@jskit-ai/assistant-voice 0.1.40 → 0.1.42

package/README.md

@@ -2,7 +2,7 @@
 
 Optional duplex voice for Vue 3/Vuetify applications. One shell controller owns
 microphone, playback, active destination and modal. Applications retain their
-conversations, permissions, message admission and history. No Vibe64 or Genesis
+conversations, permissions, message admission and history. No Genesis
 runtime is required.
 
 Install `@jskit-ai/assistant-voice` with Vue and Vuetify. Run your own speech
@@ -70,8 +70,8 @@ outside its screen. `available = false` releases the active voice session. Dispo
 the controller on app teardown or identity replacement. IDs must include the
 identity scope when different people can use the same conversation ID.
 
-Optional defaults: `mode` (`push-to-talk` or `hands-free`), `talkMode` (`tap` or
-`hold` for one-off recording), `reviewBeforeSend`, and an installed `voiceId`.
+Optional defaults: `mode` (`push-to-talk` or `hands-free`), `reviewBeforeSend`,
+and an installed `voiceId`.
 One-off recordings started through `startHeldRecording()` require review; the
 modal's push-to-talk gesture uses `reviewBeforeSend`. `avatar` and `onVisual`/`onTranscript`
 allow presentation in the surrounding app without adding another audio owner.
@@ -186,6 +186,8 @@ The engine supplies a recognizer, asynchronous synthesis and optionally a public
 voice catalogue. A host may supply synchronous `authorize(token)` and
 `handleRequest(request, response)` for its own grant policy. An authorized
 principal has `tenant` and optional `parentTenant`; both connection quotas apply.
+The catalogue and the WebSocket apply the same rule: any other result, a
+Promise or a thrown error denies.
 `revokeTenant(id)` ends current child or parent sessions as well as pending work.
 The host must reject future requests for revoked grants in its authorizer.
 
@@ -195,7 +197,7 @@ The host must reject future requests for revoked grants in its authorizer.
 client binding and one shared realtime connection for five file-backed chats.
 Text and voice retain the same conversation; an optional headless task uses the
 same runtime directly. Its README covers packed-package setup, launch and the
-local origin policy. No Vibe64, Genesis or database runtime is required.
+local origin policy. No Genesis or database runtime is required.
 
 Focused package checks run one file at a time, for example:
 `npm test --workspace packages/assistant-voice -- test/voiceController.test.js`.

package/examples/conversation/README.md

@@ -4,7 +4,7 @@ This Vue/Vuetify and Fastify application uses the supplied JSKIT conversation
 server integration, client binding and shared realtime connection. Planning,
 Notes, Research, Writing and Review have separate file-backed histories. One
 root voice modal retains its target independently of mounted text cards.
-Vibe64, Genesis and a database are not required.
+Genesis and a database are not required.
 
 ## Install and start
 

package/examples/conversation/package.json

@@ -9,14 +9,14 @@
   },
   "dependencies": {
     "@fastify/websocket": "^11.2.0",
-    "@jskit-ai/assistant-core": "0.1.239",
-    "@jskit-ai/assistant-runtime": "0.1.233",
-    "@jskit-ai/assistant-voice": "0.1.40",
-    "@jskit-ai/connectors-catalog": "0.1.63",
-    "@jskit-ai/http-runtime": "0.1.243",
-    "@jskit-ai/kernel": "0.1.245",
-    "@jskit-ai/realtime": "0.1.243",
-    "@jskit-ai/shell-web": "0.1.250",
+    "@jskit-ai/assistant-core": "0.1.241",
+    "@jskit-ai/assistant-runtime": "0.1.235",
+    "@jskit-ai/assistant-voice": "0.1.42",
+    "@jskit-ai/connectors-catalog": "0.1.65",
+    "@jskit-ai/http-runtime": "0.1.244",
+    "@jskit-ai/kernel": "0.1.246",
+    "@jskit-ai/realtime": "0.1.245",
+    "@jskit-ai/shell-web": "0.1.252",
     "@mdi/js": "^7.4.47",
     "@tanstack/vue-query": "^5.90.5",
     "fastify": "^5.8.0",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@jskit-ai/assistant-voice",
-  "version": "0.1.40",
+  "version": "0.1.42",
   "type": "module",
   "description": "Duplex voice conversations with application-owned targets, authorization and speech services.",
   "scripts": {
@@ -18,7 +18,7 @@
     "examples"
   ],
   "dependencies": {
-    "@jskit-ai/assistant-core": "0.1.239",
+    "@jskit-ai/assistant-core": "0.1.241",
     "@mdi/js": "^7.4.47",
     "ws": "^8.20.1"
   },

package/src/client/voiceConversation.js

@@ -1,11 +1,10 @@
 import { computed, onScopeDispose, ref, watch } from "vue";
-import { useVoiceTransport } from "./voiceTransport.js";
+import { createVoiceTurnId, useVoiceTransport } from "./voiceTransport.js";
 import { speechTextFromAssistant, takeStreamingSpeech, VOICE_MAX_SPEECH_TEXT_CHARACTERS } from "../shared/protocol.js";
 import { createConversationNarrationTracker } from "./conversationNarration.js";
 
 /** Created inside the controller's effect scope, independent of a mounted screen. */
 export function useVoiceConversation(binding, { socketUrl, createTransport = useVoiceTransport } = {}) {
-  const oneOffTalkMode = computed(() => binding.defaults?.talkMode === "hold" ? "hold" : "tap");
   const targetLabel = computed(() => binding.label || "Assistant");
   const readAloud = ref(binding.defaults?.readAloud === true);
   const readAloudChangePending = computed(() => binding.readAloudChangePending === true);
@@ -43,9 +42,6 @@ export function useVoiceConversation(binding, { socketUrl, createTransport = use
   const microphoneMuted = voice.microphoneMuted;
   const capturing = computed(() => voice.captureState.value !== "idle");
   const speechActive = computed(() => Boolean(voice.activeSpeechTurnId.value));
-  const talkLabel = computed(() => voice.listening.value ? "Send recording"
-    : capturing.value ? "Cancel recording" : `Talk to ${targetLabel.value}`);
-  const liveLabel = computed(() => starting.value ? "Cancel voice startup" : live.value ? "End voice" : "Start voice");
   const callConnection = computed(() => {
     if (voice.reconnecting.value) return "Voice interrupted";
     if (error.value || voice.error.value) return "Voice needs attention";
@@ -71,19 +67,8 @@ export function useVoiceConversation(binding, { socketUrl, createTransport = use
     if (voice.listening.value) return "Listening to you";
     return callMode.value === "push-to-talk" ? "Hold to speak" : "Preparing microphone…";
   });
-  const callDetail = computed(() => {
-    if (voice.reconnecting.value) return "Waiting for the voice connection. Your unfinished words remain available for review.";
-    if (error.value || voice.error.value) return "Review your message below or try voice again.";
-    if (starting.value || live.value && ["connecting", "opening"].includes(voice.captureState.value)) return "Allow microphone access if your browser asks.";
-    if (!live.value) return "";
-    if (microphoneMuted.value) return "Unmute when you want to speak. You can still listen.";
-    if (!readAloud.value) return "Sound is off. Answers still appear in chat.";
-    if (callMode.value === "push-to-talk") return "Hold to speak. Release to send. The conversation stays open.";
-    return "You can speak naturally, including while I reply.";
-  });
   const callModeBusy = computed(() => starting.value || changingCallMode.value || pushHolding.value || sending.value || Boolean(pendingTranscript.value)
     || capturing.value && Boolean(voice.partialTranscript.value.trim()) || capturing.value && !voice.listening.value);
-  const microphoneStatus = computed(() => microphoneMuted.value ? "Microphone muted" : voice.listening.value ? "Microphone on" : "Microphone off");
   const avatarVisual = computed(() => ({
     avatar: binding.avatar,
     state: voice.speaking.value ? "speaking" : pushHolding.value && voice.listening.value ? "listening" : binding.state.status === "working" ? "thinking"
@@ -274,7 +259,7 @@ export function useVoiceConversation(binding, { socketUrl, createTransport = use
         if (!text) continue;
         if (!reply.started) {
           reply.started = true;
-          reply.turnId = crypto.randomUUID();
+          reply.turnId = createVoiceTurnId("speak");
           const accepted = await voice.speak(text, reply.turnId, { stream: true });
           if (disposed) return;
           if (reply.cancelled) {
@@ -872,10 +857,10 @@ export function useVoiceConversation(binding, { socketUrl, createTransport = use
       (sending.value || Boolean(pendingTranscript.value) || capturing.value));
   const hasUnsentSpeech = computed(() => capturing.value || starting.value || sending.value || Boolean(pendingTranscript.value));
   return {
-    oneOffTalkMode, targetLabel, voice, readAloud, readAloudChangePending, error, pendingTranscript, heldReview, holding, sending,
+    targetLabel, voice, readAloud, readAloudChangePending, error, pendingTranscript, heldReview, holding, sending,
     live, starting, callMode, changingCallMode, pushHolding, microphoneMuted, capturing,
-    speechActive, talkLabel, liveLabel, callConnection, callStatus, callDetail, callModeBusy,
-    microphoneStatus, avatarVisual, callAudioLevel, status, heldTranscript, voiceWords, voiceAnswer,
+    speechActive, callConnection, callStatus, callModeBusy,
+    avatarVisual, callAudioLevel, status, heldTranscript, voiceWords, voiceAnswer,
     stopSpeech, inviteSpeech, toggleReadAloud, enableSound, soundPreparing, deliverTranscript, cancelRecording,
     discardRecording, dismissTranscript, talk, toggleMicrophoneMuted, changeCallMode,
     startPushToTalk, finishPushToTalk, cancelPushToTalk, toggleLive, toggleHandsFree, startHeldRecording,

package/src/client/voicePlayback.js

@@ -1,3 +1,5 @@
+import { VOICE_OUTPUT_SAMPLE_RATE } from "../shared/protocol.js";
+
 const MOUTH_POSES = new Set(["closed", "open", "round", "small", "wide"]);
 
 function pcm16ArrayBufferToFloat32(buffer) {
@@ -9,10 +11,10 @@ function pcm16ArrayBufferToFloat32(buffer) {
   return samples;
 }
 
-function normalizeSpeechSegment(message = {}, sampleRate = 22_050) {
+function normalizeSpeechSegment(message = {}, sampleRate = VOICE_OUTPUT_SAMPLE_RATE) {
   const sampleCount = Math.max(0, Math.floor(Number(message.sampleCount) || 0));
   const durationMs = sampleCount
-    ? (sampleCount / Math.max(1, Number(sampleRate) || 22_050)) * 1_000
+    ? (sampleCount / Math.max(1, Number(sampleRate) || VOICE_OUTPUT_SAMPLE_RATE)) * 1_000
     : Math.max(0, Number(message.durationMs) || 0);
   const cues = (Array.isArray(message.cues) ? message.cues : [])
     .map((cue) => ({

package/src/client/voiceTransport.js

@@ -1,6 +1,12 @@
 import { computed, onScopeDispose, ref, watch } from "vue";
 
-import { VOICE_MAX_SYNTHESIS_TIMEOUT_MS, speechTextFromAssistant } from "../shared/protocol.js";
+import {
+  VOICE_DEFAULT_SYNTHESIS_TIMEOUT_MS,
+  VOICE_INPUT_SAMPLE_RATE,
+  VOICE_MAX_SYNTHESIS_TIMEOUT_MS,
+  VOICE_OUTPUT_SAMPLE_RATE,
+  speechTextFromAssistant
+} from "../shared/protocol.js";
 import { createMicrophoneCapture } from "./audioCapture.js";
 import {
   normalizeSpeechSegment,
@@ -8,12 +14,11 @@ import {
   visemeCueAtTime
 } from "./voicePlayback.js";
 
-const INPUT_SAMPLE_RATE = 16_000;
 const WEBSOCKET_OPEN = 1;
 let microphoneOwner = null;
 const speechOwner = ref(null);
 
-function voiceTurnId(prefix = "turn") {
+function createVoiceTurnId(prefix = "turn") {
   const id = globalThis.crypto?.randomUUID?.() ||
     `${Date.now().toString(36)}-${Math.random().toString(36).slice(2)}`;
   return `${prefix}-${id}`;
@@ -63,7 +68,7 @@ function useVoiceTransport({ socketUrl, speechEnabled = true, interruptSpeechOnL
   let heartbeatTimer;
   let lastPongAt = 0;
   let lastSpeechProgressAt = 0;
-  let synthesisTimeoutMs = 20_000;
+  let synthesisTimeoutMs = VOICE_DEFAULT_SYNTHESIS_TIMEOUT_MS;
   let lastStartedSegmentIndex = -1;
   let microphone = null;
   let preserveBufferedMicrophone = false;
@@ -83,7 +88,7 @@ function useVoiceTransport({ socketUrl, speechEnabled = true, interruptSpeechOnL
   let playbackQueue = Promise.resolve();
   let queuedPlaybackFrames = 0;
   let speechStreamEnded = false;
-  let speechSampleRate = 22_050;
+  let speechSampleRate = VOICE_OUTPUT_SAMPLE_RATE;
   let receivingTurnId = "";
   let receivingSegment = null;
   const speechInputEnded = ref(true);
@@ -407,7 +412,7 @@ function useVoiceTransport({ socketUrl, speechEnabled = true, interruptSpeechOnL
     stopAfterSegment = Infinity;
     bufferedSeconds.value = 0;
     synthesisPending.value = 0;
-    synthesisTimeoutMs = 20_000;
+    synthesisTimeoutMs = VOICE_DEFAULT_SYNTHESIS_TIMEOUT_MS;
     lastStartedSegmentIndex = -1;
     receivingTurnId = "";
     receivingSegment = null;
@@ -474,7 +479,7 @@ function useVoiceTransport({ socketUrl, speechEnabled = true, interruptSpeechOnL
       if (!message.turnId || message.turnId !== activeSpeechTurnId.value) {
         return;
       }
-      speechSampleRate = Number(message.sampleRate) || 22_050;
+      speechSampleRate = Number(message.sampleRate) || VOICE_OUTPUT_SAMPLE_RATE;
       speechStreamEnded = false;
       speechState.value = "thinking";
       nextPlaybackTime = 0;
@@ -484,8 +489,8 @@ function useVoiceTransport({ socketUrl, speechEnabled = true, interruptSpeechOnL
       if (!speechStreamEnded && Number.isSafeInteger(message.segmentIndex) &&
           message.segmentIndex > lastStartedSegmentIndex) {
         lastStartedSegmentIndex = message.segmentIndex;
-        synthesisTimeoutMs = 20_000;
-        if (Number.isSafeInteger(message.synthesisTimeoutMs) && message.synthesisTimeoutMs > 20_000 &&
+        synthesisTimeoutMs = VOICE_DEFAULT_SYNTHESIS_TIMEOUT_MS;
+        if (Number.isSafeInteger(message.synthesisTimeoutMs) && message.synthesisTimeoutMs > VOICE_DEFAULT_SYNTHESIS_TIMEOUT_MS &&
             message.synthesisTimeoutMs <= VOICE_MAX_SYNTHESIS_TIMEOUT_MS) {
           // The engine acquired its queue slot. Allow its bounded cold job until
           // first PCM; repeated metadata and pongs are not synthesis progress.
@@ -498,7 +503,7 @@ function useVoiceTransport({ socketUrl, speechEnabled = true, interruptSpeechOnL
       return;
     }
     if (type === "speech.chunk.end" && message.turnId === activeSpeechTurnId.value) {
-      synthesisTimeoutMs = 20_000;
+      synthesisTimeoutMs = VOICE_DEFAULT_SYNTHESIS_TIMEOUT_MS;
       lastSpeechProgressAt = Date.now();
       synthesisPending.value = Math.max(0, synthesisPending.value - 1);
       return;
@@ -532,7 +537,7 @@ function useVoiceTransport({ socketUrl, speechEnabled = true, interruptSpeechOnL
   function handleSocketMessage(event) {
     if (event.data instanceof ArrayBuffer) {
       if (speechOutputEnabled.value && activeSpeechTurnId.value && receivingTurnId === activeSpeechTurnId.value && !speechStreamEnded) {
-        synthesisTimeoutMs = 20_000;
+        synthesisTimeoutMs = VOICE_DEFAULT_SYNTHESIS_TIMEOUT_MS;
         lastSpeechProgressAt = Date.now();
         enqueuePcm(event.data);
       }
@@ -652,7 +657,7 @@ function useVoiceTransport({ socketUrl, speechEnabled = true, interruptSpeechOnL
     microphoneClaim = Symbol("voice microphone");
     microphoneOwner = microphoneClaim;
     preserveBufferedMicrophone = false;
-    const turnId = voiceTurnId("listen");
+    const turnId = createVoiceTurnId("listen");
     pendingListenTurnId = turnId;
     if (interruptSpeechOnListen) stopSpeaking();
     captureState.value = "connecting";
@@ -678,7 +683,7 @@ function useVoiceTransport({ socketUrl, speechEnabled = true, interruptSpeechOnL
               socket.send(frame);
             }
           },
-          targetSampleRate: INPUT_SAMPLE_RATE
+          targetSampleRate: VOICE_INPUT_SAMPLE_RATE
         });
         openingMicrophones.add(opening);
         try { capture = await opening; }
@@ -703,7 +708,7 @@ function useVoiceTransport({ socketUrl, speechEnabled = true, interruptSpeechOnL
       try {
         endpoint.value = null;
         completedUtterance.value = null;
-        sendControl({ sampleRate: INPUT_SAMPLE_RATE, turnId, type: "listen.start", ...(continuous ? { continuous: true } : {}) });
+        sendControl({ sampleRate: VOICE_INPUT_SAMPLE_RATE, turnId, type: "listen.start", ...(continuous ? { continuous: true } : {}) });
         captureState.value = "listening";
       } catch (controlError) {
         activeListenTurnId.value = "";
@@ -738,7 +743,7 @@ function useVoiceTransport({ socketUrl, speechEnabled = true, interruptSpeechOnL
     sendControl({ turnId, type: "listen.stop" });
   }
 
-  async function speak(text, turnId = voiceTurnId("speak"), { stream = false } = {}) {
+  async function speak(text, turnId = createVoiceTurnId("speak"), { stream = false } = {}) {
     const speechText = speechTextFromAssistant(text);
     if (unmounted || !speechOutputEnabled.value || !speechText) {
       return false;
@@ -958,6 +963,6 @@ function useVoiceTransport({ socketUrl, speechEnabled = true, interruptSpeechOnL
 }
 
 export {
-  useVoiceTransport,
-  voiceTurnId
+  createVoiceTurnId,
+  useVoiceTransport
 };

package/src/server/index.js

@@ -2,4 +2,3 @@ export { createVoiceDaemon } from "./voiceDaemon.js";
 export { registerVoiceProxyRoute, readVoiceCatalogue } from "./voiceProxy.js";
 export { resolveVoiceProxyConfig, readVoiceAccessToken, voiceEndpoint } from "./voiceProxyConfig.js";
 export { createVoiceAccessToken, verifyVoiceAccessToken, bearerVoiceAccessToken } from "./voiceAccessToken.js";
-export { normalizeVoiceCloseMetadata } from "./voiceSocketClose.js";

package/src/server/voiceDaemon.js

@@ -4,6 +4,7 @@ import { WebSocketServer } from "ws";
 
 import {
   VOICE_DAEMON_WEBSOCKET_PATH,
+  VOICE_DEFAULT_SYNTHESIS_TIMEOUT_MS,
   VOICE_INPUT_SAMPLE_RATE,
   VOICE_MAX_AUDIO_FRAME_BYTES,
   VOICE_MAX_CONTROL_BYTES,
@@ -147,6 +148,13 @@ function createVoiceDaemon({
   const sockets = new Set();
   const principals = new WeakMap();
   const speechQueue = createBoundedSerialQueue({ maximumQueued: maximumQueuedSpeech });
+  // One acceptance rule for every request: the host authorizer is synchronous
+  // and must name a tenant; anything else, including a thrown error, denies.
+  function principalFor(request) {
+    let principal;
+    try { principal = authorize(bearerVoiceAccessToken(request.headers), request); } catch { return null; }
+    return principal && typeof principal.tenant === "string" && principal.tenant ? principal : null;
+  }
   let ready = false;
   let startedAt = "";
   let stopping = false;
@@ -167,9 +175,7 @@ function createVoiceDaemon({
     }
     try {
       if (request.method === "GET" && requestPathname(request) === "/voices") {
-        let principal;
-        try { principal = await authorize(bearerVoiceAccessToken(request.headers), request); } catch { /* Deny invalid credentials. */ }
-        if (!principal) { writeJsonResponse(response, 401, { ok: false }); return; }
+        if (!principalFor(request)) { writeJsonResponse(response, 401, { ok: false }); return; }
         if (!ready || stopping) { writeJsonResponse(response, 503, { ok: false }); return; }
         writeJsonResponse(response, 200, voiceCatalogue());
         return;
@@ -359,7 +365,7 @@ function createVoiceDaemon({
               const segmentIndex = active.segmentIndex++;
               const synthesisTimeoutMs = engine.synthesisTimeoutMs;
               sendJson(socket, "speech.segment.start", { segmentIndex, turnId,
-                ...(Number.isSafeInteger(synthesisTimeoutMs) && synthesisTimeoutMs > 20_000 &&
+                ...(Number.isSafeInteger(synthesisTimeoutMs) && synthesisTimeoutMs > VOICE_DEFAULT_SYNTHESIS_TIMEOUT_MS &&
                     synthesisTimeoutMs <= VOICE_MAX_SYNTHESIS_TIMEOUT_MS ? { synthesisTimeoutMs } : {}) });
               let sampleCount = 0;
               const result = text.trim() ? await engine.synthesize(text.trim(), {
@@ -511,10 +517,8 @@ function createVoiceDaemon({
       socket.destroy();
       return;
     }
-    let verified;
-    try { verified = authorize(bearerVoiceAccessToken(request.headers), request); }
-    catch { verified = null; }
-    if (!verified || typeof verified.tenant !== "string" || !verified.tenant) {
+    const verified = principalFor(request);
+    if (!verified) {
       socket.write("HTTP/1.1 401 Unauthorized\r\nConnection: close\r\n\r\n");
       socket.destroy();
       return;

package/src/server/voiceProxy.js

@@ -4,13 +4,41 @@ import {
   VOICE_MAX_AUDIO_FRAME_BYTES,
   VOICE_MAX_CONTROL_BYTES,
   assertVoiceAudioFrame,
-  parseVoiceClientControl
+  parseVoiceClientControl,
+  voiceServerMessage
 } from "../shared/protocol.js";
-import { normalizeVoiceCloseMetadata } from "./voiceSocketClose.js";
 
 const WEBSOCKET_CONNECTING = 0;
 const WEBSOCKET_OPEN = 1;
 
+const RESERVED_CLOSE_CODES = new Set([1004, 1005, 1006, 1015]);
+
+function proxyCloseCode(code) {
+  const numericCode = Number(code);
+  const isStandardCode = Number.isInteger(numericCode)
+    && numericCode >= 1000
+    && numericCode <= 1014
+    && !RESERVED_CLOSE_CODES.has(numericCode);
+  const isApplicationCode = Number.isInteger(numericCode)
+    && numericCode >= 3000
+    && numericCode <= 4999;
+  return isStandardCode || isApplicationCode ? numericCode : 1000;
+}
+
+function proxyCloseReason(reason) {
+  const encoded = Buffer.from(String(reason || ""), "utf8");
+  return encoded.byteLength <= 123
+    ? encoded.toString("utf8")
+    : encoded.subarray(0, 120).toString("utf8");
+}
+
+function normalizeVoiceCloseMetadata(code, reason) {
+  return Object.freeze({
+    code: proxyCloseCode(code),
+    reason: proxyCloseReason(reason)
+  });
+}
+
 async function readVoiceCatalogue(config, { fetchImpl = fetch } = {}) {
   try {
     if (!config?.available) throw new Error("Voice is not configured.");
@@ -39,7 +67,7 @@ function closeSocket(socket, code, reason) {
 
 function sendSocketError(socket, code, message) {
   if (socket?.readyState === WEBSOCKET_OPEN) {
-    socket.send(JSON.stringify({ code, message, type: "error" }));
+    socket.send(voiceServerMessage("error", { code, message }));
   }
 }
 

package/src/shared/protocol.js

@@ -5,6 +5,7 @@ const VOICE_OUTPUT_SAMPLE_RATE = 22_050;
 const VOICE_MAX_CONTROL_BYTES = 16 * 1024;
 const VOICE_MAX_AUDIO_FRAME_BYTES = 64 * 1024;
 const VOICE_MAX_UTTERANCE_SECONDS = 60;
+const VOICE_DEFAULT_SYNTHESIS_TIMEOUT_MS = 20_000;
 const VOICE_MAX_SYNTHESIS_TIMEOUT_MS = 120_000;
 const VOICE_MAX_SPEECH_TEXT_CHARACTERS = 4_000;
 const VOICE_MAX_SPEECH_CHUNK_CHARACTERS = 280;
@@ -12,6 +13,7 @@ const VOICE_SPEECH_EM_DASH_PAUSE_MS = 140;
 const VOICE_SPEECH_SENTENCE_PAUSE_MS = 220;
 const VOICE_SPEECH_LINE_PAUSE_MS = 350;
 const VOICE_SPEECH_PARAGRAPH_PAUSE_MS = 650;
+const VOICE_ID_PATTERN = /^[A-Za-z0-9][A-Za-z0-9_-]{0,63}$/u;
 const VOICE_TURN_ID_PATTERN = /^[a-zA-Z0-9][a-zA-Z0-9._:-]{0,127}$/u;
 const VOICE_CLIENT_CONTROL_TYPES = new Set([
   "cancel",
@@ -95,7 +97,7 @@ function parseVoiceClientControl(raw = "") {
     return Object.freeze({ type, turnId, revision: input.revision });
   }
   if (type === "speak.start" || type === "speak.append") {
-    if (input.voiceId !== undefined && (type !== "speak.start" || typeof input.voiceId !== "string" || !/^[A-Za-z0-9][A-Za-z0-9_-]{0,63}$/u.test(input.voiceId))) {
+    if (input.voiceId !== undefined && (type !== "speak.start" || typeof input.voiceId !== "string" || !VOICE_ID_PATTERN.test(input.voiceId))) {
       throw voiceProtocolError("voice_selection_invalid", "Choose an available voice before starting speech.");
     }
     const text = String(input.text || "");
@@ -150,6 +152,56 @@ function float32ToPcm16Le(samples = new Float32Array()) {
   return bytes;
 }
 
+const BUILT_IN_RECOGNITION_PHRASES = Object.freeze([
+  [/\bc plus plus\b/giu, "C++"],
+  [/\b(?:j|jay)\s+(?:s|ess)\s+kit\b/giu, "JSKIT"],
+  [/\bnode\s+(?:(?:j|jay)\s+(?:s|ess)|js)\b/giu, "Node.js"],
+  [/\bpostgres\s+(?:s\s+)?q\s+l\b/giu, "PostgreSQL"],
+  [/\bweb socket\b/giu, "WebSocket"],
+  [/\bsherpa onnx\b/giu, "sherpa-onnx"]
+]);
+const BUILT_IN_RECOGNITION_SPELLINGS = Object.freeze({
+  acl: "ACL",
+  ai: "AI",
+  api: "API",
+  cmake: "CMake",
+  codex: "Codex",
+  env: "ENV",
+  erd: "ERD",
+  fastify: "Fastify",
+  git: "Git",
+  github: "GitHub",
+  http: "HTTP",
+  https: "HTTPS",
+  hmac: "HMAC",
+  id: "ID",
+  javascript: "JavaScript",
+  jskit: "JSKIT",
+  json: "JSON",
+  mariadb: "MariaDB",
+  mysql: "MySQL",
+  npm: "npm",
+  onnx: "ONNX",
+  opencode: "OpenCode",
+  openai: "OpenAI",
+  playwright: "Playwright",
+  postgresql: "PostgreSQL",
+  pty: "PTY",
+  sql: "SQL",
+  systemd: "systemd",
+  typescript: "TypeScript",
+  uuid: "UUID",
+  vitest: "Vitest",
+  vue: "Vue",
+  vuetify: "Vuetify",
+  websocket: "WebSocket",
+  zipformer: "Zipformer"
+});
+const BUILT_IN_RECOGNITION_WORD_PATTERN = new RegExp(
+  `\\b(?:${Object.keys(BUILT_IN_RECOGNITION_SPELLINGS).join("|")})\\b`,
+  "giu"
+);
+
 function normalizeRecognizedText(value = "", vocabulary = {}) {
   const text = String(value || "").replace(/\s+/gu, " ").trim();
   if (!text || text !== text.toUpperCase() || !/[A-Z]/u.test(text)) {
@@ -158,50 +210,13 @@ function normalizeRecognizedText(value = "", vocabulary = {}) {
   const normalized = text.toLowerCase().replace(/(^|[.!?]\s+)([a-z])/gu, (_match, prefix, letter) => (
     `${prefix}${letter.toUpperCase()}`
   ));
-  let result = normalized
-    .replace(/\bc plus plus\b/giu, "C++")
-    .replace(/\b(?:j|jay)\s+(?:s|ess)\s+kit\b/giu, "JSKIT")
-    .replace(/\bnode\s+(?:(?:j|jay)\s+(?:s|ess)|js)\b/giu, "Node.js")
-    .replace(/\bpostgres\s+(?:s\s+)?q\s+l\b/giu, "PostgreSQL")
-    .replace(/\bweb socket\b/giu, "WebSocket")
-    .replace(/\bsherpa onnx\b/giu, "sherpa-onnx")
-    .replace(/\b(?:ai|sql|api|json|http|https|mysql|mariadb|postgresql|erd|id|uuid|env|pty|acl|hmac|onnx|npm|jskit|opencode|openai|javascript|typescript|vue|vuetify|fastify|playwright|vitest|github|git|systemd|cmake|codex|zipformer|websocket)\b/giu, (word) => ({
-    acl: "ACL",
-    ai: "AI",
-    api: "API",
-    cmake: "CMake",
-    codex: "Codex",
-    env: "ENV",
-    erd: "ERD",
-    fastify: "Fastify",
-    git: "Git",
-    github: "GitHub",
-    http: "HTTP",
-    https: "HTTPS",
-    hmac: "HMAC",
-    id: "ID",
-    javascript: "JavaScript",
-    jskit: "JSKIT",
-    json: "JSON",
-    mariadb: "MariaDB",
-    mysql: "MySQL",
-    npm: "npm",
-    onnx: "ONNX",
-    opencode: "OpenCode",
-    openai: "OpenAI",
-    playwright: "Playwright",
-    postgresql: "PostgreSQL",
-    pty: "PTY",
-    sql: "SQL",
-    systemd: "systemd",
-    typescript: "TypeScript",
-    uuid: "UUID",
-    vitest: "Vitest",
-    vue: "Vue",
-    vuetify: "Vuetify",
-    websocket: "WebSocket",
-    zipformer: "Zipformer"
-  })[word.toLowerCase()] || word);
+  let result = normalized;
+  for (const [pattern, spelling] of BUILT_IN_RECOGNITION_PHRASES) {
+    result = result.replace(pattern, spelling);
+  }
+  result = result.replace(BUILT_IN_RECOGNITION_WORD_PATTERN, (word) => (
+    BUILT_IN_RECOGNITION_SPELLINGS[word.toLowerCase()] || word
+  ));
   for (const [phrase, spelling] of Object.entries(vocabulary)) {
     const literal = phrase.replace(/[.*+?^${}()|[\]\\]/gu, "\\$&");
     result = result.replace(new RegExp(`\\b${literal}\\b`, "giu"), () => spelling);
@@ -416,6 +431,8 @@ function voiceServerMessage(type = "", fields = {}) {
 
 export {
   VOICE_DAEMON_WEBSOCKET_PATH,
+  VOICE_DEFAULT_SYNTHESIS_TIMEOUT_MS,
+  VOICE_ID_PATTERN,
   VOICE_INPUT_SAMPLE_RATE,
   VOICE_MAX_AUDIO_FRAME_BYTES,
   VOICE_MAX_CONTROL_BYTES,

package/src/server/voiceSocketClose.js

@@ -1,31 +0,0 @@
-const RESERVED_CLOSE_CODES = new Set([1004, 1005, 1006, 1015]);
-
-function proxyCloseCode(code) {
-  const numericCode = Number(code);
-  const isStandardCode = Number.isInteger(numericCode)
-    && numericCode >= 1000
-    && numericCode <= 1014
-    && !RESERVED_CLOSE_CODES.has(numericCode);
-  const isApplicationCode = Number.isInteger(numericCode)
-    && numericCode >= 3000
-    && numericCode <= 4999;
-  return isStandardCode || isApplicationCode ? numericCode : 1000;
-}
-
-function proxyCloseReason(reason) {
-  const encoded = Buffer.from(String(reason || ""), "utf8");
-  return encoded.byteLength <= 123
-    ? encoded.toString("utf8")
-    : encoded.subarray(0, 120).toString("utf8");
-}
-
-function normalizeVoiceCloseMetadata(code, reason) {
-  return Object.freeze({
-    code: proxyCloseCode(code),
-    reason: proxyCloseReason(reason)
-  });
-}
-
-export {
-  normalizeVoiceCloseMetadata
-};