fix(voice): Gemini voice messages follow a language change

The Gemini session translated each status and error message once, when it
happened, so after a language change the voice control kept showing it in
the old language. The state event now also carries the English catalog text
as sourceDetail, and the control words it when shown. The local server's
known token errors stay catalog text until then instead of being translated
on arrival.

docs/CURRENT-STATE.md

@@ -1135,7 +1135,7 @@ Source factories have dedicated `layers/<family>/source` exports. ALPR and earth
 
 Voice controls bind to a protocol-independent session factory. The default WebRTC adapter preserves the existing Realtime connection, push-to-talk, cost controls and radio handoff. Session subscriptions expose state, transcript, action call/progress/result, interruption, completion and turn-metrics events; stopping or replacing a session cancels pending actions. An active session's normal Stop publishes one idle transition, including adapters that emit their own terminal state; a restart triggered by that transition keeps ownership of its new startup state. Alternate adapters can use the same controls and action runner.
 
-The standalone app offers OpenAI and Gemini Live through the **AI Provider** setting, shared by voice and typed chat. Gemini keeps its permanent API key on the local server and mints a short-lived, single-use credential with server-owned instructions and the standalone action catalog. With no Gemini key the token route answers `503` with code `API_KEY_MISSING`, and the voice control says `Add a Gemini API key in Provider Settings` without the microphone-permission hint. Its native WebSocket adapter supports open microphone and push-to-talk input, streaming audio, captions and the existing action runner. In push-to-talk, a tool call that arrives with no input, reply or tool result outstanding is answered with a failure and never runs, and neither does any other call in the reply it starts; open-microphone turns start on Gemini's server-side activity detection, which reports no boundary to check. Switching providers ends the old conversation and releases its microphone and connection; only the input mode carries over. Gemini uses bounded shutdown draining and explicit manual restart, without automatic reconnection or a promise of uninterrupted session duration. Hosts that do not opt into the provider facade retain their existing default or custom adapter.
+The standalone app offers OpenAI and Gemini Live through the **AI Provider** setting, shared by voice and typed chat. Gemini keeps its permanent API key on the local server and mints a short-lived, single-use credential with server-owned instructions and the standalone action catalog. With no Gemini key the token route answers `503` with code `API_KEY_MISSING`, and the voice control says `Add a Gemini API key in Provider Settings` without the microphone-permission hint. Gemini's status and error messages keep their English catalog text on the session's `state` event (`sourceDetail`, beside the worded `detail`), so the voice control words them again after a language change. Its native WebSocket adapter supports open microphone and push-to-talk input, streaming audio, captions and the existing action runner. In push-to-talk, a tool call that arrives with no input, reply or tool result outstanding is answered with a failure and never runs, and neither does any other call in the reply it starts; open-microphone turns start on Gemini's server-side activity detection, which reports no boundary to check. Switching providers ends the old conversation and releases its microphone and connection; only the input mode carries over. Gemini uses bounded shutdown draining and explicit manual restart, without automatic reconnection or a promise of uninterrupted session duration. Hosts that do not opt into the provider facade retain their existing default or custom adapter.
 
 Voice action argument schemas have one portable owner under `src/voice/actionSchemas.js`. The Realtime provider builds 33 tools using separate description-only metadata; a session is offered 32 of them unless `OVERPASS_UPSTREAMS` is set, which adds `osm_query` and its one prompt sentence (`server/providers/openai/tools.js`). Description customization cannot replace argument types, enum values or required fields.
 

src/voice/geminiSession.js

@@ -71,7 +71,9 @@ export function geminiSocketUrl({ token, websocketUrl }) {
   return url.href;
 }
 
-// Known local-server failures are translated; unexpected diagnostics stay intact.
+// The local server's known failures, marked for translation. Each is kept as
+// its English catalog text, which the status words when it is shown, so a
+// language change words it again; unexpected diagnostics stay intact.
 function tokenErrorMessage(error) {
   switch (error) {
     case msg('Gemini voice could not connect. Try again.'):
@@ -85,7 +87,6 @@ function tokenErrorMessage(error) {
     case msg('Token request is too large'):
     case msg('Use an empty JSON token request'):
     case msg('Method not allowed'):
-      return t(error);
     default:
       return error;
   }
@@ -369,10 +370,14 @@ export class GeminiSessionController {
       detail = this.input.pushToTalkKeyHeld
         ? msg('Release Space to send')
         : msg('Hold Space to talk');
+    // `detail` is worded in the current language; `sourceDetail` keeps the
+    // English catalog text, so a view can word it again after a language
+    // change.
     const event = {
       type: 'state',
       state: status,
       detail: translateKnown(detail),
+      ...(typeof detail === 'string' && { sourceDetail: detail }),
     };
     if (recovery) event.recovery = translateKnown(recovery);
     if (missingKey) event.missingKey = true;

src/voice/geminiSession.test.mjs

@@ -4,6 +4,7 @@ import { geminiSocketUrl } from './geminiSession.js';
 import { createVoiceSession } from './session.js';
 import { createPointerTracker } from './pointerContext.js';
 import { withToolCatalog } from './gevRealtime.js';
+import { setLocale, t } from '../i18n/index.js';
 import {
   harness,
   tick,
@@ -3662,3 +3663,28 @@ test('a missing server key reaches the error state flagged, other token failures
     assert.equal(h.sockets.length, 0);
   }
 });
+
+test('an error keeps its English catalog text beside the worded detail', async (t2) => {
+  t2.after(() => setLocale('en'));
+  setLocale('es');
+  const h = harness({
+    fetchImpl: async () =>
+      Response.json(
+        { error: 'Gemini quota is unavailable. Try again later.' },
+        { status: 503 },
+      ),
+  });
+  await h.c.start();
+  const error = h.events
+    .filter((event) => event.type === 'state' && event.state === 'error')
+    .at(-1);
+  assert.equal(
+    error.sourceDetail,
+    'Gemini quota is unavailable. Try again later.',
+  );
+  assert.equal(
+    error.detail,
+    t('Gemini quota is unavailable. Try again later.'),
+  );
+  assert.notEqual(error.detail, error.sourceDetail);
+});

src/voice/providerCommands.test.mjs

@@ -11,6 +11,7 @@ import { createGevActionRunner } from './gevActions.js';
 import { createPointerTracker } from './pointerContext.js';
 import { createReferentRegistry } from './referents.js';
 import { createTestVoiceUi } from '../testSupport/voiceUi.mjs';
+import { setLocale, t } from '../i18n/index.js';
 
 const PROVIDER_KEY = 'gev.voice.provider';
 const settle = () => new Promise((resolve) => setImmediate(resolve));
@@ -897,3 +898,33 @@ test('a missing-key error hides the microphone hint; other errors keep it', asyn
   assert.equal(ui.state.error.text, 'Socket closed');
   assert.equal(ui.state.error.hint, true);
 });
+
+test('a provider message is worded again after a language change', async (t2) => {
+  t2.after(() => setLocale('en'));
+  const f = compositionFixture(t2, {});
+  f.select('gemini');
+  await settle();
+  const instance = f.instances.at(-1);
+  const ui = f.views.at(-1);
+  const quota = 'Gemini quota is unavailable. Try again later.';
+  instance.hooks.emit({
+    type: 'state',
+    state: 'error',
+    detail: quota,
+    sourceDetail: quota,
+  });
+  assert.equal(ui.state.error.text, quota);
+  setLocale('es');
+  assert.equal(ui.state.error.text, t(quota));
+  assert.notEqual(ui.state.error.text, quota);
+  setLocale('en');
+  instance.hooks.emit({
+    type: 'state',
+    state: 'listening',
+    detail: 'Ask or command',
+    sourceDetail: 'Ask or command',
+  });
+  setLocale('es');
+  assert.equal(ui.state.detail.text, t('Ask or command'));
+  assert.notEqual(ui.state.detail.text, 'Ask or command');
+});

src/voice/sessionCommands.js

@@ -7,7 +7,7 @@ import { createVoiceSession, isStoppable } from './session.js';
 import { VoiceCardControls } from './voiceCard.js';
 import { PointerReticle } from './pointerReticle.js';
 import { VOICE_PROVIDERS } from './providerCapabilities.js';
-import { t } from '../i18n/index.js';
+import { t, translateKnown } from '../i18n/index.js';
 
 function createDefaultReticle(pointer) {
   const element = pointer?.element;
@@ -114,11 +114,18 @@ export function createVoiceCommands({
     // States are machine values; only the label is translated.
     ui.update('status', { text: () => voiceStatusLabel(event.state) });
     // The portable session reports 'Voice off' in English: show the idle
-    // label instead. Other details are provider messages, shown as they come.
+    // label instead. Other details are provider messages; one that keeps its
+    // English catalog text is worded again in the current language.
     const idle = event.state === 'idle';
-    const detail = event.detail === 'Voice off' ? '' : event.detail;
+    const source = event.sourceDetail ?? event.detail;
+    const detail = () =>
+      source === 'Voice off'
+        ? ''
+        : event.sourceDetail
+          ? translateKnown(event.sourceDetail)
+          : event.detail;
     ui.update('detail', {
-      text: () => detail || (idle ? t('Voice off') : t('Voice active')),
+      text: () => detail() || (idle ? t('Voice off') : t('Voice active')),
     });
     // A typed chat session is not the microphone.
     ui.update('button', {
@@ -127,7 +134,7 @@ export function createVoiceCommands({
     ui.update('error', {
       text: () =>
         event.state === 'error'
-          ? event.detail || t('Voice could not be started.')
+          ? detail() || t('Voice could not be started.')
           : '',
       // "Check microphone permission" is wrong advice for a missing key.
       hint: () => !(event.state === 'error' && event.missingKey === true),