Add generated UI notification sounds

.gitignore

@@ -7,3 +7,7 @@
 /halls/**/*.json
 /static/**/*.d.ts
 /static/third-party/tasks-vision
+/distro-packages/Arch-Linux/skald-git/pkg/
+/distro-packages/Arch-Linux/skald-git/skald/
+/distro-packages/Arch-Linux/skald-git/src/
+/distro-packages/Arch-Linux/skald-git/*.pkg.tar.*

static/skald.html

@@ -164,12 +164,12 @@
             </select>
   
             <form>
-              <input id="preprocessingbox" type="checkbox" checked/>
+              <input id="preprocessingbox" type="checkbox"/>
               <label for="preprocessingbox">Noise suppression</label>
             </form>
 
             <form>
-              <input id="hqaudiobox" type="checkbox"/>
+              <input id="hqaudiobox" type="checkbox" checked/>
               <label for="hqaudiobox">High-quality audio</label>
             </form>
 
@@ -179,6 +179,11 @@
         <fieldset>
           <legend>Other Settings</legend>
 
+          <form>
+            <input id="notificationsoundsbox" type="checkbox" checked/>
+            <label for="notificationsoundsbox">Notification sounds</label>
+          </form>
+
           <form id="sendform">
             <label for="sendselect" class="sidenav-label-first">Send:</label>
             <select id="sendselect" class="select select-inline">

static/skald.js

@@ -76,6 +76,7 @@ let probingState = null;
 
  * @property {boolean} [preprocessing]
  * @property {boolean} [hqaudio]
+ * @property {boolean} [notificationSounds]
  * @property {boolean} [forceRelay]
  */
 
@@ -248,10 +249,150 @@ function reflectSettings() {
         store = true;
     }
 
+    if(settings.hasOwnProperty('notificationSounds')) {
+        getInputElement('notificationsoundsbox').checked =
+            settings.notificationSounds;
+    } else {
+        settings.notificationSounds =
+            getInputElement('notificationsoundsbox').checked;
+        store = true;
+    }
+
     if(store)
         storeSettings(settings);
 }
 
+/** @type {AudioContext} */
+let notificationAudioContext = null;
+let userNotificationSoundsReady = false;
+
+function notificationSoundsEnabled() {
+    return !!getSettings().notificationSounds;
+}
+
+function getNotificationAudioContext() {
+    if(!notificationSoundsEnabled())
+        return null;
+
+    let AudioContextConstructor =
+        window.AudioContext || window.webkitAudioContext;
+    if(!AudioContextConstructor)
+        return null;
+
+    if(!notificationAudioContext)
+        notificationAudioContext = new AudioContextConstructor();
+
+    if(notificationAudioContext.state === 'suspended')
+        notificationAudioContext.resume().catch(e => {
+            console.warn('Could not resume notification audio:', e);
+        });
+
+    return notificationAudioContext;
+}
+
+/**
+ * @param {number} start
+ * @param {number} duration
+ * @param {number} fromFrequency
+ * @param {number} toFrequency
+ * @param {number} gainValue
+ * @param {OscillatorType} [type]
+ */
+function scheduleGlide(start, duration, fromFrequency, toFrequency, gainValue, type) {
+    let context = getNotificationAudioContext();
+    if(!context)
+        return;
+
+    let oscillator = context.createOscillator();
+    let gain = context.createGain();
+    oscillator.type = type || 'sine';
+    oscillator.frequency.setValueAtTime(fromFrequency, start);
+    oscillator.frequency.exponentialRampToValueAtTime(toFrequency, start + duration);
+    gain.gain.setValueAtTime(0, start);
+    gain.gain.linearRampToValueAtTime(gainValue, start + 0.02);
+    gain.gain.exponentialRampToValueAtTime(0.0001, start + duration);
+    oscillator.connect(gain);
+    gain.connect(context.destination);
+    oscillator.start(start);
+    oscillator.stop(start + duration + 0.02);
+}
+
+/**
+ * @param {number} offset
+ * @param {number} duration
+ * @param {number} frequency
+ * @param {number} gainValue
+ * @param {OscillatorType} [type]
+ */
+function scheduleTone(offset, duration, frequency, gainValue, type) {
+    let context = getNotificationAudioContext();
+    if(!context)
+        return;
+
+    let start = context.currentTime + offset;
+    let oscillator = context.createOscillator();
+    let gain = context.createGain();
+    oscillator.type = type || 'sine';
+    oscillator.frequency.setValueAtTime(frequency, start);
+    gain.gain.setValueAtTime(0, start);
+    gain.gain.linearRampToValueAtTime(gainValue, start + 0.015);
+    gain.gain.exponentialRampToValueAtTime(0.0001, start + duration);
+    oscillator.connect(gain);
+    gain.connect(context.destination);
+    oscillator.start(start);
+    oscillator.stop(start + duration + 0.02);
+}
+
+/**
+ * @param {string} name
+ */
+function playNotificationSound(name) {
+    let context = getNotificationAudioContext();
+    if(!context)
+        return;
+
+    let now = context.currentTime;
+    switch(name) {
+    case 'hand-raised':
+        scheduleGlide(now, 0.36, 420, 1320, 0.035, 'sine');
+        break;
+    case 'hand-lowered':
+        scheduleGlide(now, 0.34, 1180, 360, 0.03, 'sine');
+        break;
+    case 'user-joined':
+        scheduleTone(0, 0.18, 660, 0.035, 'sine');
+        scheduleTone(0.18, 0.26, 880, 0.032, 'sine');
+        break;
+    case 'user-left':
+        scheduleTone(0, 0.16, 660, 0.025, 'triangle');
+        scheduleTone(0.14, 0.22, 440, 0.022, 'triangle');
+        break;
+    case 'recording-started':
+        scheduleTone(0, 0.16, 523.25, 0.03, 'sine');
+        scheduleTone(0.09, 0.18, 659.25, 0.026, 'sine');
+        scheduleTone(0.18, 0.24, 783.99, 0.024, 'sine');
+        break;
+    case 'recording-stopped':
+        scheduleTone(0, 0.16, 783.99, 0.026, 'sine');
+        scheduleTone(0.12, 0.18, 659.25, 0.024, 'sine');
+        scheduleTone(0.24, 0.24, 392, 0.022, 'sine');
+        break;
+    case 'chat-message':
+        scheduleTone(0, 0.09, 880, 0.018, 'triangle');
+        scheduleTone(0.08, 0.11, 1174.66, 0.014, 'triangle');
+        break;
+    default:
+        break;
+    }
+}
+
+function unlockNotificationAudio() {
+    getNotificationAudioContext();
+}
+
+document.addEventListener('pointerdown', unlockNotificationAudio);
+document.addEventListener('keydown', unlockNotificationAudio);
+
 /**
  * Returns true if we should use the mobile layout.  This should be kept
  * in sync with the CSS.
@@ -589,6 +730,15 @@ getInputElement('hqaudiobox').onchange = function(e) {
     updateSettings({hqaudio: this.checked});
 };
 
+getInputElement('notificationsoundsbox').onchange = function(e) {
+    e.preventDefault();
+    if(!(this instanceof HTMLInputElement))
+        throw new Error('Unexpected type for this');
+    updateSettings({notificationSounds: this.checked});
+    if(this.checked)
+        getNotificationAudioContext();
+};
+
 document.getElementById('mutebutton').onclick = function(e) {
     e.preventDefault();
     let localMute = getSettings().localMute;
@@ -1053,7 +1203,7 @@ async function addLocalMedia(localId) {
     /** @type{boolean|MediaTrackConstraints} */
     let audio = false;
     if(settings.audio === 'default') {
-        audio = true;
+        audio = {};
     } else if(settings.audio && settings.audio !== 'off') {
         audio = {deviceId: settings.audio};
     }
@@ -1609,7 +1759,7 @@ function addUser(id, userinfo) {
     user.classList.add("user-p");
     user.setAttribute('role', 'button');
     user.setAttribute('tabindex', '0');
-    setUserStatus(id, user, userinfo);
+    setUserStatus(id, user, userinfo, false);
     user.addEventListener('click', function(e) {
         let elt = e.currentTarget;
         if(!elt || !(elt instanceof HTMLElement))
@@ -1664,32 +1814,36 @@ function changeUser(id, userinfo) {
         console.warn('Unknown user ' + id);
         return;
     }
-    setUserStatus(id, elt, userinfo);
+    setUserStatus(id, elt, userinfo, true);
 }
 
 /**
  * @param {string} id
  * @param {HTMLElement} elt
  * @param {user} userinfo
+ * @param {boolean} notify
  */
-function setUserStatus(id, elt, userinfo) {
+function setUserStatus(id, elt, userinfo, notify) {
     let username = userinfo.username ? userinfo.username : '(anon)';
     elt.textContent = username;
 
     let wasRaised = elt.classList.contains('user-status-raisehand');
+    let isRemote = id !== (serverConnection && serverConnection.id);
     if(userinfo.data.raisehand) {
         elt.classList.add('user-status-raisehand');
         elt.setAttribute('aria-label', `${username} (hand raised)`);
-        // Announce when hand is newly raised (not on initial load)
-        if(!wasRaised && id !== (serverConnection && serverConnection.id)) {
+        if(notify && !wasRaised) {
             let announcement = document.getElementById('chat-announcements');
-            if(announcement) {
+            if(announcement && isRemote) {
                 announcement.textContent = `${username} raised their hand`;
             }
+            playNotificationSound('hand-raised');
         }
     } else {
         elt.classList.remove('user-status-raisehand');
         elt.setAttribute('aria-label', username);
+        if(notify && wasRaised)
+            playNotificationSound('hand-lowered');
     }
 
     let microphone=false;
@@ -1727,10 +1881,14 @@ function gotUser(id, kind) {
     switch(kind) {
     case 'add':
         addUser(id, serverConnection.users[id]);
+        if(userNotificationSoundsReady && id !== serverConnection.id)
+            playNotificationSound('user-joined');
         if(Object.keys(serverConnection.users).length == 3)
             reconsiderSendParameters();
         break;
     case 'delete':
+        if(userNotificationSoundsReady && id !== serverConnection.id)
+            playNotificationSound('user-left');
         delUser(id);
         if(Object.keys(serverConnection.users).length < 3)
             scheduleReconsiderParameters();
@@ -1869,6 +2027,9 @@ async function gotJoined(kind, hall, perms, status, data, error, message) {
                           serverConnection.username
         );
         openSafariStream();
+        window.setTimeout(() => {
+            userNotificationSoundsReady = true;
+        }, 1000);
         if(kind === 'change')
             return;
         break;
@@ -2121,10 +2282,14 @@ function gotUserMessage(id, dest, username, time, privileged, kind, error, messa
             return;
         }
         let text = '' + message;
-        if(text === 'Recording started')
+        if(text === 'Recording started') {
             announceUrgent(text);
-        else
+            playNotificationSound('recording-started');
+        } else {
             announceChat(text);
+            if(text === 'Recording stopped')
+                playNotificationSound('recording-stopped');
+        }
         displayMessage(text);
         break;
     }
@@ -2462,6 +2627,8 @@ function addToChatbox(id, peerId, dest, nick, time, privileged, history, kind, m
         if(serverConnection && peerId === serverConnection.id)
             speaker = 'You';
         announceChat(speaker ? `${speaker}: ${messageText}` : messageText);
+        if(serverConnection && peerId && peerId !== serverConnection.id)
+            playNotificationSound('chat-message');
     }
 
     return;

todo.txt

@@ -1,238 +1,154 @@
-# Skald Hard Fork - Master TODO
+# Skald Recording Quality Roadmap
+#
+# Current state:
+#   The original hard-fork checklist is retired.  Skald is live on the user's
+#   server, Galene rooms have been migrated as Skald halls, and basic recording
+#   has been verified with real users.
 #
 # Goal:
-#   Turn this Galene-derived repository into Skald: a new, audio-only,
-#   hall-based conferencing server. This is a hard fork, so compatibility with
-#   Galene names, URLs, video behavior, and recording format is intentionally
-#   not the default.
+#   Make Skald's single mixed Ogg Opus recordings sound good, stay reliable
+#   under larger audio-only halls, and avoid avoidable CPU/process overhead.
 #
 # Legend:
-#   - item  = needs doing
-#   + item  = implemented, needs testing
-#   X item  = tested and done
+#   - item  = needs doing.
+#   + item  = implemented, needs testing.
+#   X item  = tested and done.
 #
-# Non-negotiable target state:
-#   - Project name is Skald everywhere.
-#   - Halls replace groups/rooms everywhere user-facing and in internal code
-#     where practical.
-#   - Video is removed, not hidden.
-#   - Incoming video is rejected.
-#   - The browser UI is audio-only and accessible.
-#   - Recordings are single mixed Ogg Opus audio files.
-#   - No legacy Galene compatibility paths or fallback shims unless explicitly
-#     added later by choice.
+# Constraints:
+#   - Keep recordings as one mixed Ogg Opus file per hall recording session.
+#   - Preserve clear recording start/stop announcements and operator feedback.
+#   - Do not block WebRTC packet handling on recording work.
+#   - Favor small, testable stages over one large rewrite.
+#   - Keep live-server safety in mind; staged improvements should be deployable
+#     independently.
 
 # ============================================================================
-# PHASE 0: BASELINE INVENTORY AND GUARDRAILS
+# PHASE 0: BASELINE AND MEASUREMENT
 # ============================================================================
 
-X Record the current upstream/fork point in CHANGES or a new fork note
-X Capture current baseline: `go test ./...` result before behavior changes
-X Capture current baseline: `go build ./...` result before behavior changes
-X Inventory generated/static assets, including static/example and third-party assets
-X Inventory all public URLs: /group/, /public-groups.json, /recordings/, /galene-api/
-X Inventory all config/data paths: groups/, data/, recordings/, skaldctl.json
-X Inventory all Galene binary/service names in docs, scripts, examples, and tests
-X Decide whether existing stored data will be migrated manually or treated as incompatible
-X Keep the old LICENSE intact and preserve required attribution
+X Confirm the current live deployment can start and stop a Skald recording.
+X Confirm migrated Galene rooms work as Skald halls on meet.stormux.org.
+- Document the current recording pipeline in skald.md for maintainers:
+  incoming Opus RTP -> per-source ffmpeg decoder -> PCM mixer -> ffmpeg Opus
+  encoder -> Ogg Opus file.
+- Add lightweight recording metrics in logs or debug output:
+  active recorded sources, dropped source frames, mixer write errors, ffmpeg
+  decoder starts/stops, and encoder failures.
+- Add a synthetic recording benchmark or smoke tool that can simulate multiple
+  active Opus sources without needing browser clients.
+- Record baseline CPU, process count, and output quality for 1, 5, 10, and 25
+  active sources on a representative server.
+- Define practical target limits for "normal hall", "large moderated hall",
+  and "stress test" recording scenarios.
 
 # ============================================================================
-# PHASE 1: FOUNDATION AND TOTAL REBRANDING
+# PHASE 1: MIX QUALITY BEFORE PIPELINE REWRITE
 # ============================================================================
 
-X Rename Go module: github.com/jech/galene -> git.stormux.org/storm/skald
-X Update all internal Go import paths across source and tests
-X Rename galene.go -> skald.go and keep the main package entrypoint working
-X Rename galenectl/ directory and binary to skaldctl/
-X Rename static/galene.css -> static/skald.css
-X Rename static/galene.js -> static/skald.js
-X Rename static/galene.html -> static/skald.html
-X Update webserver static serving to load skald.html and renamed assets
-X Update all HTML, JS, CSS, docs, tests, and examples that reference renamed assets
-X Rename galene-*.md docs to skald-*.md
-X Replace "Galene", "Galène", and "galene" with "Skald" and "skald" where they refer to this project
-X Keep historical upstream references only where they are attribution, changelog history, or license context
-X Rename default binary, service, and install examples from galene to skald
-X Rename galenectl config file to skaldctl.json
-X Update CHANGES with Skald fork point and first Skald development section
-X Verify `go test ./...` compiles after the pure project rename pass
-X Search for remaining Galene names and classify each survivor as attribution/history or a bug
+- Add mixer tests for multiple simultaneous PCM sources, including clipping
+  cases and quiet/loud source combinations.
+- Replace raw summing with a safer mix strategy:
+  per-source attenuation, headroom, or another deterministic approach that
+  reduces clipping when multiple users speak.
+- Add a simple limiter to prevent harsh clipping in the final mixed PCM.
+- Add tests proving the limiter preserves normal speech-level samples and
+  bounds overloaded samples.
+- Add optional per-source level tracking so logs or future UI can identify
+  sources that dominate the recording.
+- Verify recordings with 2-5 simultaneous speakers sound cleaner than the
+  current clamped sum.
+- Verify the limiter does not create pumping, distortion, or excessive volume
+  loss for a single speaker.
 
 # ============================================================================
-# PHASE 2: HALL TERMINOLOGY AND DATA MODEL
+# PHASE 2: SOURCE LIFECYCLE AND SILENCE HANDLING
 # ============================================================================
 
-X Rename Go package group/ to hall/ if the package-wide rename stays tractable
-X Rename package declarations, types, variables, funcs, tests, and comments from group to hall
-X Rename Group type to Hall and update all compile errors intentionally
-X Rename group description files/concepts to hall description files/concepts
-X Rename default directory flag: -groups -> -halls
-X Rename default data directory: groups/ -> halls/
-X Rename public listing concept from public groups to public halls
-X Rename subgroups to subhalls everywhere, including JSON fields and commands
-X Rename auto-subgroups -> auto-subhalls
-X Rename allow-subgroups -> allow-subhalls
-X Rename include-subgroups -> include-subhalls
-X Rename group action names to hall action names where exposed to clients
-X Rename chat/help commands such as /subgroups to /subhalls
-X Update tests to use generic placeholder names, not real usernames
-X Verify no runtime user-facing "group", "groups", "room", or "rooms" remain except attribution/history
-X Verify old group JSON/config paths are not silently accepted unless explicitly reintroduced later
+- Track whether each recording source is active, silent, stalled, or closed.
+- Treat missing frames as silence without blocking the mixer.
+- Drop or de-prioritize stale silent sources after a short timeout.
+- Ensure mute/unmute transitions do not leave stale audio in the mixed output.
+- Ensure participant join/leave during recording never stops the recording file.
+- Add tests for source join, leave, mute, unmute, stall, and recovery during an
+  active recording.
+- Verify a long recording remains valid when sources repeatedly join and leave.
 
 # ============================================================================
-# PHASE 3: URLS, API, PROTOCOL, AND CONTROL TOOL
+# PHASE 3: REDUCE PROCESS OVERHEAD
 # ============================================================================
 
-X Rename browser hall URLs from /group/name/ to /hall/name/
-X Rename /public-groups.json to /public-halls.json
-X Rename /galene-api/ to /skald-api/
-X Rename API collection paths from .groups to .halls
-X Rename protocol join fields from group to hall
-X Rename protocol callbacks/events from joined group semantics to joined hall semantics
-X Update static/protocol.js public API: ServerConnection.join and related callbacks
-X Update static/management.js to use /skald-api/v0/.halls/
-X Update static/mainpage.js to open /hall/ URLs and fetch /public-halls.json
-X Update stats endpoint naming only if it exposes Skald/group terminology
-X Update skaldctl command names: create-hall, update-hall, delete-hall, list-halls
-X Update skaldctl flags: -hall, -include-subhalls, -auto-subhalls, etc.
-X Update skaldctl help text and error messages
-X Update API and webserver tests for /hall/ and /skald-api/
-X Verify old /group/, /public-groups.json, /galene-api/, and .groups paths fail clearly
-X Verify runtime protocol backward compatibility with Skald is intentionally broken
+- Evaluate Go-native Opus decode options against the current ffmpeg decoder
+  subprocesses:
+  sound quality, maintenance risk, packaging impact, CGO requirements, and CPU.
+- Decide whether in-process Opus decoding is worth the dependency and packaging
+  tradeoff.
+- If keeping ffmpeg decoders, pool or supervise decoder processes more
+  deliberately and improve failure recovery.
+- If replacing ffmpeg decoders, implement one in-process decoder path behind
+  tests before removing the subprocess path.
+- Verify recording still handles dynamic payload type, RTP sequence behavior,
+  packet loss, and source restarts.
+- Measure CPU and process-count improvement with 10+ active sources.
+- Keep one final encoder path to Ogg Opus unless there is a proven reason to
+  replace it too.
 
 # ============================================================================
-# PHASE 4: BACKEND - STRIP VIDEO COMPLETELY
+# PHASE 4: ENCODER AND CONTAINER HARDENING
 # ============================================================================
 
-+ Remove video codec constants and helpers from codecs/ package
-+ Remove or rewrite video-only codec tests
-+ Strip video/vp8, video/vp9, video/av1, and video/h264 from SDP negotiation
-+ Ensure audio/opus remains negotiated correctly
-+ Reject incoming video m-lines/transceivers at WebRTC peer connection setup
-+ Reject or close incoming video tracks if a browser still sends them
-+ Ensure rejection is explicit in logs without crashing the hall
-+ Remove outgoing video track routing from rtpconn/ and hall/
-+ Remove simulcast and SVC handling
-+ Remove video bandwidth estimation and video throughput selection
-+ Remove video-specific RTCP behavior such as PLI/FIR/keyframe requests
-+ Remove screenshare backend support
-+ Remove presentation/file-as-video support if it depends on video tracks
-+ Remove background blur support and background-blur-worker.js from the served app
-X Remove video fields from hall descriptions, stats, API responses, and protocol docs
-+ Check WHIP/WHEP-style endpoints for video assumptions and either make audio-only or remove
-+ Ensure browsers that request audio+video get a working audio-only session or a clear video rejection
-X Verify two browser clients can join one hall and exchange audio only
-X Verify a client attempting camera or screenshare cannot publish video
+- Decide whether the final Ogg Opus encoder should remain ffmpeg or move
+  in-process later.
+- Make encoder startup errors operator-visible and log the exact ffmpeg error.
+- Ensure encoder stdin writes cannot block the mixer indefinitely.
+- Ensure recorder close handles encoder failure, stdin close failure, and wait
+  failure without leaving misleading successful output.
+- Verify interrupted or failed recordings are either clearly marked as failed or
+  produce a playable partial file.
+- Verify completed files with ffprobe or equivalent in tests where available.
+- Verify completed files play in common players after long recordings.
 
 # ============================================================================
-# PHASE 5: FRONTEND - AUDIO-ONLY USER EXPERIENCE
+# PHASE 5: LARGE-HALL OPERATIONS
 # ============================================================================
 
-+ Remove all <video> elements and video containers from skald.html
-+ Remove camera buttons and presentation buttons tied to camera/video behavior
-+ Remove screenshare button and related menu commands
-+ Remove video quality, resolution, throughput, simulcast, and collapse controls
-+ Remove video device selector and camera permission flow
-+ Remove hideVideo(), showVideo(), getMaxVideoThroughput(), and related JS paths
-+ Remove background blur UI, worker references, and CSS
-+ Keep microphone mute/unmute behavior intact
-+ Keep audio input and output device selection intact
-+ Keep chat, user list, moderation, locking, recording controls, and status messages
-+ Rename visible text in skald.html, index.html, stats.html, mainpage.js, management.js, and protocol errors
-+ Replace public groups UI with public halls UI
-+ Remove video CSS from skald.css and common.css
-- Check layout after removing the video pane; no dead space or hidden required controls
-- Preserve keyboard access with Tab and Shift+Tab through all controls
-- Ensure controls have labels or accessible names after icon/button cleanup
-X Verify audio-only UI loads without 404s for renamed assets
-X Verify joining, muting, chat, recording controls, and leaving work from keyboard
+- Add documentation for recommended large-hall permissions:
+  most users observe/message, only presenters publish audio, operators control
+  recording.
+- Document expected scaling limits in terms of active speakers, not just total
+  connected users.
+- Add guidance for max-clients, TURN bandwidth, CPU monitoring, and recording
+  expectations for 50, 100, and 250 user events.
+- Add an operator-facing warning or log message when active recorded sources
+  exceed a conservative threshold.
+- Test a large moderated hall with many listeners and a few active speakers.
+- Test a stress hall with many active speakers and document the observed limit.
 
 # ============================================================================
-# PHASE 6: RECORDING REWRITE - SINGLE MIXED OGG OPUS
+# PHASE 6: FINAL CLEANUP AND RELEASE READINESS
 # ============================================================================
 
-+ Remove diskwriter dependency on github.com/at-wat/ebml-go and WebM/Matroska writing
-+ Remove per-track/per-participant WebM recording logic
-+ Remove video recording fields such as hasVideo, width, height, keyframes, and video MIME handling
-X Decide final mixer implementation before coding:
-  * Decode all active Opus RTP audio to 48 kHz PCM
-  * Maintain a per-source jitter/timing buffer so join/leave does not corrupt the mix
-  * Mix to stereo with clipping prevention or limiting
-  * Insert silence for missing frames rather than blocking the mixer
-  * Encode mixed PCM to Ogg Opus
-+ Evaluate Go-native decode/encode/container libraries versus ffmpeg subprocess
-+ If using ffmpeg, add a startup check and clear admin/operator error when ffmpeg is missing
-+ If using ffmpeg, use nonblocking stdin writes and drain stderr safely
-X If using ffmpeg, command shape starts from:
-  ffmpeg -f s16le -ar 48000 -ac 2 -i pipe:0 -c:a libopus -b:a 128k output.ogg
-+ Ensure recording work cannot block WebRTC packet handling
-+ Name recording files skald-recording-YYYYMMDD-HHMMSS.ogg
-+ Avoid usernames or live personal identifiers in recording filenames
-+ Store one recording file per hall recording session
-+ Close recordings cleanly on explicit stop
-+ Close recordings cleanly when the hall empties
-+ Handle participant join/leave during recording without stopping the file
-+ Handle mute/unmute during recording without stale audio
-X Handle recording start when no one is speaking
-X Add focused mixer/recording tests with synthetic audio where possible
-X Verify produced Ogg Opus file with `ffprobe` or equivalent
-- Verify produced file plays in a standard audio player
+- Update skald.md and skald-install.md with the final recording architecture.
+- Update package dependencies if recording dependency choices change.
+- Run focused recording tests and webserver recording-control tests.
+- Run `go test ./...` in a clean tree without distro package build artifacts.
+- Run live browser testing for recording start, stop, join, leave, mute, and
+  playback.
+- Do a final review for accidental regressions in recording announcements,
+  hall permissions, and migrated-hall compatibility.
 
 # ============================================================================
-# PHASE 7: CONFIGURATION, INSTALLATION, AND DEPLOYMENT SURFACES
+# PHASE 7: WEB UI AUDIO DEFAULTS AND EVENT SOUNDS
 # ============================================================================
 
-X Update command-line flags and help output for Skald and halls
-X Update default examples for halls/, data/, recordings/, and skaldctl.json
-X Update systemd service examples to skald user, binary, and paths
-X Update reverse proxy examples to /hall/ and /skald-api/
-X Update TURN/STUN docs only where project name or URLs changed
-X Update Docker/container or packaging files if present later
-X Update .gitignore for renamed binaries, config files, and generated artifacts
-X Verify there are no stale install commands that build or run skald/skaldctl
-
-# ============================================================================
-# PHASE 8: DOCUMENTATION
-# ============================================================================
-
-X Rewrite README.md to describe Skald as audio-only hall conferencing
-X Rewrite skald-install.md for new project name, binary names, and hall paths
-X Rewrite skald.md admin/usage docs for audio-only behavior
-X Rewrite skald-client.md for audio-only client authors
-X Rewrite skald-protocol.md for hall terminology and audio-only tracks
-X Rewrite skald-api.md for /skald-api/ and .halls paths
-X Remove video, camera, screenshare, simulcast, SVC, and WebM recording instructions
-X Add Ogg Opus recording documentation
-X Document that incoming video is rejected by design
-X Document that Skald protocol/API compatibility is intentionally broken
-X Document any required external ffmpeg dependency if the recorder uses ffmpeg
-X Check docs for old real-person sample usernames and replace with placeholders
-
-# ============================================================================
-# PHASE 9: VERIFICATION AND CLEANUP
-# ============================================================================
-
-X Run `go test ./...` and fix failures caused by the fork work
-X Run `go vet ./...` and resolve real issues
-X Run focused webserver/API tests for /hall/ and /skald-api/
-X Run focused protocol/frontend checks if the project has a JS typecheck/test path
-X Search source/docs for remaining Galene names
-X Search source/docs for remaining /group/, .groups, public-groups, galenectl, and galene-api
-X Search source/docs for remaining video, camera, screenshare, simulcast, WebM, and Matroska references
-X Search source/docs for remaining group/room terminology in user-facing surfaces
-X Classify allowed survivors as attribution/history/third-party icon names only
-X Build skald binary
-X Build skaldctl binary
-X Run a local server and verify static pages load without 404s
-X Test: create a hall with skaldctl
-X Test: list halls with skaldctl
-X Test: join one hall from two browsers and confirm two-way audio
-X Test: attempt to publish video and confirm rejection
-X Test: start recording and confirm one Ogg Opus file appears
-- Test: join a third participant during recording and confirm the mix continues
-- Test: mute/unmute during recording and confirm the output reflects it
-X Test: leave all participants and confirm recording stops cleanly
-- Test: play the recording in a standard audio player
-X Test: restart server and verify halls/config load from the new paths
-X Remove obsolete files that are no longer served or built
-- Do a final diff review for accidental compatibility shims or stale Skald naming
++ Default new sessions to high-quality audio.
++ Default new sessions to no browser noise suppression, echo cancellation, or
+  automatic gain control.
++ Ensure the default microphone path also receives explicit audio constraints.
++ Add generated browser notification sounds without binary sound assets.
++ Add a user-facing Notification sounds setting.
++ Play generated sounds for recording start, recording stop, raised hand,
+  lowered hand, user join, user leave, and incoming chat message.
+- Verify sounds play after login in Chromium.
+- Verify sounds play after login in Firefox.
+- Verify sounds never replace screen-reader live-region announcements.
+- Tune sound levels and shapes with live users.