dsh-live-voice 0.8.0 → 0.8.1

package/README.md

@@ -4,7 +4,7 @@ Local-first speech-to-text (STT), text-to-speech (TTS), and meeting audio transc
 
 ## Installation
 
-Tested with **DeepSeek Harness v0.2.2-rc.1**. The package manifest's `dshTestedVersion` field records the tested version.
+Tested with **DeepSeek Harness v0.2.2-rc.3**. The package manifest's `dshTestedVersion` field records the tested version.
 
 Install the prebuilt package from npm; no source build is required. The Git repository contains source only.
 

package/lib/asr/asr.worker.js

@@ -36682,7 +36682,7 @@ async function createQwenRuntime(options) {
 }
 
 // src/modules/recognition/local/assets.ts
-var browserAsrAssetBase = () => "/api/dsh-live-voice/asr/" + encodeURIComponent(true ? "0.8.0" : "test") + "/";
+var browserAsrAssetBase = () => "/api/dsh-live-voice/asr/" + encodeURIComponent(true ? "0.8.1" : "test") + "/";
 
 // src/modules/recognition/local/modelCatalog.ts
 var browserAsrModels = [

package/lib/asr/manifest.json

@@ -1,5 +1,5 @@
 {
-  "version": "0.8.0",
+  "version": "0.8.1",
   "assets": [
     "asr.worker.js",
     "manifest.json",

package/lib/client.js

@@ -36,89 +36,6 @@ __export(apply_exports, {
 module.exports = __toCommonJS(apply_exports);
 var import_react80 = __toESM(require("react"), 1);
 
-// src/modules/conversation/models/inactivity.ts
-var VoiceInactivity = class {
-  constructor(options) {
-    this.options = options;
-  }
-  options;
-  snapshot = { phase: "idle", remaining: 0 };
-  listeners = /* @__PURE__ */ new Set();
-  active = false;
-  lastText = 0;
-  lastTick = 0;
-  deadline = 0;
-  busySince = 0;
-  warningText = "";
-  echoUntil = 0;
-  now = () => this.options.now?.() ?? Date.now();
-  getSnapshot = () => this.snapshot;
-  subscribe = (listener) => {
-    this.listeners.add(listener);
-    return () => {
-      this.listeners.delete(listener);
-    };
-  };
-  patch(phase, remaining = 0) {
-    if (this.snapshot.phase === phase && this.snapshot.remaining === remaining) return;
-    this.snapshot = { phase, remaining };
-    for (const listener of this.listeners) listener();
-  }
-  keepAlive = () => {
-    this.lastText = this.now();
-    this.deadline = 0;
-    this.busySince = 0;
-    this.patch("idle");
-  };
-  /** Filter only matching warning echo; unrelated replies remain eligible. */
-  recognized(text) {
-    const normalize = (value) => value.toLocaleLowerCase().replace(/[^\p{L}\p{N}]+/gu, " ").trim();
-    if (!text.trim()) return true;
-    const heard = normalize(text);
-    const own = normalize(this.warningText);
-    if (this.now() < this.echoUntil && heard.length >= 24 && heard.length >= own.length / 2 && own.includes(heard))
-      return false;
-    this.keepAlive();
-    return true;
-  }
-  tick(active, busy, settings) {
-    const now2 = this.now();
-    const gap = this.lastTick && now2 - this.lastTick > 1e4;
-    this.lastTick = now2;
-    if (!settings.autoEndEnabled || !active) {
-      this.active = false;
-      if (this.snapshot.phase !== "ended" || active) this.patch("idle");
-      return;
-    }
-    if (!this.active) {
-      this.active = true;
-      this.keepAlive();
-    }
-    const total = settings.autoEndMinutes * 6e4;
-    const warning = settings.autoEndWarningMinutes * 6e4;
-    if (gap && now2 - this.lastText >= total - warning) this.lastText = now2 - (total - warning);
-    if (now2 - this.lastText < total - warning) return;
-    if (!this.deadline || gap) {
-      this.deadline = Math.max(this.lastText + total, now2 + warning);
-      this.patch("warning", Math.ceil((this.deadline - now2) / 1e3));
-      this.echoUntil = now2 + 3e4;
-      this.warningText = this.options.warn(this.snapshot.remaining);
-    }
-    this.patch("warning", Math.max(0, Math.ceil((this.deadline - now2) / 1e3)));
-    if (now2 < this.deadline) return;
-    if (busy) {
-      if (!this.busySince) this.busySince = now2;
-      if (now2 - this.busySince < 3e4) return;
-    }
-    this.active = false;
-    this.patch("ended");
-    this.options.end();
-  }
-};
-
-// src/modules/conversation/components/InactivityNotice.tsx
-var import_react3 = __toESM(require("react"), 1);
-
 // src/app/client/i18n/catalogs/en.ts
 var en = {
   "dsh-live-voice.commons.connection.contactingHost": "Contacting DSH host\u2026",
@@ -3197,6 +3114,143 @@ var appDictionaries = Object.freeze({
   zh: zh_default
 });
 
+// src/app/client/i18n/registerDshLocales.ts
+var LIVE_VOICE_LOCALE_NAMESPACE = "dsh-live-voice";
+var LIVE_VOICE_LANGUAGES = liveVoiceLanguageDefinitions;
+var liveVoiceDictionaries = appDictionaries;
+function createFallbackTranslator(locale = "en") {
+  return (key, params = {}) => {
+    const template = liveVoiceDictionaries[locale]?.[key] ?? liveVoiceDictionaries.en[key] ?? key;
+    return String(template).replace(
+      /\{([A-Za-z0-9_]+)\}/g,
+      (_2, name) => String(params[name] ?? "{" + name + "}")
+    );
+  };
+}
+function registerDshLocales(ctx) {
+  if (!ctx.locale) return createFallbackTranslator();
+  const registeredLanguages = new Set(
+    (ctx.locale.getSnapshot?.().locales || []).map(
+      (language) => language.id.toLowerCase()
+    )
+  );
+  for (const language of LIVE_VOICE_LANGUAGES) {
+    if (registeredLanguages.has(language.id.toLowerCase())) continue;
+    ctx.effect(
+      () => ctx.locale.addLanguage(language),
+      "dsh-live-voice: " + language.id + " language"
+    );
+    registeredLanguages.add(language.id.toLowerCase());
+  }
+  for (const [locale, dictionary] of Object.entries(liveVoiceDictionaries))
+    ctx.effect(
+      () => ctx.locale.register(LIVE_VOICE_LOCALE_NAMESPACE, locale, dictionary),
+      "dsh-live-voice: " + locale + " dictionary"
+    );
+  return ctx.locale.bind(LIVE_VOICE_LOCALE_NAMESPACE);
+}
+var registerLiveVoiceLocales = registerDshLocales;
+
+// src/app/client/i18n/spokenWarning.ts
+function warningLanguage(language) {
+  const value = language?.trim().replaceAll("_", "-").toLowerCase();
+  if (!value) return void 0;
+  if (value === "pt" || value.startsWith("pt-")) return "pt-BR";
+  const base = value.split("-")[0];
+  return base && base in appDictionaries ? base : void 0;
+}
+function spokenInactivityWarning(ttsLanguage, uiLanguage, seconds) {
+  const spoken = warningLanguage(ttsLanguage);
+  const locale = spoken ?? warningLanguage(uiLanguage) ?? "en";
+  return {
+    text: createFallbackTranslator(locale)("dsh-live-voice.settings.autoEnd.warning", { seconds }),
+    lang: spoken ? ttsLanguage.replaceAll("_", "-") : locale
+  };
+}
+
+// src/modules/conversation/models/inactivity.ts
+var VoiceInactivity = class {
+  constructor(options) {
+    this.options = options;
+  }
+  options;
+  snapshot = { phase: "idle", remaining: 0 };
+  listeners = /* @__PURE__ */ new Set();
+  active = false;
+  lastText = 0;
+  lastTick = 0;
+  deadline = 0;
+  busySince = 0;
+  warningText = "";
+  echoUntil = 0;
+  now = () => this.options.now?.() ?? Date.now();
+  getSnapshot = () => this.snapshot;
+  subscribe = (listener) => {
+    this.listeners.add(listener);
+    return () => {
+      this.listeners.delete(listener);
+    };
+  };
+  patch(phase, remaining = 0) {
+    if (this.snapshot.phase === phase && this.snapshot.remaining === remaining) return;
+    this.snapshot = { phase, remaining };
+    for (const listener of this.listeners) listener();
+  }
+  keepAlive = () => {
+    this.lastText = this.now();
+    this.deadline = 0;
+    this.busySince = 0;
+    this.patch("idle");
+  };
+  /** Filter only matching warning echo; unrelated replies remain eligible. */
+  recognized(text) {
+    const normalize = (value) => value.toLocaleLowerCase().replace(/[^\p{L}\p{N}]+/gu, " ").trim();
+    if (!text.trim()) return true;
+    const heard = normalize(text);
+    const own = normalize(this.warningText);
+    if (this.now() < this.echoUntil && heard.length >= 24 && heard.length >= own.length / 2 && own.includes(heard))
+      return false;
+    this.keepAlive();
+    return true;
+  }
+  tick(active, busy, settings) {
+    const now2 = this.now();
+    const gap = this.lastTick && now2 - this.lastTick > 1e4;
+    this.lastTick = now2;
+    if (!settings.autoEndEnabled || !active) {
+      this.active = false;
+      if (this.snapshot.phase !== "ended" || active) this.patch("idle");
+      return;
+    }
+    if (!this.active) {
+      this.active = true;
+      this.keepAlive();
+    }
+    const total = settings.autoEndMinutes * 6e4;
+    const warning = settings.autoEndWarningMinutes * 6e4;
+    if (gap && now2 - this.lastText >= total - warning) this.lastText = now2 - (total - warning);
+    if (now2 - this.lastText < total - warning) return;
+    if (!this.deadline || gap) {
+      this.deadline = Math.max(this.lastText + total, now2 + warning);
+      this.patch("warning", Math.ceil((this.deadline - now2) / 1e3));
+      this.echoUntil = now2 + 3e4;
+      this.warningText = this.options.warn(this.snapshot.remaining);
+    }
+    this.patch("warning", Math.max(0, Math.ceil((this.deadline - now2) / 1e3)));
+    if (now2 < this.deadline) return;
+    if (busy) {
+      if (!this.busySince) this.busySince = now2;
+      if (now2 - this.busySince < 3e4) return;
+    }
+    this.active = false;
+    this.patch("ended");
+    this.options.end();
+  }
+};
+
+// src/modules/conversation/components/InactivityNotice.tsx
+var import_react3 = __toESM(require("react"), 1);
+
 // src/app/client/i18n/DshLanguageBoundary.tsx
 var import_react2 = __toESM(require("react"), 1);
 
@@ -3261,43 +3315,6 @@ function createTranslationRuntime({ availableLangs: availableLangs2, defaultLang
   return { TranslationProvider, useTranslation };
 }
 
-// src/app/client/i18n/registerDshLocales.ts
-var LIVE_VOICE_LOCALE_NAMESPACE = "dsh-live-voice";
-var LIVE_VOICE_LANGUAGES = liveVoiceLanguageDefinitions;
-var liveVoiceDictionaries = appDictionaries;
-function createFallbackTranslator(locale = "en") {
-  return (key, params = {}) => {
-    const template = liveVoiceDictionaries[locale]?.[key] ?? liveVoiceDictionaries.en[key] ?? key;
-    return String(template).replace(
-      /\{([A-Za-z0-9_]+)\}/g,
-      (_2, name) => String(params[name] ?? "{" + name + "}")
-    );
-  };
-}
-function registerDshLocales(ctx) {
-  if (!ctx.locale) return createFallbackTranslator();
-  const registeredLanguages = new Set(
-    (ctx.locale.getSnapshot?.().locales || []).map(
-      (language) => language.id.toLowerCase()
-    )
-  );
-  for (const language of LIVE_VOICE_LANGUAGES) {
-    if (registeredLanguages.has(language.id.toLowerCase())) continue;
-    ctx.effect(
-      () => ctx.locale.addLanguage(language),
-      "dsh-live-voice: " + language.id + " language"
-    );
-    registeredLanguages.add(language.id.toLowerCase());
-  }
-  for (const [locale, dictionary] of Object.entries(liveVoiceDictionaries))
-    ctx.effect(
-      () => ctx.locale.register(LIVE_VOICE_LOCALE_NAMESPACE, locale, dictionary),
-      "dsh-live-voice: " + locale + " dictionary"
-    );
-  return ctx.locale.bind(LIVE_VOICE_LOCALE_NAMESPACE);
-}
-var registerLiveVoiceLocales = registerDshLocales;
-
 // src/app/client/i18n/runtime.tsx
 function interpolate(template, params = {}) {
   return template.replace(
@@ -3497,7 +3514,7 @@ var SessionSpeechMemory = class {
 };
 
 // src/modules/recognition/vad/assets.ts
-var sileroVadAssetBase = () => "/api/dsh-live-voice/vad/" + encodeURIComponent(true ? "0.8.0" : "test") + "/";
+var sileroVadAssetBase = () => "/api/dsh-live-voice/vad/" + encodeURIComponent(true ? "0.8.1" : "test") + "/";
 async function sileroVadAvailable(fetchImpl = globalThis.fetch) {
   try {
     const response = await fetchImpl(sileroVadAssetBase() + "availability", {
@@ -6907,7 +6924,7 @@ async function removeBrowserModel(key, storage = caches) {
 }
 
 // src/modules/recognition/local/assets.ts
-var browserAsrAssetBase = () => "/api/dsh-live-voice/asr/" + encodeURIComponent(true ? "0.8.0" : "test") + "/";
+var browserAsrAssetBase = () => "/api/dsh-live-voice/asr/" + encodeURIComponent(true ? "0.8.1" : "test") + "/";
 
 // src/modules/recognition/local/runtimeIdentity.ts
 var workerProtocol = "asr-worker-v3-ort-1.31.0-dev.20260914-8d85527a0";
@@ -11551,9 +11568,9 @@ var VoiceCoordinator = class _VoiceCoordinator {
     }
     this.prefetchItems.clear();
   }
-  _speechPreparationKey() {
+  _speechPreparationKey(item = {}) {
     const { engine, voice, lang, rate, outputDeviceId } = this.snapshot.settings;
-    return JSON.stringify([engine, voice, lang, rate, outputDeviceId]);
+    return JSON.stringify([engine, voice, item.lang ?? lang, rate, outputDeviceId]);
   }
   _resetStaleSpeechItem(item, key) {
     if (item.preparationKey === key) return;
@@ -11574,13 +11591,13 @@ var VoiceCoordinator = class _VoiceCoordinator {
       lang: this.snapshot.settings.lang,
       outputDeviceId: this.snapshot.settings.outputDeviceId
     };
-    const key = this._speechPreparationKey();
     for (const item of this.queue.slice(0, SPEECH_QUEUE_PREFETCH_LIMIT)) {
+      const key = this._speechPreparationKey(item);
       this._resetStaleSpeechItem(item, key);
       if (item.prepared || item.preparing) continue;
       this.prefetchItems.add(item);
       const signal = this.prefetchController.signal;
-      item.preparing = engine.prepare(item.text, { ...options, signal }).then((prepared) => {
+      item.preparing = engine.prepare(item.text, { ...options, lang: item.lang ?? options.lang, signal }).then((prepared) => {
         if (item.preparationKey === key) item.preparing = null;
         if (signal.aborted || item.preparationKey !== key || !this.queue.includes(item) && !this.speechHistory.includes(item)) {
           prepared.dispose?.();
@@ -11957,15 +11974,15 @@ var VoiceCoordinator = class _VoiceCoordinator {
     this._syncSpeechSegments();
     return this.play(first, messageId, true);
   }
-  async play(text, messageId, manual = false) {
+  async play(text, messageId, manual = false, speechOptions = {}) {
     if (this.disposed || !text.trim()) return;
     if (this.snapshot.speaking) {
-      this.queue.push({ text, id: messageId, manual });
+      this.queue.push({ text, id: messageId, manual, lang: speechOptions.lang });
       this._prefetchSpeech();
       this._syncSpeechSegments();
       return;
     }
-    const item = this.activeSpeechItem?.text === text ? this.activeSpeechItem : { text, id: messageId, manual };
+    const item = this.activeSpeechItem?.text === text ? this.activeSpeechItem : { text, id: messageId, manual, lang: speechOptions.lang };
     if (!this.speechHistory.includes(item)) this.speechHistory.push(item);
     this.speechCursor = this.speechHistory.indexOf(item);
     const epoch = ++this.speechEpoch;
@@ -11984,7 +12001,7 @@ var VoiceCoordinator = class _VoiceCoordinator {
     let playbackStarted = false;
     let playbackFinished = false;
     const engine = this.engines[this.snapshot.settings.engine];
-    this._resetStaleSpeechItem(item, this._speechPreparationKey());
+    this._resetStaleSpeechItem(item, this._speechPreparationKey(item));
     item.engine = engine;
     if (!engine) {
       this.queue.unshift(item);
@@ -12040,6 +12057,7 @@ var VoiceCoordinator = class _VoiceCoordinator {
             this.patch({ speechLoading: false });
           }
         },
+        lang: item.lang ?? this.snapshot.settings.lang,
         voice: this.snapshot.settings.voice || void 0,
         rate: this.snapshot.settings.rate,
         volume: this.snapshot.settings.volume,
@@ -15246,8 +15264,8 @@ function createDeviceOptions(devices, kind, systemDefault, fallback) {
 var import_react42 = __toESM(require("react"), 1);
 
 // src/modules/settings/services/releases.ts
-var CURRENT_VERSION = "0.8.0";
-var TESTED_DSH_VERSION = "0.2.2-rc.1";
+var CURRENT_VERSION = "0.8.1";
+var TESTED_DSH_VERSION = "0.2.2-rc.3";
 var REPOSITORY_URL = "https://github.com/victorwads/dsh-live-voice";
 var RELEASES_URL = `${REPOSITORY_URL}/releases`;
 var LATEST_RELEASE_API_URL = "https://api.github.com/repos/victorwads/dsh-live-voice/releases/latest";
@@ -18156,11 +18174,15 @@ function apply(ctx) {
   let inactivityEnding = false;
   const inactivity = new VoiceInactivity({
     warn: (seconds) => {
-      const text = t("dsh-live-voice.settings.autoEnd.warning", { seconds });
       const c = meetingEntry?.controller;
+      const { text, lang } = spokenInactivityWarning(
+        c?.getSnapshot().settings.lang ?? preferences.getSnapshot().lang,
+        ctx.locale?.getSnapshot?.().active,
+        seconds
+      );
       const s = c?.getSnapshot();
       if (c && !s.speaking && !s.paused && !s.speechRunActive && !c.queue.length)
-        run(c, c.play(text, null, true));
+        run(c, c.play(text, null, true, { lang }));
       return text;
     },
     end: () => {

package/lib/server.js

@@ -2066,7 +2066,7 @@ var VAD_ASSET_TYPES = {
 };
 var MAX_VAD_ASSET_BYTES = 32 * 1024 * 1024;
 function registerVadAssetRoutes(ctx, options = {}) {
-  const version = options.version ?? (true ? "0.8.0" : "test");
+  const version = options.version ?? (true ? "0.8.1" : "test");
   if (!/^[a-zA-Z0-9][a-zA-Z0-9.+-]{0,127}$/.test(version))
     throw new Error("Invalid VAD asset version");
   const directory = options.directory ?? fileURLToPath(new URL("./vad/", import.meta.url));
@@ -2160,7 +2160,7 @@ import { lstat as lstat2, readFile as readFile5 } from "node:fs/promises";
 import { fileURLToPath as fileURLToPath2 } from "node:url";
 import { join as join8 } from "node:path";
 function registerBrowserAsrAssets(ctx, options = {}) {
-  const version = options.version ?? (true ? "0.8.0" : "test");
+  const version = options.version ?? (true ? "0.8.1" : "test");
   if (!/^[a-zA-Z0-9][a-zA-Z0-9.+-]{0,127}$/.test(version)) throw new Error("invalid-version");
   const base = "/api/dsh-live-voice/asr/" + encodeURIComponent(version) + "/";
   const directory = options.directory ?? fileURLToPath2(new URL("./asr/", import.meta.url));

package/lib/vad/manifest.json

@@ -1,6 +1,6 @@
 {
   "schemaVersion": 1,
-  "pluginVersion": "0.8.0",
+  "pluginVersion": "0.8.1",
   "dependencies": {
     "@ricky0123/vad-web": "0.0.31",
     "onnxruntime-web": "1.22.0"
@@ -41,7 +41,7 @@
     },
     "vad.worker.js": {
       "bytes": 79337,
-      "sha256": "543b52073b17a67ecaa1ff72350ecd6f1f4e67399420ef961f1be5a809ab022a"
+      "sha256": "7bc4a97041cb5740bb0254387de63f69d5e869c020d9dd82096f5d8b4fbfdf06"
     }
   }
 }

package/lib/vad/vad.worker.js

@@ -1632,7 +1632,7 @@ var vo = Ke;
 Object.defineProperty(I.versions, "web", { value: $t, enumerable: true });
 
 // src/modules/recognition/vad/assets.ts
-var sileroVadAssetBase = () => "/api/dsh-live-voice/vad/" + encodeURIComponent(true ? "0.8.0" : "test") + "/";
+var sileroVadAssetBase = () => "/api/dsh-live-voice/vad/" + encodeURIComponent(true ? "0.8.1" : "test") + "/";
 
 // src/modules/recognition/vad/StreamingResampler.ts
 var StreamingResampler = class {

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "dsh-live-voice",
-  "version": "0.8.0",
-  "dshTestedVersion": "0.2.2-rc.1",
+  "version": "0.8.1",
+  "dshTestedVersion": "0.2.2-rc.3",
   "type": "module",
   "scripts": {
     "build": "node --experimental-strip-types ../../scripts/build.ts --package live-voice",

package/src/app/client/apply.tsx

@@ -1,5 +1,6 @@
 // @ts-nocheck
 import React from 'react';
+import { spokenInactivityWarning } from './i18n/spokenWarning.js';
 import { VoiceInactivity } from '../../modules/conversation/models/inactivity.js';
 import { InactivityNotice } from '../../modules/conversation/components/InactivityNotice.js';
 import {
@@ -97,12 +98,16 @@ export function apply(ctx) {
   let inactivityEnding = false;
   const inactivity = new VoiceInactivity({
     warn: (seconds) => {
-      const text = t('dsh-live-voice.settings.autoEnd.warning', { seconds });
       const c = meetingEntry?.controller;
+      const { text, lang } = spokenInactivityWarning(
+        c?.getSnapshot().settings.lang ?? preferences.getSnapshot().lang,
+        ctx.locale?.getSnapshot?.().active,
+        seconds,
+      );
       // Do not replace assistant playback or a manually paused speech queue.
       const s = c?.getSnapshot();
       if (c && !s.speaking && !s.paused && !s.speechRunActive && !c.queue.length)
-        run(c, c.play(text, null, true));
+        run(c, c.play(text, null, true, { lang }));
       return text;
     },
     end: () => {

package/src/app/client/i18n/spokenWarning.ts

@@ -0,0 +1,21 @@
+import { appDictionaries, type LiveVoiceLocale } from './catalogs/index.js';
+import { createFallbackTranslator } from './registerDshLocales.js';
+export function warningLanguage(language: string | undefined): LiveVoiceLocale | undefined {
+  const value = language?.trim().replaceAll('_', '-').toLowerCase();
+  if (!value) return undefined;
+  if (value === 'pt' || value.startsWith('pt-')) return 'pt-BR';
+  const base = value.split('-')[0];
+  return base && base in appDictionaries ? (base as LiveVoiceLocale) : undefined;
+}
+export function spokenInactivityWarning(
+  ttsLanguage: string | undefined,
+  uiLanguage: string | undefined,
+  seconds: number,
+) {
+  const spoken = warningLanguage(ttsLanguage);
+  const locale = spoken ?? warningLanguage(uiLanguage) ?? 'en';
+  return {
+    text: createFallbackTranslator(locale)('dsh-live-voice.settings.autoEnd.warning', { seconds }),
+    lang: spoken ? ttsLanguage!.replaceAll('_', '-') : locale,
+  };
+}

package/src/modules/core/coordinator.ts

@@ -1077,9 +1077,9 @@ export class VoiceCoordinator {
     this.prefetchItems.clear();
   }
 
-  _speechPreparationKey() {
+  _speechPreparationKey(item = {}) {
     const { engine, voice, lang, rate, outputDeviceId } = this.snapshot.settings;
-    return JSON.stringify([engine, voice, lang, rate, outputDeviceId]);
+    return JSON.stringify([engine, voice, item.lang ?? lang, rate, outputDeviceId]);
   }
   _resetStaleSpeechItem(item, key) {
     if (item.preparationKey === key) return;
@@ -1100,14 +1100,14 @@ export class VoiceCoordinator {
       lang: this.snapshot.settings.lang,
       outputDeviceId: this.snapshot.settings.outputDeviceId,
     };
-    const key = this._speechPreparationKey();
     for (const item of this.queue.slice(0, SPEECH_QUEUE_PREFETCH_LIMIT)) {
+      const key = this._speechPreparationKey(item);
       this._resetStaleSpeechItem(item, key);
       if (item.prepared || item.preparing) continue;
       this.prefetchItems.add(item);
       const signal = this.prefetchController.signal;
       item.preparing = engine
-        .prepare(item.text, { ...options, signal })
+        .prepare(item.text, { ...options, lang: item.lang ?? options.lang, signal })
         .then((prepared) => {
           if (item.preparationKey === key) item.preparing = null;
           if (
@@ -1542,10 +1542,10 @@ export class VoiceCoordinator {
     this._syncSpeechSegments();
     return this.play(first, messageId, true);
   }
-  async play(text, messageId, manual = false) {
+  async play(text, messageId, manual = false, speechOptions = {}) {
     if (this.disposed || !text.trim()) return;
     if (this.snapshot.speaking) {
-      this.queue.push({ text, id: messageId, manual });
+      this.queue.push({ text, id: messageId, manual, lang: speechOptions.lang });
       this._prefetchSpeech();
       this._syncSpeechSegments();
       return;
@@ -1553,7 +1553,7 @@ export class VoiceCoordinator {
     const item =
       this.activeSpeechItem?.text === text
         ? this.activeSpeechItem
-        : { text, id: messageId, manual };
+        : { text, id: messageId, manual, lang: speechOptions.lang };
     if (!this.speechHistory.includes(item)) this.speechHistory.push(item);
     this.speechCursor = this.speechHistory.indexOf(item);
     const epoch = ++this.speechEpoch;
@@ -1572,7 +1572,7 @@ export class VoiceCoordinator {
     let playbackStarted = false;
     let playbackFinished = false;
     const engine = this.engines[this.snapshot.settings.engine];
-    this._resetStaleSpeechItem(item, this._speechPreparationKey());
+    this._resetStaleSpeechItem(item, this._speechPreparationKey(item));
     item.engine = engine;
     if (!engine) {
       this.queue.unshift(item);
@@ -1635,6 +1635,7 @@ export class VoiceCoordinator {
             this.patch({ speechLoading: false });
           }
         },
+        lang: item.lang ?? this.snapshot.settings.lang,
         voice: this.snapshot.settings.voice || undefined,
         rate: this.snapshot.settings.rate,
         volume: this.snapshot.settings.volume,