@smartspace/chat-ui 1.14.4-pr.453.43dda12 → 1.14.4-pr.453.d6dd83a

package/dist/index.d.ts

@@ -127,8 +127,8 @@ type UseDictationOptions = {
     onInterim?: (text: string) => void;
     /**
      * Hard stop, so a forgotten mic can't run for an hour. Keep it under the
-     * token's ~10 minute life: the token is fetched once per session and never
-     * refreshed, so a longer cap would end the session on an auth failure.
+     * token's ~10 minute life: the token is never refreshed once the session is
+     * running, so a longer cap would end the session on an auth failure.
      */
     maxDurationMs?: number;
     /** Stop after this long without any speech. */
@@ -136,7 +136,7 @@ type UseDictationOptions = {
 };
 /**
  * Streams microphone audio to Azure AI Speech and hands back recognised
- * phrases. Owns the whole session: mic permission + stream, lazy SDK load,
+ * phrases. Owns the whole session: mic permission + stream, SDK loading,
  * recogniser lifecycle, idle/max timers, and teardown on stop or unmount.
  * Interim (not yet final) text goes to `onInterim` for display only; only final
  * phrases reach `onPhrase`.
@@ -481,10 +481,12 @@ interface ChatService {
      * Speech-to-text (composer dictation). Both are optional so existing
      * service implementations keep compiling — the composer renders a
      * microphone only when both are provided AND the config says enabled.
-     * `getSpeechConfig` is read once per session. `getSpeechToken` is called once
-     * per dictation session, before the SDK connects — the token it returns is
-     * handed over as-is and never refreshed, so a session is capped to stay
-     * inside its life. No caching wanted: a stale token cannot be renewed.
+     * `getSpeechConfig` is read once per session. `getSpeechToken` is called
+     * when a dictation session starts (in parallel with the mic permission
+     * prompt — and again if that token aged out while the prompt sat open). The
+     * token is handed to the SDK as-is and never refreshed mid-session, so a
+     * session is capped to stay inside its life. No caching wanted: a stale
+     * token cannot be renewed.
      */
     getSpeechConfig?(): Promise<SpeechConfig>;
     getSpeechToken?(): Promise<SpeechToken>;

package/dist/index.js

@@ -2557,6 +2557,17 @@ var MarkdownEditor = forwardRef((props, ref) => {
 });
 MarkdownEditor.displayName = "MarkdownEditor";
 var TOKEN_EXPIRY_MARGIN_MS = 15e3;
+var STALE_TOKEN_FLOOR_MS = 6e4;
+function kickOff(fn) {
+  let promise;
+  try {
+    promise = Promise.resolve(fn());
+  } catch (e) {
+    promise = Promise.reject(e);
+  }
+  promise.catch(() => void 0);
+  return promise;
+}
 var isSupported = () => typeof window !== "undefined" && window.isSecureContext && !!navigator.mediaDevices?.getUserMedia;
 function classifyTokenFailure(error) {
   const code3 = error?.code;
@@ -2588,6 +2599,18 @@ function useDictation({
   const clearInterim = useCallback(() => onInterimRef.current?.(""), []);
   const supported = isSupported();
   const available = supported && !!region && !!getToken;
+  useEffect(() => {
+    if (!available) return;
+    const nav = navigator;
+    if (nav.connection?.saveData) return;
+    const warm = () => void kickOff(() => import('microsoft-cognitiveservices-speech-sdk'));
+    if (typeof window.requestIdleCallback === "function") {
+      const id2 = window.requestIdleCallback(warm, { timeout: 2e3 });
+      return () => window.cancelIdleCallback(id2);
+    }
+    const id = window.setTimeout(warm, 1500);
+    return () => window.clearTimeout(id);
+  }, [available]);
   const stop = useCallback(() => {
     generationRef.current += 1;
     const session = sessionRef.current;
@@ -2609,15 +2632,22 @@ function useDictation({
     const superseded = () => generationRef.current !== generation;
     setError(null);
     setState("starting");
-    let stream;
-    try {
-      stream = await navigator.mediaDevices.getUserMedia({
+    const streamPromise = kickOff(
+      () => navigator.mediaDevices.getUserMedia({
         audio: {
           echoCancellation: true,
           noiseSuppression: true,
           autoGainControl: true
         }
-      });
+      })
+    );
+    const tokenPromise = kickOff(getToken);
+    const sdkPromise = kickOff(
+      () => import('microsoft-cognitiveservices-speech-sdk')
+    );
+    let stream;
+    try {
+      stream = await streamPromise;
     } catch (e) {
       if (superseded()) return;
       const name = e?.name;
@@ -2633,7 +2663,10 @@ function useDictation({
     }
     let token;
     try {
-      token = await getToken();
+      token = await tokenPromise;
+      if (Date.parse(token.expiresOn) - Date.now() < STALE_TOKEN_FLOOR_MS) {
+        token = await getToken();
+      }
     } catch (e) {
       stream.getTracks().forEach((t) => t.stop());
       if (superseded()) return;
@@ -2648,7 +2681,7 @@ function useDictation({
     let recognizer = null;
     let published = null;
     try {
-      const sdk = await import('microsoft-cognitiveservices-speech-sdk');
+      const sdk = await sdkPromise;
       if (superseded()) {
         stream.getTracks().forEach((t) => t.stop());
         return;