feat(kiosk): audio on a display: radio stations (incl. operator-ticked home-network streams) through a pinned stream relay, display tickets, transport verbs, ducking, now playing; voice turn: optional STT prompt/hotwords, repetition-loop and tool-syntax guards; kiosk 0.3.6

bundles/kiosk/manifest.json

@@ -1,7 +1,7 @@
 {
   "id": "kiosk",
   "name": "Kiosk display",
-  "version": "0.2.2",
+  "version": "0.3.6",
   "type": "mcp-server",
   "author": "Crow",
   "category": "hardware",

bundles/kiosk/panel/kiosk.js

@@ -66,8 +66,10 @@ export const CLIENT_SCRIPT = `
     if (b.bot !== a.bot && b.bot) { p.bound_bot_id = b.bot; any = true; }
     if (b.stt !== a.stt) { p.stt_profile_id = b.stt; any = true; }
     if (b.tts !== a.tts) { p.tts_profile_id = b.tts; any = true; }
-    if (b.fu !== a.fu || b.mem !== a.mem || b.vad !== a.vad || b.sm !== a.sm || b.th !== a.th || b.pf !== a.pf) {
+    if (b.fu !== a.fu || b.mem !== a.mem || b.vad !== a.vad || b.sm !== a.sm || b.th !== a.th || b.pf !== a.pf || b.mv !== a.mv || b.pml !== a.pml) {
       p.kiosk_settings = { follow_up: b.fu, memory_integration: b.mem };
+      if (b.mv !== a.mv) p.kiosk_settings.max_volume = b.mv;
+      if (b.pml !== a.pml && b.pml !== null) p.kiosk_settings.pause_media_on_listen = b.pml;
       if (b.vad !== a.vad) p.kiosk_settings.vad_hangover_ms = b.vad;
       if (b.sm !== a.sm) p.kiosk_settings.stt_model = b.sm;
       if (b.th !== a.th) p.kiosk_settings.theme = b.th;
@@ -164,16 +166,25 @@ export const CLIENT_SCRIPT = `
     var pf = pick([['', S.profile_unset], ['pi3', S.profile_pi3], ['phone', S.profile_phone], ['tablet', S.profile_tablet], ['desktop', S.profile_desktop]], ks.profile || '');
     var vad = el('input'); vad.type = 'number'; vad.min = '300'; vad.max = '900'; vad.step = '50'; vad.value = String(ks.vad_hangover_ms || 450);
     function vadValue() { var n = parseInt(vad.value, 10); return isFinite(n) ? Math.min(900, Math.max(300, n)) : (initial ? initial.vad : (ks.vad_hangover_ms || 450)); }
-    function current() { return { bot: bot.value, stt: stt.value, tts: tts.value, fu: fu.checked, mem: mem.checked, vad: vadValue(), sm: sm.value, th: th.value, pf: pf.value }; }
+    var mv = pick([10, 20, 30, 40, 50, 60, 70, 80, 90, 100].map(function (n) { return [String(n), n + '%']; }), String(ks.max_volume || 100));
+    // F3: shown as it takes effect — the stored choice, else on for a phone or a tablet. Sent only once the operator
+    // ticks or unticks it (a default is never written); until then it follows the display type chosen here.
+    var handheld = function (v) { return v === 'phone' || v === 'tablet'; };
+    var pmlSet = typeof ks.pause_media_on_listen === 'boolean';
+    var pml = el('input'); pml.type = 'checkbox'; pml.checked = pmlSet ? ks.pause_media_on_listen : handheld(ks.profile);
+    pml.addEventListener('change', function () { pmlSet = true; });
+    pf.addEventListener('change', function () { if (!pmlSet) pml.checked = handheld(pf.value); });
+    function current() { return { bot: bot.value, stt: stt.value, tts: tts.value, fu: fu.checked, mem: mem.checked, vad: vadValue(), sm: sm.value, th: th.value, pf: pf.value, mv: parseInt(mv.value, 10) || 100, pml: pmlSet ? pml.checked : null }; }
     var initial = null;
     initial = current();
     var fit = fitLine(data.bots, bot, function () { return mem.checked; });
     mem.addEventListener('change', fit.show);
-    [[S.bot, bot], [S.stt, stt], [S.stt_model, sm], [S.tts, tts], [S.profile, pf], [S.vad_wait, vad], [S.follow_up, fu], [S.memory, mem], [S.theme, th]].forEach(function (pair) { var l = el('label', null, pair[0]); l.appendChild(pair[1]); card.appendChild(l); if (pair[1] === bot) card.appendChild(fit.node); });
+    [[S.bot, bot], [S.stt, stt], [S.stt_model, sm], [S.tts, tts], [S.profile, pf], [S.vad_wait, vad], [S.follow_up, fu], [S.memory, mem], [S.theme, th], [S.max_volume, mv], [S.pause_media_on_listen, pml]].forEach(function (pair) { var l = el('label', null, pair[0]); l.appendChild(pair[1]); card.appendChild(l); if (pair[1] === bot) card.appendChild(fit.node); });
     card.appendChild(el('p', 'kk-dim', S.vad_wait_hint));
     card.appendChild(el('p', 'kk-dim', S.memory_warn + ' ' + S.memory_hint));
     card.appendChild(el('p', 'kk-dim', S.profile_hint + (ks.profile_source === 'guessed' ? ' ' + S.profile_guessed : '')));
     card.appendChild(el('p', 'kk-dim', S.theme_hint));
+    card.appendChild(el('p', 'kk-dim', S.pause_media_hint));
     var msg = el('span', 'kk-msg');
     var save = el('button', 'btn btn-primary btn-sm', S.save); save.type = 'button';
     save.addEventListener('click', function () {
@@ -207,6 +218,65 @@ export const CLIENT_SCRIPT = `
     return card;
   }
 
+  /** Station presets: this instance's own list (name, spoken names, stream address). Saved as a whole; one bad row refuses the save. */
+  function renderStations() {
+    var box = document.getElementById('kk-stations'); if (!box) return;
+    api('GET', '/api/kiosk/admin/stations').then(function (j) {
+      clear(box);
+      box.appendChild(el('h2', null, S.stations_title));
+      box.appendChild(el('p', 'kk-dim', S.stations_intro));
+      var rows = el('div', 'kk-stations');
+      var msg = el('p', 'kk-msg');
+      function row(st) {
+        var r = el('div', 'kk-station');
+        var name = el('input'); name.maxLength = 60; name.placeholder = S.station_name; name.value = st.name || '';
+        var al = el('input'); al.maxLength = 220; al.placeholder = S.station_aliases; al.value = (st.aliases || []).join(', ');
+        var url = el('input'); url.maxLength = 500; url.placeholder = 'https://'; url.value = st.url || ''; url.type = 'url';
+        var loc = el('input'); loc.type = 'checkbox'; loc.checked = st.local === true; loc.title = S.station_local_hint;
+        var locL = el('label', 'kk-local', S.station_local); locL.insertBefore(loc, locL.firstChild);
+        var res = el('span', 'kk-msg');
+        var test = el('button', 'btn btn-secondary btn-sm', S.station_test); test.type = 'button';
+        test.addEventListener('click', function () {
+          res.textContent = S.station_testing;
+          api('POST', '/api/kiosk/admin/stations/test', { url: url.value, local: loc.checked }).then(function (t) {
+            res.textContent = t.ok ? fill(S.station_test_ok, { type: t.content_type || '' }) : (S['station_err_' + t.error] || S.station_err_unreachable);
+          });
+        });
+        var del = el('button', 'btn btn-secondary btn-sm', S.station_remove); del.type = 'button';
+        del.addEventListener('click', function () { rows.removeChild(r); });
+        [name, al, url, locL, test, del, res].forEach(function (n) { r.appendChild(n); });
+        r.read = function () { var v = { name: name.value, aliases: al.value.split(',').map(function (x) { return x.trim(); }).filter(Boolean).slice(0, 5), url: url.value.trim() }; if (loc.checked) v.local = true; return v; };
+        return r;
+      }
+      (j.stations || []).forEach(function (st) { rows.appendChild(row(st)); });
+      var add = el('button', 'btn btn-secondary btn-sm', S.station_add); add.type = 'button';
+      add.addEventListener('click', function () { if (rows.children.length < (j.max || 50)) rows.appendChild(row({})); });
+      var save = el('button', 'btn btn-primary btn-sm', S.save); save.type = 'button';
+      save.addEventListener('click', function () {
+        var list = [];
+        for (var i = 0; i < rows.children.length; i++) { var v = rows.children[i].read(); if (v.name || v.url) list.push(v); }
+        // A home-network station is checked against its policy (the same Test) before anything is saved.
+        var local = list.filter(function (x) { return x.local; });
+        var checks = local.map(function (x) { return api('POST', '/api/kiosk/admin/stations/test', { url: x.url, local: true }); });
+        Promise.all(checks).then(function (results) {
+          for (var j = 0; j < results.length; j++) {
+            if (!results[j].ok) { msg.textContent = fill(S.station_local_refused, { name: local[j].name }) + ' ' + (S['station_err_' + results[j].error] || S.station_err_unreachable); return; }
+          }
+          api('POST', '/api/kiosk/admin/stations', { stations: list }).then(function (k) {
+            msg.textContent = k.ok ? S.saved : k.error === 'local_check_failed' ? fill(S.station_local_refused, { name: k.station || '' }) + ' ' + (S['station_err_' + k.reason] || S.station_err_unreachable)
+              : k.error === 'command_name' ? fill(S.station_command_name, { word: k.word || '' }) : (S['station_' + k.error] || k.error || '');
+            if (k.ok) renderStations();
+          });
+        });
+      });
+      box.appendChild(rows);
+      // A station saved before command words were refused keeps playing; it is left out of the speech hint and named here.
+      if (j.command_names && j.command_names.length) box.appendChild(el('p', 'kk-warn', fill(S.station_command_saved, { words: j.command_names.join(', ') })));
+      var bar = el('div', 'kk-bar'); [add, save, msg].forEach(function (n) { bar.appendChild(n); });
+      box.appendChild(bar);
+    });
+  }
+
   function render(data) {
     state = data;
     renderPair(data);
@@ -222,11 +292,12 @@ export const CLIENT_SCRIPT = `
     api('GET', '/api/kiosk/admin/displays').then(function (j) {
       if (!j.devices || !state) return;
       var changed = JSON.stringify(j.pending) !== JSON.stringify(state.pending) || j.devices.length !== state.devices.length;
-      if (changed && !document.activeElement.closest('#kk-root form, #kk-root section, #kk-dash')) render(j);
+      if (changed && !document.activeElement.closest('#kk-root form, #kk-root section, #kk-dash, #kk-stations')) render(j);
       else state = j;
     });
   }
   load();
+  renderStations();
   window.__kkRefresh = setInterval(refreshPending, 5000);
 })();
 `;
@@ -247,6 +318,10 @@ const STYLES = `
   .kk-bar { display: flex; gap: 8px; align-items: center; margin-top: 8px; }
   .kk-form { display: grid; gap: 4px; max-width: 420px; }
   .kk-form input[name=code] { font-size: 1.4rem; letter-spacing: .15em; width: 9ch; }
+  .kk-station { display: flex; flex-wrap: wrap; gap: 6px; align-items: center; margin: 6px 0; }
+  .kk-station input { min-width: 10ch; flex: 1 1 12ch; }
+  .kk-station input[type=url] { flex: 2 1 24ch; }
+  .kk-station .kk-local { margin: 0; gap: 4px; }
   .kk-row { font-family: ui-monospace, monospace; font-size: .8rem; margin: 2px 0; }
 `;
 
@@ -273,6 +348,7 @@ export default {
         <div id="kk-pair" class="kk-card"></div>
         <div id="kk-dash" class="kk-card" hidden></div>
         <div id="kk-devices"></div>
+        <div id="kk-stations" class="kk-card"></div>
       </div>
       <script type="application/json" id="kk-strings">${json}</script>
       <script>${CLIENT_SCRIPT}<\/script>`;

bundles/kiosk/panel/routes.js

@@ -71,6 +71,7 @@ const runtime = createKioskRuntime({
 // Smoke 2026-10-04: the first inference after a whisper start took 8.2 s even with the
 // model preloaded, so warm every paired display's STT now (retried until whisper answers).
 runtime.bootWarmup().catch(() => {});
+runtime.migrateVolumeCaps().catch(() => {});   // rev 7b (M2): old linear caps → the 5 dB scale, once
 
 export default function kioskRouter(dashboardAuth) { return runtime.router(dashboardAuth); }
 export function setupWebSocket(server) { return runtime.attachUpgrade(server); }

bundles/kiosk/public/audio.js

@@ -1,9 +1,26 @@
 /** Mic capture (AEC/NS/AGC on, spec §7.3) and in-page playback (so Chromium's echo canceller sees it). */
 import { playStartPerfTime } from "./metrics.js";
 
+/** The capture worklet is added to an AudioContext once: a phone opens the mic again at every tap (F6). */
+const worklets = new WeakSet();
+/**
+ * Re-smoke 2026-10-07 (G12): on a phone the reply's sources sometimes never ended (the output clock stalled
+ * while Android switched audio mode), so the page held the music and the bird until Stop. The player now
+ * bounds every reply: past the scheduled end + DRAIN_GRACE_MS it stops what is left and drains (onStall says so);
+ * an mp3 decode that does not answer within DECODE_MAX_MS is dropped.
+ */
+export const DRAIN_GRACE_MS = 2000;
+/** Review of rev 7 (M1): a reply is cut only once its audio clock has stopped moving — a late but running sink (a Bluetooth
+ *  speaker waking) re-arms the check — and never held past twice its length plus STALL_CAP_EXTRA_MS. */
+export const STALL_CAP_EXTRA_MS = 5000;
+export const DECODE_MAX_MS = 5000;
 export async function openMic(ctx, onFrame) {
   const stream = await navigator.mediaDevices.getUserMedia({ audio: { echoCancellation: true, noiseSuppression: true, autoGainControl: true, channelCount: 1 } });
-  await ctx.audioWorklet.addModule("/display/assets/pcm-worklet.js");
+  if (!worklets.has(ctx)) {
+    try { await ctx.audioWorklet.addModule("/display/assets/pcm-worklet.js"); }
+    catch (err) { stream.getTracks().forEach((t) => t.stop()); throw err; }
+    worklets.add(ctx);
+  }
   const src = ctx.createMediaStreamSource(stream);
   const node = new AudioWorkletNode(ctx, "pcm-capture", { numberOfInputs: 1, numberOfOutputs: 0 });
   node.port.onmessage = (e) => onFrame(e.data.pcm, e.data.rms, performance.now());
@@ -15,6 +32,31 @@ export async function openMic(ctx, onFrame) {
   };
 }
 
+/**
+ * One microphone at a time (review H3): every caller of acquire() during an open waits for the SAME open, so a
+ * second tap can never start a second stream. release() during an open closes what that open returns. open() →
+ * Promise<mic with close()>; now() → ms (the open time is kept for the turn's metrics).
+ */
+export function createMicGate(open, now = () => performance.now()) {
+  let mic = null, pending = null, dropped = false, lastMs = null;
+  return {
+    get current() { return mic; },
+    opening: () => pending !== null,
+    acquire() {
+      if (mic) return Promise.resolve(mic);
+      if (pending) { dropped = false; return pending; }
+      dropped = false;
+      const t0 = now();
+      pending = Promise.resolve().then(open).then(
+        (m) => { pending = null; if (dropped) { try { m.close(); } catch {} return null; } mic = m; lastMs = now() - t0; return m; },
+        (err) => { pending = null; throw err; });
+      return pending;
+    },
+    release() { if (pending) dropped = true; if (mic) { const m = mic; mic = null; try { m.close(); } catch {} } },
+    takeOpenMs() { const v = lastMs; lastMs = null; return v; },
+  };
+}
+
 /**
  * Sequential playback of the server's sentence buffers. `playing` covers buffers still
  * being decoded (the server's idle can arrive before an mp3 decode finishes, ruling F3).
@@ -23,13 +65,13 @@ export async function openMic(ctx, onFrame) {
  * frames still in flight are dropped until the next begin() (tts_start). onFirstPlay(at, tag) carries the
  * tag given to begin() so the page attributes the play time to the right turn.
  */
-export function createPlayer(ctx, { onLevel, onFirstPlay, onDrained }) {
+export function createPlayer(ctx, { onLevel, onFirstPlay, onDrained, onStall = () => {} }) {
   const analyser = ctx.createAnalyser();
   analyser.fftSize = 256;
   analyser.connect(ctx.destination);
   const data = new Uint8Array(analyser.fftSize);
   const sources = new Set();
-  let codec = "pcm", rate = 24000, nextAt = 0, first = true, tag = null, level = null, chain = Promise.resolve(), gen = 0, pending = 0, muted = false;
+  let codec = "pcm", rate = 24000, nextAt = 0, first = true, tag = null, level = null, chain = Promise.resolve(), gen = 0, pending = 0, muted = false, watch = null, lastClock = 0, capAt = 0, audioMs = 0;
   const startLevel = () => {
     if (level) return;
     level = setInterval(() => {                     // ~15 Hz beak level, only while audio plays
@@ -40,7 +82,30 @@ export function createPlayer(ctx, { onLevel, onFirstPlay, onDrained }) {
     }, 66);
   };
   const stopLevel = () => { clearInterval(level); level = null; onLevel(0); };
-  const maybeDrained = () => { if (!sources.size && !pending) { stopLevel(); onDrained(); } };
+  const maybeDrained = () => { if (!sources.size && !pending) { clearTimeout(watch); watch = null; capAt = 0; stopLevel(); onDrained(); } };
+  const stopAll = () => { for (const s of sources) { s.onended = null; try { s.stop(); } catch {} } sources.clear(); };
+  const left = () => Math.max(0, (nextAt - ctx.currentTime) * 1000);
+  // Looked at past the scheduled end + DRAIN_GRACE_MS: a clock that moved since the last look gets another look (M1).
+  const check = () => {
+    watch = null;
+    if (!sources.size) return;
+    const moved = ctx.currentTime > lastClock;
+    lastClock = ctx.currentTime;
+    if (moved && Date.now() < capAt) { watch = setTimeout(check, left() + DRAIN_GRACE_MS); return; }
+    onStall({ left: sources.size, ctxState: String(ctx.state || ""), clockMoved: moved });
+    stopAll(); nextAt = 0;
+    maybeDrained();
+  };
+  const arm = () => {
+    clearTimeout(watch);
+    lastClock = ctx.currentTime;
+    capAt = Math.max(capAt, Date.now() + 2 * left() + STALL_CAP_EXTRA_MS);
+    watch = setTimeout(check, left() + DRAIN_GRACE_MS);
+  };
+  const decode = (buf) => new Promise((res, rej) => {
+    const cap = setTimeout(() => rej(new Error("decode timeout")), DECODE_MAX_MS);
+    ctx.decodeAudioData(buf.slice(0)).then((ab) => { clearTimeout(cap); res(ab); }, (err) => { clearTimeout(cap); rej(err); });
+  });
   async function play(buf, g) {
     let ab;
     if (codec === "pcm") {
@@ -49,7 +114,7 @@ export function createPlayer(ctx, { onLevel, onFirstPlay, onDrained }) {
       const ch = ab.getChannelData(0);
       for (let i = 0; i < i16.length; i++) ch[i] = i16[i] / 32768;
     } else {
-      ab = await ctx.decodeAudioData(buf.slice(0));
+      ab = await decode(buf);
     }
     if (g !== gen) return;
     const src = ctx.createBufferSource();
@@ -58,8 +123,10 @@ export function createPlayer(ctx, { onLevel, onFirstPlay, onDrained }) {
     const when = Math.max(ctx.currentTime + 0.02, nextAt);
     src.start(when);
     nextAt = when + ab.duration;
+    audioMs += ab.duration * 1000;
     sources.add(src);
     src.onended = () => { sources.delete(src); maybeDrained(); };
+    arm();
     startLevel();                                   // every start: a drained inter-sentence gap stopped it
     if (first) {
       first = false;
@@ -68,7 +135,7 @@ export function createPlayer(ctx, { onLevel, onFirstPlay, onDrained }) {
   }
   return {
     // nextAt is NOT reset: an announcement that follows a turn queues after it instead of overlapping.
-    begin(c, sr, t = null) { codec = c === "mp3" ? "mp3" : "pcm"; rate = sr || 24000; first = true; tag = t; muted = false; },
+    begin(c, sr, t = null) { codec = c === "mp3" ? "mp3" : "pcm"; rate = sr || 24000; first = true; tag = t; muted = false; audioMs = 0; },
     push(buf) {
       if (muted) return;                            // in-flight audio from before a barge (until the next tts_start)
       const g = gen;
@@ -83,10 +150,12 @@ export function createPlayer(ctx, { onLevel, onFirstPlay, onDrained }) {
       gen++;
       muted = true;
       pending = 0;
-      for (const s of sources) { s.onended = null; try { s.stop(); } catch {} }
-      sources.clear(); nextAt = 0; chain = Promise.resolve(); stopLevel();
+      clearTimeout(watch); watch = null; capAt = 0;
+      stopAll(); nextAt = 0; chain = Promise.resolve(); stopLevel();
     },
     get playing() { return sources.size > 0 || pending > 0; },
     get sampling() { return level != null; },
+    /** r7b H3: audio scheduled since the last begin() (ms; counts only). */
+    get audioMs() { return Math.round(audioMs); },
   };
 }