@athenaintel/react 0.23.2 → 0.24.0

package/dist/dictation/dictation-engine.d.ts

@@ -0,0 +1,95 @@
+/**
+ * Speech-to-text engines behind the composer's dictation mode.
+ *
+ * An engine turns microphone audio into a stream of transcript updates. The
+ * SDK ships the browser's own `SpeechRecognition` engine (Chrome, Edge,
+ * Safari), which needs no backend. A host can plug in another engine (for
+ * example a relay to a server-side recognizer) through the `engine` option
+ * of `useDictation` / `<AthenaChat dictation>`; the composer UI is engine
+ * agnostic.
+ */
+export interface DictationTranscriptUpdate {
+    /** Text the engine has committed. Appended to earlier committed text. */
+    final: string;
+    /** Text still being recognized; replaced on the next update. */
+    interim: string;
+}
+export interface DictationEngineCallbacks {
+    onTranscript: (update: DictationTranscriptUpdate) => void;
+    /** The engine stopped on its own (silence timeout, network, permission). */
+    onEnd: () => void;
+    onError: (error: DictationEngineError) => void;
+}
+export interface DictationEngineSession {
+    /** Ask the engine to finish. `onEnd` fires once it has flushed. */
+    stop: () => void;
+    /** Tear the engine down without waiting for a flush. */
+    abort: () => void;
+}
+export interface DictationEngine {
+    /** Stable identifier surfaced in diagnostics (e.g. `web-speech`). */
+    readonly id: string;
+    /**
+     * Whether this engine can run in the current browser. When false the
+     * composer hides the mic instead of rendering a control that cannot work.
+     */
+    isSupported: () => boolean;
+    /**
+     * How long `stop()` may wait for the final transcript before the session
+     * is forced closed. Engines that transcribe after recording need more than
+     * the live default (1.5s).
+     */
+    readonly flushTimeoutMs?: number;
+    /** Status shown while flushing (e.g. `Transcribing…`); defaults to `Finishing…`. */
+    readonly stoppingLabel?: string;
+    /**
+     * Start recognizing. `stream` is the microphone stream the composer already
+     * opened for the waveform; engines that read audio themselves (Web Speech)
+     * ignore it, relay engines encode it.
+     */
+    start: (options: {
+        stream: MediaStream | null;
+        lang: string;
+        callbacks: DictationEngineCallbacks;
+    }) => DictationEngineSession;
+}
+export declare const DICTATION_ENGINE_ERROR_KINDS: {
+    /** The user denied the microphone, or the browser blocked it. */
+    readonly permission: 'permission';
+    /** No speech engine is available in this browser. */
+    readonly unsupported: 'unsupported';
+    /** The engine could not reach its recognizer. */
+    readonly connection: 'connection';
+    /** The engine heard nothing for its silence window. */
+    readonly noSpeech: 'no-speech';
+    readonly unknown: 'unknown';
+};
+export type DictationEngineErrorKind = (typeof DICTATION_ENGINE_ERROR_KINDS)[keyof typeof DICTATION_ENGINE_ERROR_KINDS];
+export interface DictationEngineError {
+    kind: DictationEngineErrorKind;
+    /** One line the composer can show the user. */
+    message: string;
+}
+interface SpeechRecognitionAlternativeLike {
+    transcript: string;
+}
+interface SpeechRecognitionResultLike {
+    isFinal: boolean;
+    readonly length: number;
+    [index: number]: SpeechRecognitionAlternativeLike;
+}
+/** Map a Web Speech `error` code to the engine's error vocabulary. */
+export declare function classifyWebSpeechError(code: string): DictationEngineError;
+/**
+ * Fold one Web Speech result event into `{ final, interim }`. Finalized
+ * alternatives from `resultIndex` onward are joined with a space; the rest is
+ * the interim tail. Pure so it can be tested without a browser.
+ */
+export declare function foldWebSpeechResults(results: ArrayLike<SpeechRecognitionResultLike>, resultIndex: number): DictationTranscriptUpdate;
+/**
+ * The browser's own recognizer. Zero dependencies and zero backend; Chrome
+ * and Edge stream partial results, Safari returns results in chunks, Firefox
+ * has no implementation (`isSupported()` is false there).
+ */
+export declare function createWebSpeechDictationEngine(): DictationEngine;
+export {};

package/dist/dictation/dictation-engine.js

@@ -0,0 +1,128 @@
+const DICTATION_ENGINE_ERROR_KINDS = {
+  /** The user denied the microphone, or the browser blocked it. */
+  permission: "permission",
+  /** No speech engine is available in this browser. */
+  unsupported: "unsupported",
+  /** The engine could not reach its recognizer. */
+  connection: "connection",
+  /** The engine heard nothing for its silence window. */
+  noSpeech: "no-speech",
+  unknown: "unknown"
+};
+function resolveSpeechRecognitionCtor() {
+  if (typeof window === "undefined") return null;
+  const scope = window;
+  return scope.SpeechRecognition ?? scope.webkitSpeechRecognition ?? null;
+}
+function classifyWebSpeechError(code) {
+  switch (code) {
+    case "not-allowed":
+    case "service-not-allowed":
+      return {
+        kind: DICTATION_ENGINE_ERROR_KINDS.permission,
+        message: "Microphone access was denied. Allow the microphone for this site and try again."
+      };
+    case "network":
+      return {
+        kind: DICTATION_ENGINE_ERROR_KINDS.connection,
+        message: "Speech recognition could not connect. Check your connection and try again."
+      };
+    case "no-speech":
+      return {
+        kind: DICTATION_ENGINE_ERROR_KINDS.noSpeech,
+        message: "No speech was heard."
+      };
+    case "audio-capture":
+      return {
+        kind: DICTATION_ENGINE_ERROR_KINDS.permission,
+        message: "No microphone was found."
+      };
+    case "aborted":
+      return { kind: DICTATION_ENGINE_ERROR_KINDS.unknown, message: "Dictation was interrupted." };
+    default:
+      return { kind: DICTATION_ENGINE_ERROR_KINDS.unknown, message: "Dictation stopped unexpectedly." };
+  }
+}
+function foldWebSpeechResults(results, resultIndex) {
+  var _a;
+  let final = "";
+  let interim = "";
+  for (let i = resultIndex; i < results.length; i += 1) {
+    const result = results[i];
+    const transcript = ((_a = result == null ? void 0 : result[0]) == null ? void 0 : _a.transcript) ?? "";
+    if (!transcript) continue;
+    if (result.isFinal) final = final ? `${final} ${transcript.trim()}` : transcript.trim();
+    else interim = interim ? `${interim} ${transcript.trim()}` : transcript.trim();
+  }
+  return { final, interim };
+}
+function createWebSpeechDictationEngine() {
+  return {
+    id: "web-speech",
+    isSupported: () => resolveSpeechRecognitionCtor() !== null,
+    start: ({ lang, callbacks }) => {
+      const Ctor = resolveSpeechRecognitionCtor();
+      if (!Ctor) {
+        callbacks.onError({
+          kind: DICTATION_ENGINE_ERROR_KINDS.unsupported,
+          message: "This browser has no speech recognition."
+        });
+        callbacks.onEnd();
+        return { stop: () => void 0, abort: () => void 0 };
+      }
+      const recognition = new Ctor();
+      recognition.continuous = true;
+      recognition.interimResults = true;
+      recognition.lang = lang;
+      let ended = false;
+      const end = () => {
+        if (ended) return;
+        ended = true;
+        recognition.onresult = null;
+        recognition.onerror = null;
+        recognition.onend = null;
+        callbacks.onEnd();
+      };
+      recognition.onresult = (event) => {
+        callbacks.onTranscript(foldWebSpeechResults(event.results, event.resultIndex));
+      };
+      recognition.onerror = (event) => {
+        if (event.error === "aborted") return;
+        callbacks.onError(classifyWebSpeechError(event.error));
+      };
+      recognition.onend = end;
+      try {
+        recognition.start();
+      } catch (error) {
+        callbacks.onError({
+          kind: DICTATION_ENGINE_ERROR_KINDS.unknown,
+          message: error instanceof Error ? error.message : "Dictation could not start."
+        });
+        end();
+      }
+      return {
+        stop: () => {
+          try {
+            recognition.stop();
+          } catch {
+            end();
+          }
+        },
+        abort: () => {
+          try {
+            recognition.abort();
+          } finally {
+            end();
+          }
+        }
+      };
+    }
+  };
+}
+export {
+  DICTATION_ENGINE_ERROR_KINDS,
+  classifyWebSpeechError,
+  createWebSpeechDictationEngine,
+  foldWebSpeechResults
+};
+//# sourceMappingURL=dictation-engine.js.map

package/dist/dictation/dictation-engine.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"dictation-engine.js","sources":["../../src/dictation/dictation-engine.ts"],"sourcesContent":["/**\n * Speech-to-text engines behind the composer's dictation mode.\n *\n * An engine turns microphone audio into a stream of transcript updates. The\n * SDK ships the browser's own `SpeechRecognition` engine (Chrome, Edge,\n * Safari), which needs no backend. A host can plug in another engine (for\n * example a relay to a server-side recognizer) through the `engine` option\n * of `useDictation` / `<AthenaChat dictation>`; the composer UI is engine\n * agnostic.\n */\n\nexport interface DictationTranscriptUpdate {\n  /** Text the engine has committed. Appended to earlier committed text. */\n  final: string;\n  /** Text still being recognized; replaced on the next update. */\n  interim: string;\n}\n\nexport interface DictationEngineCallbacks {\n  onTranscript: (update: DictationTranscriptUpdate) => void;\n  /** The engine stopped on its own (silence timeout, network, permission). */\n  onEnd: () => void;\n  onError: (error: DictationEngineError) => void;\n}\n\nexport interface DictationEngineSession {\n  /** Ask the engine to finish. `onEnd` fires once it has flushed. */\n  stop: () => void;\n  /** Tear the engine down without waiting for a flush. */\n  abort: () => void;\n}\n\nexport interface DictationEngine {\n  /** Stable identifier surfaced in diagnostics (e.g. `web-speech`). */\n  readonly id: string;\n  /**\n   * Whether this engine can run in the current browser. When false the\n   * composer hides the mic instead of rendering a control that cannot work.\n   */\n  isSupported: () => boolean;\n  /**\n   * How long `stop()` may wait for the final transcript before the session\n   * is forced closed. Engines that transcribe after recording need more than\n   * the live default (1.5s).\n   */\n  readonly flushTimeoutMs?: number;\n  /** Status shown while flushing (e.g. `Transcribing…`); defaults to `Finishing…`. */\n  readonly stoppingLabel?: string;\n  /**\n   * Start recognizing. `stream` is the microphone stream the composer already\n   * opened for the waveform; engines that read audio themselves (Web Speech)\n   * ignore it, relay engines encode it.\n   */\n  start: (options: {\n    stream: MediaStream | null;\n    lang: string;\n    callbacks: DictationEngineCallbacks;\n  }) => DictationEngineSession;\n}\n\nexport const DICTATION_ENGINE_ERROR_KINDS = {\n  /** The user denied the microphone, or the browser blocked it. */\n  permission: 'permission',\n  /** No speech engine is available in this browser. */\n  unsupported: 'unsupported',\n  /** The engine could not reach its recognizer. */\n  connection: 'connection',\n  /** The engine heard nothing for its silence window. */\n  noSpeech: 'no-speech',\n  unknown: 'unknown',\n} as const;\n\nexport type DictationEngineErrorKind =\n  (typeof DICTATION_ENGINE_ERROR_KINDS)[keyof typeof DICTATION_ENGINE_ERROR_KINDS];\n\nexport interface DictationEngineError {\n  kind: DictationEngineErrorKind;\n  /** One line the composer can show the user. */\n  message: string;\n}\n\n// ─── Web Speech API engine ──────────────────────────────────────────\n\n// The Web Speech API has no lib.dom entry in every TypeScript version the SDK\n// supports, and browsers still vendor-prefix the constructor.\ninterface SpeechRecognitionAlternativeLike {\n  transcript: string;\n}\ninterface SpeechRecognitionResultLike {\n  isFinal: boolean;\n  readonly length: number;\n  [index: number]: SpeechRecognitionAlternativeLike;\n}\ninterface SpeechRecognitionEventLike extends Event {\n  resultIndex: number;\n  results: ArrayLike<SpeechRecognitionResultLike>;\n}\ninterface SpeechRecognitionErrorEventLike extends Event {\n  error: string;\n  message?: string;\n}\ninterface SpeechRecognitionLike extends EventTarget {\n  continuous: boolean;\n  interimResults: boolean;\n  lang: string;\n  start: () => void;\n  stop: () => void;\n  abort: () => void;\n  onresult: ((event: SpeechRecognitionEventLike) => void) | null;\n  onerror: ((event: SpeechRecognitionErrorEventLike) => void) | null;\n  onend: (() => void) | null;\n}\ntype SpeechRecognitionCtor = new () => SpeechRecognitionLike;\n\nfunction resolveSpeechRecognitionCtor(): SpeechRecognitionCtor | null {\n  if (typeof window === 'undefined') return null;\n  const scope = window as unknown as {\n    SpeechRecognition?: SpeechRecognitionCtor;\n    webkitSpeechRecognition?: SpeechRecognitionCtor;\n  };\n  return scope.SpeechRecognition ?? scope.webkitSpeechRecognition ?? null;\n}\n\n/** Map a Web Speech `error` code to the engine's error vocabulary. */\nexport function classifyWebSpeechError(code: string): DictationEngineError {\n  switch (code) {\n    case 'not-allowed':\n    case 'service-not-allowed':\n      return {\n        kind: DICTATION_ENGINE_ERROR_KINDS.permission,\n        message: 'Microphone access was denied. Allow the microphone for this site and try again.',\n      };\n    case 'network':\n      return {\n        kind: DICTATION_ENGINE_ERROR_KINDS.connection,\n        message: 'Speech recognition could not connect. Check your connection and try again.',\n      };\n    case 'no-speech':\n      return {\n        kind: DICTATION_ENGINE_ERROR_KINDS.noSpeech,\n        message: 'No speech was heard.',\n      };\n    case 'audio-capture':\n      return {\n        kind: DICTATION_ENGINE_ERROR_KINDS.permission,\n        message: 'No microphone was found.',\n      };\n    case 'aborted':\n      return { kind: DICTATION_ENGINE_ERROR_KINDS.unknown, message: 'Dictation was interrupted.' };\n    default:\n      return { kind: DICTATION_ENGINE_ERROR_KINDS.unknown, message: 'Dictation stopped unexpectedly.' };\n  }\n}\n\n/**\n * Fold one Web Speech result event into `{ final, interim }`. Finalized\n * alternatives from `resultIndex` onward are joined with a space; the rest is\n * the interim tail. Pure so it can be tested without a browser.\n */\nexport function foldWebSpeechResults(\n  results: ArrayLike<SpeechRecognitionResultLike>,\n  resultIndex: number,\n): DictationTranscriptUpdate {\n  let final = '';\n  let interim = '';\n  for (let i = resultIndex; i < results.length; i += 1) {\n    const result = results[i];\n    const transcript = result?.[0]?.transcript ?? '';\n    if (!transcript) continue;\n    if (result.isFinal) final = final ? `${final} ${transcript.trim()}` : transcript.trim();\n    else interim = interim ? `${interim} ${transcript.trim()}` : transcript.trim();\n  }\n  return { final, interim };\n}\n\n/**\n * The browser's own recognizer. Zero dependencies and zero backend; Chrome\n * and Edge stream partial results, Safari returns results in chunks, Firefox\n * has no implementation (`isSupported()` is false there).\n */\nexport function createWebSpeechDictationEngine(): DictationEngine {\n  return {\n    id: 'web-speech',\n    isSupported: () => resolveSpeechRecognitionCtor() !== null,\n    start: ({ lang, callbacks }) => {\n      const Ctor = resolveSpeechRecognitionCtor();\n      if (!Ctor) {\n        callbacks.onError({\n          kind: DICTATION_ENGINE_ERROR_KINDS.unsupported,\n          message: 'This browser has no speech recognition.',\n        });\n        callbacks.onEnd();\n        return { stop: () => undefined, abort: () => undefined };\n      }\n      const recognition = new Ctor();\n      recognition.continuous = true;\n      recognition.interimResults = true;\n      recognition.lang = lang;\n      let ended = false;\n      const end = () => {\n        if (ended) return;\n        ended = true;\n        recognition.onresult = null;\n        recognition.onerror = null;\n        recognition.onend = null;\n        callbacks.onEnd();\n      };\n      recognition.onresult = (event) => {\n        callbacks.onTranscript(foldWebSpeechResults(event.results, event.resultIndex));\n      };\n      recognition.onerror = (event) => {\n        // `aborted` follows our own abort(); the session already ended.\n        if (event.error === 'aborted') return;\n        callbacks.onError(classifyWebSpeechError(event.error));\n      };\n      recognition.onend = end;\n      try {\n        recognition.start();\n      } catch (error) {\n        callbacks.onError({\n          kind: DICTATION_ENGINE_ERROR_KINDS.unknown,\n          message: error instanceof Error ? error.message : 'Dictation could not start.',\n        });\n        end();\n      }\n      return {\n        stop: () => {\n          try {\n            recognition.stop();\n          } catch {\n            end();\n          }\n        },\n        abort: () => {\n          try {\n            recognition.abort();\n          } finally {\n            end();\n          }\n        },\n      };\n    },\n  };\n}\n"],"names":[],"mappings":"AA4DO,MAAM,+BAA+B;AAAA;AAAA,EAE1C,YAAY;AAAA;AAAA,EAEZ,aAAa;AAAA;AAAA,EAEb,YAAY;AAAA;AAAA,EAEZ,UAAU;AAAA,EACV,SAAS;AACX;AA4CA,SAAS,+BAA6D;AACpE,MAAI,OAAO,WAAW,YAAa,QAAO;AAC1C,QAAM,QAAQ;AAId,SAAO,MAAM,qBAAqB,MAAM,2BAA2B;AACrE;AAGO,SAAS,uBAAuB,MAAoC;AACzE,UAAQ,MAAA;AAAA,IACN,KAAK;AAAA,IACL,KAAK;AACH,aAAO;AAAA,QACL,MAAM,6BAA6B;AAAA,QACnC,SAAS;AAAA,MAAA;AAAA,IAEb,KAAK;AACH,aAAO;AAAA,QACL,MAAM,6BAA6B;AAAA,QACnC,SAAS;AAAA,MAAA;AAAA,IAEb,KAAK;AACH,aAAO;AAAA,QACL,MAAM,6BAA6B;AAAA,QACnC,SAAS;AAAA,MAAA;AAAA,IAEb,KAAK;AACH,aAAO;AAAA,QACL,MAAM,6BAA6B;AAAA,QACnC,SAAS;AAAA,MAAA;AAAA,IAEb,KAAK;AACH,aAAO,EAAE,MAAM,6BAA6B,SAAS,SAAS,6BAAA;AAAA,IAChE;AACE,aAAO,EAAE,MAAM,6BAA6B,SAAS,SAAS,kCAAA;AAAA,EAAkC;AAEtG;AAOO,SAAS,qBACd,SACA,aAC2B;AAtGtB;AAuGL,MAAI,QAAQ;AACZ,MAAI,UAAU;AACd,WAAS,IAAI,aAAa,IAAI,QAAQ,QAAQ,KAAK,GAAG;AACpD,UAAM,SAAS,QAAQ,CAAC;AACxB,UAAM,eAAa,sCAAS,OAAT,mBAAa,eAAc;AAC9C,QAAI,CAAC,WAAY;AACjB,QAAI,OAAO,QAAS,SAAQ,QAAQ,GAAG,KAAK,IAAI,WAAW,KAAA,CAAM,KAAK,WAAW,KAAA;AAAA,QAC5E,WAAU,UAAU,GAAG,OAAO,IAAI,WAAW,MAAM,KAAK,WAAW,KAAA;AAAA,EAC1E;AACA,SAAO,EAAE,OAAO,QAAA;AAClB;AAOO,SAAS,iCAAkD;AAChE,SAAO;AAAA,IACL,IAAI;AAAA,IACJ,aAAa,MAAM,6BAAA,MAAmC;AAAA,IACtD,OAAO,CAAC,EAAE,MAAM,gBAAgB;AAC9B,YAAM,OAAO,6BAAA;AACb,UAAI,CAAC,MAAM;AACT,kBAAU,QAAQ;AAAA,UAChB,MAAM,6BAA6B;AAAA,UACnC,SAAS;AAAA,QAAA,CACV;AACD,kBAAU,MAAA;AACV,eAAO,EAAE,MAAM,MAAM,QAAW,OAAO,MAAM,OAAA;AAAA,MAC/C;AACA,YAAM,cAAc,IAAI,KAAA;AACxB,kBAAY,aAAa;AACzB,kBAAY,iBAAiB;AAC7B,kBAAY,OAAO;AACnB,UAAI,QAAQ;AACZ,YAAM,MAAM,MAAM;AAChB,YAAI,MAAO;AACX,gBAAQ;AACR,oBAAY,WAAW;AACvB,oBAAY,UAAU;AACtB,oBAAY,QAAQ;AACpB,kBAAU,MAAA;AAAA,MACZ;AACA,kBAAY,WAAW,CAAC,UAAU;AAChC,kBAAU,aAAa,qBAAqB,MAAM,SAAS,MAAM,WAAW,CAAC;AAAA,MAC/E;AACA,kBAAY,UAAU,CAAC,UAAU;AAE/B,YAAI,MAAM,UAAU,UAAW;AAC/B,kBAAU,QAAQ,uBAAuB,MAAM,KAAK,CAAC;AAAA,MACvD;AACA,kBAAY,QAAQ;AACpB,UAAI;AACF,oBAAY,MAAA;AAAA,MACd,SAAS,OAAO;AACd,kBAAU,QAAQ;AAAA,UAChB,MAAM,6BAA6B;AAAA,UACnC,SAAS,iBAAiB,QAAQ,MAAM,UAAU;AAAA,QAAA,CACnD;AACD,YAAA;AAAA,MACF;AACA,aAAO;AAAA,QACL,MAAM,MAAM;AACV,cAAI;AACF,wBAAY,KAAA;AAAA,UACd,QAAQ;AACN,gBAAA;AAAA,UACF;AAAA,QACF;AAAA,QACA,OAAO,MAAM;AACX,cAAI;AACF,wBAAY,MAAA;AAAA,UACd,UAAA;AACE,gBAAA;AAAA,UACF;AAAA,QACF;AAAA,MAAA;AAAA,IAEJ;AAAA,EAAA;AAEJ;"}

package/dist/dictation/dictation-state.cjs

@@ -0,0 +1,68 @@
+"use strict";
+Object.defineProperty(exports, Symbol.toStringTag, { value: "Module" });
+const DICTATION_STATUSES = {
+  idle: "idle",
+  starting: "starting",
+  listening: "listening",
+  stopping: "stopping",
+  error: "error"
+};
+const INITIAL_DICTATION_STATE = {
+  status: DICTATION_STATUSES.idle,
+  transcript: { final: "", interim: "" },
+  error: null
+};
+function joinSpeech(head, tail) {
+  const left = head.trim();
+  const right = tail.trim();
+  if (!left) return right;
+  if (!right) return left;
+  return `${left} ${right}`;
+}
+function dictationReducer(state, action) {
+  switch (action.type) {
+    case "start":
+      return { ...INITIAL_DICTATION_STATE, status: DICTATION_STATUSES.starting };
+    case "listening":
+      return state.status === DICTATION_STATUSES.starting ? { ...state, status: DICTATION_STATUSES.listening } : state;
+    case "transcript": {
+      if (state.status !== DICTATION_STATUSES.listening && state.status !== DICTATION_STATUSES.stopping) {
+        return state;
+      }
+      return {
+        ...state,
+        transcript: {
+          final: joinSpeech(state.transcript.final, action.update.final),
+          interim: action.update.interim
+        }
+      };
+    }
+    case "stop":
+      return state.status === DICTATION_STATUSES.listening || state.status === DICTATION_STATUSES.starting ? { ...state, status: DICTATION_STATUSES.stopping } : state;
+    case "ended":
+      return {
+        ...state,
+        status: state.status === DICTATION_STATUSES.error ? DICTATION_STATUSES.error : DICTATION_STATUSES.idle,
+        transcript: { final: joinSpeech(state.transcript.final, state.transcript.interim), interim: "" }
+      };
+    case "cancel":
+      return INITIAL_DICTATION_STATE;
+    case "error":
+      return { ...state, status: DICTATION_STATUSES.error, error: action.error };
+    default:
+      return state;
+  }
+}
+function spokenText(transcript) {
+  return joinSpeech(transcript.final, transcript.interim);
+}
+function composeDictationDraft(existingDraft, spoken) {
+  return joinSpeech(existingDraft, spoken);
+}
+exports.DICTATION_STATUSES = DICTATION_STATUSES;
+exports.INITIAL_DICTATION_STATE = INITIAL_DICTATION_STATE;
+exports.composeDictationDraft = composeDictationDraft;
+exports.dictationReducer = dictationReducer;
+exports.joinSpeech = joinSpeech;
+exports.spokenText = spokenText;
+//# sourceMappingURL=dictation-state.cjs.map

package/dist/dictation/dictation-state.cjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"dictation-state.cjs","sources":["../../src/dictation/dictation-state.ts"],"sourcesContent":["import type { DictationEngineError, DictationTranscriptUpdate } from './dictation-engine';\n\n/**\n * The dictation session's state machine, kept pure so the transitions are\n * testable without a microphone.\n *\n *   idle ──start──▶ starting ──engine ready──▶ listening ──stop──▶ stopping ──flushed──▶ idle\n *     ▲                │                          │                     │\n *     └────cancel──────┴──────────────────────────┴─────────────────────┘\n *\n * `error` is a terminal state shown inline until the next start or cancel.\n */\nexport const DICTATION_STATUSES = {\n  idle: 'idle',\n  starting: 'starting',\n  listening: 'listening',\n  stopping: 'stopping',\n  error: 'error',\n} as const;\n\nexport type DictationStatus = (typeof DICTATION_STATUSES)[keyof typeof DICTATION_STATUSES];\n\nexport interface DictationTranscript {\n  /** Everything the engine committed this session, space-joined. */\n  final: string;\n  /** The in-progress tail; replaced on every update. */\n  interim: string;\n}\n\nexport interface DictationState {\n  status: DictationStatus;\n  transcript: DictationTranscript;\n  error: DictationEngineError | null;\n}\n\nexport type DictationAction =\n  | { type: 'start' }\n  | { type: 'listening' }\n  | { type: 'transcript'; update: DictationTranscriptUpdate }\n  | { type: 'stop' }\n  | { type: 'ended' }\n  | { type: 'cancel' }\n  | { type: 'error'; error: DictationEngineError };\n\nexport const INITIAL_DICTATION_STATE: DictationState = {\n  status: DICTATION_STATUSES.idle,\n  transcript: { final: '', interim: '' },\n  error: null,\n};\n\n/** Join two transcript fragments with one space, ignoring empty sides. */\nexport function joinSpeech(head: string, tail: string): string {\n  const left = head.trim();\n  const right = tail.trim();\n  if (!left) return right;\n  if (!right) return left;\n  return `${left} ${right}`;\n}\n\nexport function dictationReducer(state: DictationState, action: DictationAction): DictationState {\n  switch (action.type) {\n    case 'start':\n      return { ...INITIAL_DICTATION_STATE, status: DICTATION_STATUSES.starting };\n    case 'listening':\n      return state.status === DICTATION_STATUSES.starting\n        ? { ...state, status: DICTATION_STATUSES.listening }\n        : state;\n    case 'transcript': {\n      if (state.status !== DICTATION_STATUSES.listening && state.status !== DICTATION_STATUSES.stopping) {\n        return state;\n      }\n      return {\n        ...state,\n        transcript: {\n          final: joinSpeech(state.transcript.final, action.update.final),\n          interim: action.update.interim,\n        },\n      };\n    }\n    case 'stop':\n      return state.status === DICTATION_STATUSES.listening || state.status === DICTATION_STATUSES.starting\n        ? { ...state, status: DICTATION_STATUSES.stopping }\n        : state;\n    case 'ended':\n      // A flush may still have an interim tail; commit it, the engine will\n      // not refine it any further. An error reported before the engine closed\n      // stays visible: only a start or cancel clears it.\n      return {\n        ...state,\n        status: state.status === DICTATION_STATUSES.error ? DICTATION_STATUSES.error : DICTATION_STATUSES.idle,\n        transcript: { final: joinSpeech(state.transcript.final, state.transcript.interim), interim: '' },\n      };\n    case 'cancel':\n      return INITIAL_DICTATION_STATE;\n    case 'error':\n      return { ...state, status: DICTATION_STATUSES.error, error: action.error };\n    default:\n      return state;\n  }\n}\n\n/** The text a dictation session hands to the composer once it ends. */\nexport function spokenText(transcript: DictationTranscript): string {\n  return joinSpeech(transcript.final, transcript.interim);\n}\n\n/**\n * Merge spoken text into the draft that existed before recording started.\n * Manual edits are kept; the speech is appended after one space.\n */\nexport function composeDictationDraft(existingDraft: string, spoken: string): string {\n  return joinSpeech(existingDraft, spoken);\n}\n"],"names":[],"mappings":";;AAYO,MAAM,qBAAqB;AAAA,EAChC,MAAM;AAAA,EACN,UAAU;AAAA,EACV,WAAW;AAAA,EACX,UAAU;AAAA,EACV,OAAO;AACT;AA0BO,MAAM,0BAA0C;AAAA,EACrD,QAAQ,mBAAmB;AAAA,EAC3B,YAAY,EAAE,OAAO,IAAI,SAAS,GAAA;AAAA,EAClC,OAAO;AACT;AAGO,SAAS,WAAW,MAAc,MAAsB;AAC7D,QAAM,OAAO,KAAK,KAAA;AAClB,QAAM,QAAQ,KAAK,KAAA;AACnB,MAAI,CAAC,KAAM,QAAO;AAClB,MAAI,CAAC,MAAO,QAAO;AACnB,SAAO,GAAG,IAAI,IAAI,KAAK;AACzB;AAEO,SAAS,iBAAiB,OAAuB,QAAyC;AAC/F,UAAQ,OAAO,MAAA;AAAA,IACb,KAAK;AACH,aAAO,EAAE,GAAG,yBAAyB,QAAQ,mBAAmB,SAAA;AAAA,IAClE,KAAK;AACH,aAAO,MAAM,WAAW,mBAAmB,WACvC,EAAE,GAAG,OAAO,QAAQ,mBAAmB,UAAA,IACvC;AAAA,IACN,KAAK,cAAc;AACjB,UAAI,MAAM,WAAW,mBAAmB,aAAa,MAAM,WAAW,mBAAmB,UAAU;AACjG,eAAO;AAAA,MACT;AACA,aAAO;AAAA,QACL,GAAG;AAAA,QACH,YAAY;AAAA,UACV,OAAO,WAAW,MAAM,WAAW,OAAO,OAAO,OAAO,KAAK;AAAA,UAC7D,SAAS,OAAO,OAAO;AAAA,QAAA;AAAA,MACzB;AAAA,IAEJ;AAAA,IACA,KAAK;AACH,aAAO,MAAM,WAAW,mBAAmB,aAAa,MAAM,WAAW,mBAAmB,WACxF,EAAE,GAAG,OAAO,QAAQ,mBAAmB,aACvC;AAAA,IACN,KAAK;AAIH,aAAO;AAAA,QACL,GAAG;AAAA,QACH,QAAQ,MAAM,WAAW,mBAAmB,QAAQ,mBAAmB,QAAQ,mBAAmB;AAAA,QAClG,YAAY,EAAE,OAAO,WAAW,MAAM,WAAW,OAAO,MAAM,WAAW,OAAO,GAAG,SAAS,GAAA;AAAA,MAAG;AAAA,IAEnG,KAAK;AACH,aAAO;AAAA,IACT,KAAK;AACH,aAAO,EAAE,GAAG,OAAO,QAAQ,mBAAmB,OAAO,OAAO,OAAO,MAAA;AAAA,IACrE;AACE,aAAO;AAAA,EAAA;AAEb;AAGO,SAAS,WAAW,YAAyC;AAClE,SAAO,WAAW,WAAW,OAAO,WAAW,OAAO;AACxD;AAMO,SAAS,sBAAsB,eAAuB,QAAwB;AACnF,SAAO,WAAW,eAAe,MAAM;AACzC;;;;;;;"}

package/dist/dictation/dictation-state.d.ts

@@ -0,0 +1,58 @@
+import type { DictationEngineError, DictationTranscriptUpdate } from './dictation-engine';
+/**
+ * The dictation session's state machine, kept pure so the transitions are
+ * testable without a microphone.
+ *
+ *   idle ──start──▶ starting ──engine ready──▶ listening ──stop──▶ stopping ──flushed──▶ idle
+ *     ▲                │                          │                     │
+ *     └────cancel──────┴──────────────────────────┴─────────────────────┘
+ *
+ * `error` is a terminal state shown inline until the next start or cancel.
+ */
+export declare const DICTATION_STATUSES: {
+    readonly idle: 'idle';
+    readonly starting: 'starting';
+    readonly listening: 'listening';
+    readonly stopping: 'stopping';
+    readonly error: 'error';
+};
+export type DictationStatus = (typeof DICTATION_STATUSES)[keyof typeof DICTATION_STATUSES];
+export interface DictationTranscript {
+    /** Everything the engine committed this session, space-joined. */
+    final: string;
+    /** The in-progress tail; replaced on every update. */
+    interim: string;
+}
+export interface DictationState {
+    status: DictationStatus;
+    transcript: DictationTranscript;
+    error: DictationEngineError | null;
+}
+export type DictationAction = {
+    type: 'start';
+} | {
+    type: 'listening';
+} | {
+    type: 'transcript';
+    update: DictationTranscriptUpdate;
+} | {
+    type: 'stop';
+} | {
+    type: 'ended';
+} | {
+    type: 'cancel';
+} | {
+    type: 'error';
+    error: DictationEngineError;
+};
+export declare const INITIAL_DICTATION_STATE: DictationState;
+/** Join two transcript fragments with one space, ignoring empty sides. */
+export declare function joinSpeech(head: string, tail: string): string;
+export declare function dictationReducer(state: DictationState, action: DictationAction): DictationState;
+/** The text a dictation session hands to the composer once it ends. */
+export declare function spokenText(transcript: DictationTranscript): string;
+/**
+ * Merge spoken text into the draft that existed before recording started.
+ * Manual edits are kept; the speech is appended after one space.
+ */
+export declare function composeDictationDraft(existingDraft: string, spoken: string): string;

package/dist/dictation/dictation-state.js

@@ -0,0 +1,68 @@
+const DICTATION_STATUSES = {
+  idle: "idle",
+  starting: "starting",
+  listening: "listening",
+  stopping: "stopping",
+  error: "error"
+};
+const INITIAL_DICTATION_STATE = {
+  status: DICTATION_STATUSES.idle,
+  transcript: { final: "", interim: "" },
+  error: null
+};
+function joinSpeech(head, tail) {
+  const left = head.trim();
+  const right = tail.trim();
+  if (!left) return right;
+  if (!right) return left;
+  return `${left} ${right}`;
+}
+function dictationReducer(state, action) {
+  switch (action.type) {
+    case "start":
+      return { ...INITIAL_DICTATION_STATE, status: DICTATION_STATUSES.starting };
+    case "listening":
+      return state.status === DICTATION_STATUSES.starting ? { ...state, status: DICTATION_STATUSES.listening } : state;
+    case "transcript": {
+      if (state.status !== DICTATION_STATUSES.listening && state.status !== DICTATION_STATUSES.stopping) {
+        return state;
+      }
+      return {
+        ...state,
+        transcript: {
+          final: joinSpeech(state.transcript.final, action.update.final),
+          interim: action.update.interim
+        }
+      };
+    }
+    case "stop":
+      return state.status === DICTATION_STATUSES.listening || state.status === DICTATION_STATUSES.starting ? { ...state, status: DICTATION_STATUSES.stopping } : state;
+    case "ended":
+      return {
+        ...state,
+        status: state.status === DICTATION_STATUSES.error ? DICTATION_STATUSES.error : DICTATION_STATUSES.idle,
+        transcript: { final: joinSpeech(state.transcript.final, state.transcript.interim), interim: "" }
+      };
+    case "cancel":
+      return INITIAL_DICTATION_STATE;
+    case "error":
+      return { ...state, status: DICTATION_STATUSES.error, error: action.error };
+    default:
+      return state;
+  }
+}
+function spokenText(transcript) {
+  return joinSpeech(transcript.final, transcript.interim);
+}
+function composeDictationDraft(existingDraft, spoken) {
+  return joinSpeech(existingDraft, spoken);
+}
+export {
+  DICTATION_STATUSES,
+  INITIAL_DICTATION_STATE,
+  composeDictationDraft,
+  dictationReducer,
+  joinSpeech,
+  spokenText
+};
+//# sourceMappingURL=dictation-state.js.map

package/dist/dictation/dictation-state.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"dictation-state.js","sources":["../../src/dictation/dictation-state.ts"],"sourcesContent":["import type { DictationEngineError, DictationTranscriptUpdate } from './dictation-engine';\n\n/**\n * The dictation session's state machine, kept pure so the transitions are\n * testable without a microphone.\n *\n *   idle ──start──▶ starting ──engine ready──▶ listening ──stop──▶ stopping ──flushed──▶ idle\n *     ▲                │                          │                     │\n *     └────cancel──────┴──────────────────────────┴─────────────────────┘\n *\n * `error` is a terminal state shown inline until the next start or cancel.\n */\nexport const DICTATION_STATUSES = {\n  idle: 'idle',\n  starting: 'starting',\n  listening: 'listening',\n  stopping: 'stopping',\n  error: 'error',\n} as const;\n\nexport type DictationStatus = (typeof DICTATION_STATUSES)[keyof typeof DICTATION_STATUSES];\n\nexport interface DictationTranscript {\n  /** Everything the engine committed this session, space-joined. */\n  final: string;\n  /** The in-progress tail; replaced on every update. */\n  interim: string;\n}\n\nexport interface DictationState {\n  status: DictationStatus;\n  transcript: DictationTranscript;\n  error: DictationEngineError | null;\n}\n\nexport type DictationAction =\n  | { type: 'start' }\n  | { type: 'listening' }\n  | { type: 'transcript'; update: DictationTranscriptUpdate }\n  | { type: 'stop' }\n  | { type: 'ended' }\n  | { type: 'cancel' }\n  | { type: 'error'; error: DictationEngineError };\n\nexport const INITIAL_DICTATION_STATE: DictationState = {\n  status: DICTATION_STATUSES.idle,\n  transcript: { final: '', interim: '' },\n  error: null,\n};\n\n/** Join two transcript fragments with one space, ignoring empty sides. */\nexport function joinSpeech(head: string, tail: string): string {\n  const left = head.trim();\n  const right = tail.trim();\n  if (!left) return right;\n  if (!right) return left;\n  return `${left} ${right}`;\n}\n\nexport function dictationReducer(state: DictationState, action: DictationAction): DictationState {\n  switch (action.type) {\n    case 'start':\n      return { ...INITIAL_DICTATION_STATE, status: DICTATION_STATUSES.starting };\n    case 'listening':\n      return state.status === DICTATION_STATUSES.starting\n        ? { ...state, status: DICTATION_STATUSES.listening }\n        : state;\n    case 'transcript': {\n      if (state.status !== DICTATION_STATUSES.listening && state.status !== DICTATION_STATUSES.stopping) {\n        return state;\n      }\n      return {\n        ...state,\n        transcript: {\n          final: joinSpeech(state.transcript.final, action.update.final),\n          interim: action.update.interim,\n        },\n      };\n    }\n    case 'stop':\n      return state.status === DICTATION_STATUSES.listening || state.status === DICTATION_STATUSES.starting\n        ? { ...state, status: DICTATION_STATUSES.stopping }\n        : state;\n    case 'ended':\n      // A flush may still have an interim tail; commit it, the engine will\n      // not refine it any further. An error reported before the engine closed\n      // stays visible: only a start or cancel clears it.\n      return {\n        ...state,\n        status: state.status === DICTATION_STATUSES.error ? DICTATION_STATUSES.error : DICTATION_STATUSES.idle,\n        transcript: { final: joinSpeech(state.transcript.final, state.transcript.interim), interim: '' },\n      };\n    case 'cancel':\n      return INITIAL_DICTATION_STATE;\n    case 'error':\n      return { ...state, status: DICTATION_STATUSES.error, error: action.error };\n    default:\n      return state;\n  }\n}\n\n/** The text a dictation session hands to the composer once it ends. */\nexport function spokenText(transcript: DictationTranscript): string {\n  return joinSpeech(transcript.final, transcript.interim);\n}\n\n/**\n * Merge spoken text into the draft that existed before recording started.\n * Manual edits are kept; the speech is appended after one space.\n */\nexport function composeDictationDraft(existingDraft: string, spoken: string): string {\n  return joinSpeech(existingDraft, spoken);\n}\n"],"names":[],"mappings":"AAYO,MAAM,qBAAqB;AAAA,EAChC,MAAM;AAAA,EACN,UAAU;AAAA,EACV,WAAW;AAAA,EACX,UAAU;AAAA,EACV,OAAO;AACT;AA0BO,MAAM,0BAA0C;AAAA,EACrD,QAAQ,mBAAmB;AAAA,EAC3B,YAAY,EAAE,OAAO,IAAI,SAAS,GAAA;AAAA,EAClC,OAAO;AACT;AAGO,SAAS,WAAW,MAAc,MAAsB;AAC7D,QAAM,OAAO,KAAK,KAAA;AAClB,QAAM,QAAQ,KAAK,KAAA;AACnB,MAAI,CAAC,KAAM,QAAO;AAClB,MAAI,CAAC,MAAO,QAAO;AACnB,SAAO,GAAG,IAAI,IAAI,KAAK;AACzB;AAEO,SAAS,iBAAiB,OAAuB,QAAyC;AAC/F,UAAQ,OAAO,MAAA;AAAA,IACb,KAAK;AACH,aAAO,EAAE,GAAG,yBAAyB,QAAQ,mBAAmB,SAAA;AAAA,IAClE,KAAK;AACH,aAAO,MAAM,WAAW,mBAAmB,WACvC,EAAE,GAAG,OAAO,QAAQ,mBAAmB,UAAA,IACvC;AAAA,IACN,KAAK,cAAc;AACjB,UAAI,MAAM,WAAW,mBAAmB,aAAa,MAAM,WAAW,mBAAmB,UAAU;AACjG,eAAO;AAAA,MACT;AACA,aAAO;AAAA,QACL,GAAG;AAAA,QACH,YAAY;AAAA,UACV,OAAO,WAAW,MAAM,WAAW,OAAO,OAAO,OAAO,KAAK;AAAA,UAC7D,SAAS,OAAO,OAAO;AAAA,QAAA;AAAA,MACzB;AAAA,IAEJ;AAAA,IACA,KAAK;AACH,aAAO,MAAM,WAAW,mBAAmB,aAAa,MAAM,WAAW,mBAAmB,WACxF,EAAE,GAAG,OAAO,QAAQ,mBAAmB,aACvC;AAAA,IACN,KAAK;AAIH,aAAO;AAAA,QACL,GAAG;AAAA,QACH,QAAQ,MAAM,WAAW,mBAAmB,QAAQ,mBAAmB,QAAQ,mBAAmB;AAAA,QAClG,YAAY,EAAE,OAAO,WAAW,MAAM,WAAW,OAAO,MAAM,WAAW,OAAO,GAAG,SAAS,GAAA;AAAA,MAAG;AAAA,IAEnG,KAAK;AACH,aAAO;AAAA,IACT,KAAK;AACH,aAAO,EAAE,GAAG,OAAO,QAAQ,mBAAmB,OAAO,OAAO,OAAO,MAAA;AAAA,IACrE;AACE,aAAO;AAAA,EAAA;AAEb;AAGO,SAAS,WAAW,YAAyC;AAClE,SAAO,WAAW,WAAW,OAAO,WAAW,OAAO;AACxD;AAMO,SAAS,sBAAsB,eAAuB,QAAwB;AACnF,SAAO,WAAW,eAAe,MAAM;AACzC;"}

package/dist/dictation/recorded-dictation-engine.cjs

@@ -0,0 +1,153 @@
+"use strict";
+Object.defineProperty(exports, Symbol.toStringTag, { value: "Module" });
+const dictationEngine = require("./dictation-engine.cjs");
+const DEFAULT_MIME_TYPES = ["audio/webm;codecs=opus", "audio/webm", "audio/mp4", "audio/ogg;codecs=opus"];
+const DEFAULT_FLUSH_TIMEOUT_MS = 9e4;
+const DEFAULT_MAX_RECORDING_MS = 5 * 6e4;
+function resolveMediaRecorderCtor() {
+  if (typeof window === "undefined") return null;
+  return window.MediaRecorder ?? null;
+}
+function pickRecorderMimeType(Ctor, candidates) {
+  if (!Ctor.isTypeSupported) return void 0;
+  return candidates.find((type) => {
+    var _a;
+    return (_a = Ctor.isTypeSupported) == null ? void 0 : _a.call(Ctor, type);
+  });
+}
+function transcriptionError(error) {
+  return {
+    kind: dictationEngine.DICTATION_ENGINE_ERROR_KINDS.connection,
+    message: error instanceof Error && error.message ? error.message : "Transcription failed. Try again."
+  };
+}
+function createRecordedDictationEngine(options) {
+  const {
+    transcribe,
+    mimeTypes = DEFAULT_MIME_TYPES,
+    flushTimeoutMs = DEFAULT_FLUSH_TIMEOUT_MS,
+    maxDurationMs = DEFAULT_MAX_RECORDING_MS
+  } = options;
+  return {
+    id: "recorded-upload",
+    flushTimeoutMs,
+    stoppingLabel: "Transcribing…",
+    isSupported: () => {
+      var _a;
+      return resolveMediaRecorderCtor() !== null && typeof navigator !== "undefined" && typeof ((_a = navigator.mediaDevices) == null ? void 0 : _a.getUserMedia) === "function";
+    },
+    start: ({ stream, lang, callbacks }) => {
+      const Ctor = resolveMediaRecorderCtor();
+      const noop = { stop: () => void 0, abort: () => void 0 };
+      if (!Ctor || !stream) {
+        callbacks.onError({
+          kind: !Ctor ? dictationEngine.DICTATION_ENGINE_ERROR_KINDS.unsupported : dictationEngine.DICTATION_ENGINE_ERROR_KINDS.permission,
+          message: !Ctor ? "This browser cannot record audio." : "No microphone stream was available."
+        });
+        callbacks.onEnd();
+        return noop;
+      }
+      const mimeType = pickRecorderMimeType(Ctor, mimeTypes);
+      let recorder;
+      try {
+        recorder = new Ctor(stream, mimeType ? { mimeType } : void 0);
+      } catch (error) {
+        callbacks.onError({
+          kind: dictationEngine.DICTATION_ENGINE_ERROR_KINDS.unknown,
+          message: error instanceof Error ? error.message : "Recording could not start."
+        });
+        callbacks.onEnd();
+        return noop;
+      }
+      const chunks = [];
+      const upload = new AbortController();
+      let ended = false;
+      let aborted = false;
+      let limitTimer;
+      const clearLimit = () => {
+        if (limitTimer === void 0) return;
+        clearTimeout(limitTimer);
+        limitTimer = void 0;
+      };
+      const end = () => {
+        if (ended) return;
+        ended = true;
+        clearLimit();
+        recorder.ondataavailable = null;
+        recorder.onstop = null;
+        recorder.onerror = null;
+        callbacks.onEnd();
+      };
+      const stopRecorder = () => {
+        if (recorder.state === "inactive") return false;
+        try {
+          recorder.stop();
+          return true;
+        } catch {
+          return false;
+        }
+      };
+      recorder.ondataavailable = (event) => {
+        if (event.data && event.data.size > 0) chunks.push(event.data);
+      };
+      recorder.onerror = () => {
+        callbacks.onError({
+          kind: dictationEngine.DICTATION_ENGINE_ERROR_KINDS.unknown,
+          message: "Recording stopped unexpectedly."
+        });
+        end();
+      };
+      recorder.onstop = () => {
+        if (aborted || ended) {
+          end();
+          return;
+        }
+        const clip = new Blob(chunks, { type: recorder.mimeType || mimeType || "audio/webm" });
+        if (clip.size === 0) {
+          end();
+          return;
+        }
+        void transcribe(clip, { lang, signal: upload.signal }).then((text) => {
+          if (ended) return;
+          const trimmed = text.trim();
+          if (trimmed) callbacks.onTranscript({ final: trimmed, interim: "" });
+        }).catch((error) => {
+          if (ended) return;
+          callbacks.onError(transcriptionError(error));
+        }).finally(end);
+      };
+      try {
+        recorder.start();
+      } catch (error) {
+        callbacks.onError({
+          kind: dictationEngine.DICTATION_ENGINE_ERROR_KINDS.unknown,
+          message: error instanceof Error ? error.message : "Recording could not start."
+        });
+        end();
+        return noop;
+      }
+      if (Number.isFinite(maxDurationMs) && maxDurationMs > 0) {
+        limitTimer = setTimeout(() => {
+          limitTimer = void 0;
+          if (!ended) stopRecorder();
+        }, maxDurationMs);
+      }
+      return {
+        stop: () => {
+          clearLimit();
+          if (!stopRecorder()) end();
+        },
+        abort: () => {
+          aborted = true;
+          upload.abort();
+          stopRecorder();
+          end();
+        }
+      };
+    }
+  };
+}
+exports.DEFAULT_MAX_RECORDING_MS = DEFAULT_MAX_RECORDING_MS;
+exports.createRecordedDictationEngine = createRecordedDictationEngine;
+exports.pickRecorderMimeType = pickRecorderMimeType;
+//# sourceMappingURL=recorded-dictation-engine.cjs.map

package/dist/dictation/recorded-dictation-engine.cjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"recorded-dictation-engine.cjs","sources":["../../src/dictation/recorded-dictation-engine.ts"],"sourcesContent":["import {\n  DICTATION_ENGINE_ERROR_KINDS,\n  type DictationEngine,\n  type DictationEngineError,\n} from './dictation-engine';\n\nexport interface DictationTranscribeContext {\n  /** BCP-47 tag the session was started with (host `lang` or the document's). */\n  lang: string;\n  /** Fires when the user cancels or the composer gives up waiting. */\n  signal: AbortSignal;\n}\n\n/**\n * Turn a finished recording into text. Receives the whole clip once the user\n * stops; resolves with the transcript (empty when nothing was said). Should\n * honour `signal` so a cancelled dictation does not keep uploading.\n */\nexport type DictationTranscriber = (clip: Blob, context: DictationTranscribeContext) => Promise<string>;\n\nexport interface RecordedDictationEngineOptions {\n  transcribe: DictationTranscriber;\n  /**\n   * Preferred `MediaRecorder` container, tried in order until the browser\n   * accepts one. Defaults to Opus in WebM (Chrome, Firefox, Edge) then MP4\n   * (Safari); the browser's own default is the last resort.\n   */\n  mimeTypes?: readonly string[];\n  /**\n   * Upper bound on the transcription round trip; the in-flight request is\n   * aborted when it elapses. Defaults to 90s, above agora's 60s Deepgram\n   * budget plus upload time.\n   */\n  flushTimeoutMs?: number;\n  /**\n   * Longest recording the engine accepts before it stops on its own and\n   * transcribes what it has, exactly as if Stop had been pressed. Bounds a\n   * mic left open by mistake: every recorded minute is a minute billed by the\n   * recognizer, and agora refuses clips over its size cap. Defaults to\n   * 5 minutes; pass `Infinity` to disable.\n   */\n  maxDurationMs?: number;\n}\n\nconst DEFAULT_MIME_TYPES = ['audio/webm;codecs=opus', 'audio/webm', 'audio/mp4', 'audio/ogg;codecs=opus'];\nconst DEFAULT_FLUSH_TIMEOUT_MS = 90_000;\nexport const DEFAULT_MAX_RECORDING_MS = 5 * 60_000;\n\n// lib.dom typings for MediaRecorder vary across the TypeScript versions the\n// SDK supports; describe only what the engine touches.\ninterface MediaRecorderLike extends EventTarget {\n  readonly state: 'inactive' | 'recording' | 'paused';\n  readonly mimeType: string;\n  start: (timeslice?: number) => void;\n  stop: () => void;\n  ondataavailable: ((event: { data: Blob }) => void) | null;\n  onstop: (() => void) | null;\n  onerror: ((event: Event) => void) | null;\n}\ninterface MediaRecorderCtor {\n  new (stream: MediaStream, options?: { mimeType?: string }): MediaRecorderLike;\n  isTypeSupported?: (type: string) => boolean;\n}\n\nfunction resolveMediaRecorderCtor(): MediaRecorderCtor | null {\n  if (typeof window === 'undefined') return null;\n  return (window as unknown as { MediaRecorder?: MediaRecorderCtor }).MediaRecorder ?? null;\n}\n\n/** First container the recorder accepts, or undefined for the browser default. */\nexport function pickRecorderMimeType(\n  Ctor: Pick<MediaRecorderCtor, 'isTypeSupported'>,\n  candidates: readonly string[],\n): string | undefined {\n  if (!Ctor.isTypeSupported) return undefined;\n  return candidates.find((type) => Ctor.isTypeSupported?.(type));\n}\n\nfunction transcriptionError(error: unknown): DictationEngineError {\n  return {\n    kind: DICTATION_ENGINE_ERROR_KINDS.connection,\n    message: error instanceof Error && error.message ? error.message : 'Transcription failed. Try again.',\n  };\n}\n\n/**\n * Record-then-transcribe: capture the microphone with `MediaRecorder` while\n * the user talks, then hand the finished clip to `transcribe` on stop. No\n * interim words are shown; the composer shows a \"transcribing\" state until\n * the text lands. Works in every browser with `MediaRecorder`, so the mic is\n * available where Web Speech is not (Firefox) and the quality is whatever the\n * server-side recognizer provides.\n */\nexport function createRecordedDictationEngine(options: RecordedDictationEngineOptions): DictationEngine {\n  const {\n    transcribe,\n    mimeTypes = DEFAULT_MIME_TYPES,\n    flushTimeoutMs = DEFAULT_FLUSH_TIMEOUT_MS,\n    maxDurationMs = DEFAULT_MAX_RECORDING_MS,\n  } = options;\n  return {\n    id: 'recorded-upload',\n    flushTimeoutMs,\n    stoppingLabel: 'Transcribing…',\n    isSupported: () =>\n      resolveMediaRecorderCtor() !== null &&\n      typeof navigator !== 'undefined' &&\n      typeof navigator.mediaDevices?.getUserMedia === 'function',\n    start: ({ stream, lang, callbacks }) => {\n      const Ctor = resolveMediaRecorderCtor();\n      const noop = { stop: () => undefined, abort: () => undefined };\n      if (!Ctor || !stream) {\n        callbacks.onError({\n          kind: !Ctor ? DICTATION_ENGINE_ERROR_KINDS.unsupported : DICTATION_ENGINE_ERROR_KINDS.permission,\n          message: !Ctor ? 'This browser cannot record audio.' : 'No microphone stream was available.',\n        });\n        callbacks.onEnd();\n        return noop;\n      }\n\n      const mimeType = pickRecorderMimeType(Ctor, mimeTypes);\n      let recorder: MediaRecorderLike;\n      try {\n        recorder = new Ctor(stream, mimeType ? { mimeType } : undefined);\n      } catch (error) {\n        callbacks.onError({\n          kind: DICTATION_ENGINE_ERROR_KINDS.unknown,\n          message: error instanceof Error ? error.message : 'Recording could not start.',\n        });\n        callbacks.onEnd();\n        return noop;\n      }\n\n      const chunks: Blob[] = [];\n      const upload = new AbortController();\n      let ended = false;\n      let aborted = false;\n      let limitTimer: ReturnType<typeof setTimeout> | undefined;\n      const clearLimit = () => {\n        if (limitTimer === undefined) return;\n        clearTimeout(limitTimer);\n        limitTimer = undefined;\n      };\n      const end = () => {\n        if (ended) return;\n        ended = true;\n        clearLimit();\n        recorder.ondataavailable = null;\n        recorder.onstop = null;\n        recorder.onerror = null;\n        callbacks.onEnd();\n      };\n      const stopRecorder = () => {\n        if (recorder.state === 'inactive') return false;\n        try {\n          recorder.stop();\n          return true;\n        } catch {\n          return false;\n        }\n      };\n\n      recorder.ondataavailable = (event) => {\n        if (event.data && event.data.size > 0) chunks.push(event.data);\n      };\n      recorder.onerror = () => {\n        callbacks.onError({\n          kind: DICTATION_ENGINE_ERROR_KINDS.unknown,\n          message: 'Recording stopped unexpectedly.',\n        });\n        end();\n      };\n      recorder.onstop = () => {\n        if (aborted || ended) {\n          end();\n          return;\n        }\n        const clip = new Blob(chunks, { type: recorder.mimeType || mimeType || 'audio/webm' });\n        if (clip.size === 0) {\n          end();\n          return;\n        }\n        void transcribe(clip, { lang, signal: upload.signal })\n          .then((text) => {\n            if (ended) return;\n            const trimmed = text.trim();\n            if (trimmed) callbacks.onTranscript({ final: trimmed, interim: '' });\n          })\n          .catch((error: unknown) => {\n            if (ended) return;\n            callbacks.onError(transcriptionError(error));\n          })\n          .finally(end);\n      };\n\n      try {\n        recorder.start();\n      } catch (error) {\n        callbacks.onError({\n          kind: DICTATION_ENGINE_ERROR_KINDS.unknown,\n          message: error instanceof Error ? error.message : 'Recording could not start.',\n        });\n        end();\n        return noop;\n      }\n\n      if (Number.isFinite(maxDurationMs) && maxDurationMs > 0) {\n        limitTimer = setTimeout(() => {\n          limitTimer = undefined;\n          // Same path as the user's Stop: onstop transcribes the clip and the\n          // words land in the draft. A recorder that is already inactive is a\n          // stop in flight, which must run to completion untouched.\n          if (!ended) stopRecorder();\n        }, maxDurationMs);\n      }\n\n      return {\n        stop: () => {\n          clearLimit();\n          if (!stopRecorder()) end();\n        },\n        abort: () => {\n          aborted = true;\n          upload.abort();\n          stopRecorder();\n          end();\n        },\n      };\n    },\n  };\n}\n"],"names":["DICTATION_ENGINE_ERROR_KINDS"],"mappings":";;;AA4CA,MAAM,qBAAqB,CAAC,0BAA0B,cAAc,aAAa,uBAAuB;AACxG,MAAM,2BAA2B;AAC1B,MAAM,2BAA2B,IAAI;AAkB5C,SAAS,2BAAqD;AAC5D,MAAI,OAAO,WAAW,YAAa,QAAO;AAC1C,SAAQ,OAA4D,iBAAiB;AACvF;AAGO,SAAS,qBACd,MACA,YACoB;AACpB,MAAI,CAAC,KAAK,gBAAiB,QAAO;AAClC,SAAO,WAAW,KAAK,CAAC;;AAAS,sBAAK,oBAAL,8BAAuB;AAAA,GAAK;AAC/D;AAEA,SAAS,mBAAmB,OAAsC;AAChE,SAAO;AAAA,IACL,MAAMA,gBAAAA,6BAA6B;AAAA,IACnC,SAAS,iBAAiB,SAAS,MAAM,UAAU,MAAM,UAAU;AAAA,EAAA;AAEvE;AAUO,SAAS,8BAA8B,SAA0D;AACtG,QAAM;AAAA,IACJ;AAAA,IACA,YAAY;AAAA,IACZ,iBAAiB;AAAA,IACjB,gBAAgB;AAAA,EAAA,IACd;AACJ,SAAO;AAAA,IACL,IAAI;AAAA,IACJ;AAAA,IACA,eAAe;AAAA,IACf,aAAa,MAAA;;AACX,4CAA+B,QAC/B,OAAO,cAAc,eACrB,SAAO,eAAU,iBAAV,mBAAwB,kBAAiB;AAAA;AAAA,IAClD,OAAO,CAAC,EAAE,QAAQ,MAAM,gBAAgB;AACtC,YAAM,OAAO,yBAAA;AACb,YAAM,OAAO,EAAE,MAAM,MAAM,QAAW,OAAO,MAAM,OAAA;AACnD,UAAI,CAAC,QAAQ,CAAC,QAAQ;AACpB,kBAAU,QAAQ;AAAA,UAChB,MAAM,CAAC,OAAOA,gBAAAA,6BAA6B,cAAcA,gBAAAA,6BAA6B;AAAA,UACtF,SAAS,CAAC,OAAO,sCAAsC;AAAA,QAAA,CACxD;AACD,kBAAU,MAAA;AACV,eAAO;AAAA,MACT;AAEA,YAAM,WAAW,qBAAqB,MAAM,SAAS;AACrD,UAAI;AACJ,UAAI;AACF,mBAAW,IAAI,KAAK,QAAQ,WAAW,EAAE,SAAA,IAAa,MAAS;AAAA,MACjE,SAAS,OAAO;AACd,kBAAU,QAAQ;AAAA,UAChB,MAAMA,gBAAAA,6BAA6B;AAAA,UACnC,SAAS,iBAAiB,QAAQ,MAAM,UAAU;AAAA,QAAA,CACnD;AACD,kBAAU,MAAA;AACV,eAAO;AAAA,MACT;AAEA,YAAM,SAAiB,CAAA;AACvB,YAAM,SAAS,IAAI,gBAAA;AACnB,UAAI,QAAQ;AACZ,UAAI,UAAU;AACd,UAAI;AACJ,YAAM,aAAa,MAAM;AACvB,YAAI,eAAe,OAAW;AAC9B,qBAAa,UAAU;AACvB,qBAAa;AAAA,MACf;AACA,YAAM,MAAM,MAAM;AAChB,YAAI,MAAO;AACX,gBAAQ;AACR,mBAAA;AACA,iBAAS,kBAAkB;AAC3B,iBAAS,SAAS;AAClB,iBAAS,UAAU;AACnB,kBAAU,MAAA;AAAA,MACZ;AACA,YAAM,eAAe,MAAM;AACzB,YAAI,SAAS,UAAU,WAAY,QAAO;AAC1C,YAAI;AACF,mBAAS,KAAA;AACT,iBAAO;AAAA,QACT,QAAQ;AACN,iBAAO;AAAA,QACT;AAAA,MACF;AAEA,eAAS,kBAAkB,CAAC,UAAU;AACpC,YAAI,MAAM,QAAQ,MAAM,KAAK,OAAO,EAAG,QAAO,KAAK,MAAM,IAAI;AAAA,MAC/D;AACA,eAAS,UAAU,MAAM;AACvB,kBAAU,QAAQ;AAAA,UAChB,MAAMA,gBAAAA,6BAA6B;AAAA,UACnC,SAAS;AAAA,QAAA,CACV;AACD,YAAA;AAAA,MACF;AACA,eAAS,SAAS,MAAM;AACtB,YAAI,WAAW,OAAO;AACpB,cAAA;AACA;AAAA,QACF;AACA,cAAM,OAAO,IAAI,KAAK,QAAQ,EAAE,MAAM,SAAS,YAAY,YAAY,cAAc;AACrF,YAAI,KAAK,SAAS,GAAG;AACnB,cAAA;AACA;AAAA,QACF;AACA,aAAK,WAAW,MAAM,EAAE,MAAM,QAAQ,OAAO,QAAQ,EAClD,KAAK,CAAC,SAAS;AACd,cAAI,MAAO;AACX,gBAAM,UAAU,KAAK,KAAA;AACrB,cAAI,mBAAmB,aAAa,EAAE,OAAO,SAAS,SAAS,IAAI;AAAA,QACrE,CAAC,EACA,MAAM,CAAC,UAAmB;AACzB,cAAI,MAAO;AACX,oBAAU,QAAQ,mBAAmB,KAAK,CAAC;AAAA,QAC7C,CAAC,EACA,QAAQ,GAAG;AAAA,MAChB;AAEA,UAAI;AACF,iBAAS,MAAA;AAAA,MACX,SAAS,OAAO;AACd,kBAAU,QAAQ;AAAA,UAChB,MAAMA,gBAAAA,6BAA6B;AAAA,UACnC,SAAS,iBAAiB,QAAQ,MAAM,UAAU;AAAA,QAAA,CACnD;AACD,YAAA;AACA,eAAO;AAAA,MACT;AAEA,UAAI,OAAO,SAAS,aAAa,KAAK,gBAAgB,GAAG;AACvD,qBAAa,WAAW,MAAM;AAC5B,uBAAa;AAIb,cAAI,CAAC,MAAO,cAAA;AAAA,QACd,GAAG,aAAa;AAAA,MAClB;AAEA,aAAO;AAAA,QACL,MAAM,MAAM;AACV,qBAAA;AACA,cAAI,CAAC,aAAA,EAAgB,KAAA;AAAA,QACvB;AAAA,QACA,OAAO,MAAM;AACX,oBAAU;AACV,iBAAO,MAAA;AACP,uBAAA;AACA,cAAA;AAAA,QACF;AAAA,MAAA;AAAA,IAEJ;AAAA,EAAA;AAEJ;;;;"}