\n\n  try {\n    return await loading;\n  } catch (error) {\n    // A failed download must not poison every later attempt.\n    instances.delete(key);\n    throw error;\n  }\n}\n\n/** Drops cached model instances. Mainly useful in tests and long-lived workers. */\nexport function resetKokoroCache(): void {\n  instances.clear();\n}\n\n/**\n * Synthesizes `text` to raw samples, chunking anything longer than the model's\n * comfortable context.\n */\nexport async function synthesizeSamples(\n  text: string,\n  options: KokoroNodeOptions = {}\n): Promise<KokoroAudio> {\n  const trimmed = (text ?? \"\").trim();\n  if (!trimmed) throw new Error(\"Text is required\");\n\n  const chunks = splitTextSmart(trimmed, options.maxChunkLength ?? DEFAULT_MAX_CHUNK)\n    .map((chunk: string) => chunk.trim())\n    .filter((chunk: string) => chunk.length > 0);\n\n  const tts = await loadKokoroTTS(options);\n  const voice = options.voice ?? DEFAULT_VOICE;\n  const speed = options.speed ?? 1;\n\n  const rendered: Float32Array[] = [];\n  let sampleRate = 24000;\n\n  for (let index = 0; index < chunks.length; index += 1) {\n    options.onChunk?.({ index, total: chunks.length, text: chunks[index] });\n    const audio = await tts.generate(chunks[index], { voice, speed });\n    rendered.push(audio.audio);\n    sampleRate = audio.sampling_rate ?? sampleRate;\n  }\n\n  const gapMs = options.gapMs ?? DEFAULT_GAP_MS;\n  return {\n    samples: concatSamples(rendered, Math.round((gapMs / 1000) * sampleRate)),\n    sampleRate,\n  };\n}\n\n/** Synthesizes `text` and encodes it as a 16-bit PCM WAV file. */\nexport async function synthesizeWav(\n  text: string,\n  options: KokoroNodeOptions = {}\n): Promise<TTSResult> {\n  const { samples, sampleRate } = await synthesizeSamples(text, options);\n  return { audio: encodeWav(samples, sampleRate), contentType: \"audio/wav\" };\n}\n"],"mappings":";;AAoBA,IAAa,IAAgB,sRAO7B;AAiBA,SAAgB,EAAoB,GAAoC;CACtE,IAAM,CAAC,GAAY,KAAc,GAC3B,IAAS,EAAG,MAAM,CAAC,KAAK;CAE9B,OAAO;EACL;EACA,MAAM,EAAO,OAAO,CAAC,CAAC,CAAC,YAAY,IAAI,EAAO,MAAM,CAAC;EACrD,QAAQ,MAAe,MAAM,oBAAoB;EACjD,QAAQ,MAAe,MAAM,SAAS;CACxC;AACF;AAGA,IAAa,IAAoB;CAC/B;CAAS;CAAW;CAAS;CAAS;CAAW;CACjD;CAAQ;CAAQ;CAAW;CAAU;CAAQ;AAC/C,GCjDM,IAAe;AASrB,SAAgB,EACd,GACA,IAAiB,GACH;CACd,IAAM,IAAO,KAAK,IAAI,EAAO,SAAS,GAAG,CAAC,IAAI,KAAK,IAAI,GAAgB,CAAC,GAClE,IAAQ,EAAO,QAAQ,GAAK,MAAU,IAAM,EAAM,QAAQ,CAAC,IAAI,GAC/D,IAAM,IAAI,aAAa,CAAK,GAE9B,IAAS;CAOb,OANA,EAAO,SAAS,GAAO,MAAM;EAG3B,AAFA,EAAI,IAAI,GAAO,CAAM,GACrB,KAAU,EAAM,QACZ,IAAI,EAAO,SAAS,MAAG,KAAU,KAAK,IAAI,GAAgB,CAAC;CACjE,CAAC,GAEM;AACT;AASA,SAAgB,EACd,GACA,IAAa,MACb,IAAW,GACE;CACb,IAAM,oBAAS,IAAI,YAAY,IAAe,EAAQ,SAAS,CAAC,GAC1D,IAAO,IAAI,SAAS,CAAM,GAE1B,KAAe,GAAgB,MAAwB;EAC3D,KAAK,IAAI,IAAI,GAAG,IAAI,EAAM,QAAQ,KAAK,GACrC,EAAK,SAAS,IAAS,GAAG,EAAM,WAAW,CAAC,CAAC;CAEjD,GAEM,IAAW,IAAa,IAAW;CAczC,AAZA,EAAY,GAAG,MAAM,GACrB,EAAK,UAAU,GAAG,KAAK,EAAQ,SAAS,GAAG,EAAI,GAC/C,EAAY,GAAG,MAAM,GACrB,EAAY,IAAI,MAAM,GACtB,EAAK,UAAU,IAAI,IAAI,EAAI,GAC3B,EAAK,UAAU,IAAI,GAAG,EAAI,GAC1B,EAAK,UAAU,IAAI,GAAU,EAAI,GACjC,EAAK,UAAU,IAAI,GAAY,EAAI,GACnC,EAAK,UAAU,IAAI,GAAU,EAAI,GACjC,EAAK,UAAU,IAAI,IAAW,GAAG,EAAI,GACrC,EAAK,UAAU,IAAI,IAAI,EAAI,GAC3B,EAAY,IAAI,MAAM,GACtB,EAAK,UAAU,IAAI,EAAQ,SAAS,GAAG,EAAI;CAE3C,KAAK,IAAI,IAAI,GAAG,IAAI,EAAQ,QAAQ,KAAK,GAAG;EAG1C,IAAM,IAAS,KAAK,IAAI,IAAI,KAAK,IAAI,GAAG,EAAQ,EAAE,CAAC;EACnD,EAAK,SAAS,IAAe,IAAI,GAAG,IAAS,IAAI,IAAS,QAAS,IAAS,OAAQ,EAAI;CAC1F;CAEA,OAAO;AACT;AAGA,SAAgB,EAAmB,GAAuB,GAA4B;CACpF,OAAO,IAAa,IAAI,EAAQ,SAAS,IAAa;AACxD;;;ACtCA,IAAM,IAAgB,uCAChB,IAAgB,YAChB,IAA6B,MAC7B,IAA+B,OAC/B,IAAoB,KACpB,IAAiB,KAGjB,oBAAY,IAAI,IAA0B;AAGhD,SAAS,IAA2B;CAClC,IAAM,IAAI;CACV,OAAO,GAAQ,EAAE,SAAS,UAAU,QAAQ,EAAE,OAAO,EAAE;AACzD;AASA,eAAsB,EAAc,IAA6B,CAAC,GAAiB;CACjF,IAAM,IAAQ,EAAQ,SAAS,GACzB,IAAQ,EAAQ,SAAS,GACzB,IAAS,EAAQ,WAAW,EAAgB,IAAI,IAAiB,SACjE,IAAM,GAAG,EAAM,GAAG,EAAM,GAAG,KAE3B,IAAS,EAAU,IAAI,CAAG;CAChC,IAAI,GAAQ,OAAO;CAEnB,IAAM,KAAW,YAAY;EAC3B,IAAI;EACJ,IAAI;GAIF,CAAC,iBAAgB,MAAM;;IAA0B;;EACnD,SAAS,GAAO;GACd,MAAU,MACR,4HAEqB,aAAiB,QAAQ,EAAM,UAAU,OAAO,CAAK,EAAE,EAC9E;EACF;EAEA,OAAO,EAAU,gBAAgB,GAAO;GACtC;GACA;GACA,mBAAmB,EAAQ;EAC7B,CAAC;CACH,EAAA,CAAG;CAEH,EAAU,IAAI,GAAK,CAAO;CAE1B,IAAI;EACF,OAAO,MAAM;CACf,SAAS,GAAO;EAGd,MADA,EAAU,OAAO,CAAG,GACd;CACR;AACF;AAGA,SAAgB,IAAyB;CACvC,EAAU,MAAM;AAClB;AAMA,eAAsB,EACpB,GACA,IAA6B,CAAC,GACR;CACtB,IAAM,KAAW,KAAQ,GAAA,CAAI,KAAK;CAClC,IAAI,CAAC,GAAS,MAAU,MAAM,kBAAkB;CAEhD,IAAM,IAAS,EAAe,GAAS,EAAQ,kBAAkB,CAAiB,CAAC,CAChF,KAAK,MAAkB,EAAM,KAAK,CAAC,CAAC,CACpC,QAAQ,MAAkB,EAAM,SAAS,CAAC,GAEvC,IAAM,MAAM,EAAc,CAAO,GACjC,IAAQ,EAAQ,SAAS,GACzB,IAAQ,EAAQ,SAAS,GAEzB,IAA2B,CAAC,GAC9B,IAAa;CAEjB,KAAK,IAAI,IAAQ,GAAG,IAAQ,EAAO,QAAQ,KAAS,GAAG;EACrD,EAAQ,UAAU;GAAE;GAAO,OAAO,EAAO;GAAQ,MAAM,EAAO;EAAO,CAAC;EACtE,IAAM,IAAQ,MAAM,EAAI,SAAS,EAAO,IAAQ;GAAE;GAAO;EAAM,CAAC;EAEhE,AADA,EAAS,KAAK,EAAM,KAAK,GACzB,IAAa,EAAM,iBAAiB;CACtC;CAEA,IAAM,IAAQ,EAAQ,SAAS;CAC/B,OAAO;EACL,SAAS,EAAc,GAAU,KAAK,MAAO,IAAQ,MAAQ,CAAU,CAAC;EACxE;CACF;AACF;AAGA,eAAsB,EACpB,GACA,IAA6B,CAAC,GACV;CACpB,IAAM,EAAE,YAAS,kBAAe,MAAM,EAAkB,GAAM,CAAO;CACrE,OAAO;EAAE,OAAO,EAAU,GAAS,CAAU;EAAG,aAAa;CAAY;AAC3E"}