feat(tts): per-sentence glow synced to audio via Web Audio pause detection

When Auto-Play TTS reads a message, wrap its sentences in spans and advance a
glow highlight on detected inter-sentence silences (RMS dips), with a
char-proportional fallback. Comma/clause pauses are detected too, reserved for
a future word-level highlight. Tunable via window.GLOW.
This commit is contained in:
russell@unturf.com 2026-05-29 11:52:08 -04:00
parent 57a5b6f243
commit 0fdb673a70
No known key found for this signature in database
2 changed files with 160 additions and 0 deletions

View file

@ -1381,3 +1381,13 @@ a:hover {
font-size: 12px;
color: var(--text-secondary);
}
/* Per-sentence glow while TTS reads a message aloud */
.tts-sentence {
border-radius: 3px;
transition: background-color 0.12s ease, box-shadow 0.12s ease, color 0.12s ease;
}
.tts-sentence.tts-reading {
background: rgba(255, 214, 92, 0.30);
box-shadow: 0 0 10px 2px rgba(255, 200, 70, 0.55);
}

View file

@ -962,6 +962,154 @@ async function fetchTTSStreaming(cleanText, model, voice) {
};
}
// ---- Per-sentence glow synced to TTS audio via Web Audio pause detection ----
// Listens to the audio's RMS energy and advances the highlighted sentence when
// a long inter-sentence silence is heard. Shorter (comma/clause) pauses are
// detected too, reserved for a future word-level "bouncing ball" glow. Falls
// back to char-proportional timing when Web Audio is unavailable/suspended.
// Tune these live from the browser console (window.GLOW):
window.GLOW = window.GLOW || {
silenceRms: 0.015, // RMS below this counts as silence
sentencePauseMs: 280, // silence at least this long advances one sentence
commaPauseMs: 90, // shorter pauses (clause/comma) — detected, not yet shown
fftSize: 1024
};
let _glowAudioCtx = null;
function _getGlowCtx() {
if (!_glowAudioCtx) {
try { _glowAudioCtx = new (window.AudioContext || window.webkitAudioContext)(); }
catch (e) { return null; }
}
if (_glowAudioCtx.state === 'suspended') { _glowAudioCtx.resume().catch(function () {}); }
return _glowAudioCtx;
}
function wrapSentencesForGlow(container) {
if (!container || container.dataset.glowWrapped === '1') return;
const walker = document.createTreeWalker(container, NodeFilter.SHOW_TEXT, null);
const nodes = [];
let t;
while ((t = walker.nextNode())) { if (t.nodeValue && t.nodeValue.trim()) nodes.push(t); }
if (!nodes.length) return;
const full = nodes.map(function (x) { return x.nodeValue; }).join('');
const parts = full.match(/[^.!?]+[.!?]*\s*/g) || [full];
const ranges = [];
const weights = [];
let start = 0;
for (let i = 0; i < parts.length; i++) {
ranges.push([start, start + parts[i].length]);
start += parts[i].length;
weights.push(parts[i].replace(/\s+/g, ' ').trim().length || 1);
}
function sentAt(pos) {
for (let s = 0; s < ranges.length; s++) { if (pos < ranges[s][1]) return s; }
return ranges.length - 1;
}
let off = 0;
nodes.forEach(function (node) {
const text = node.nodeValue;
const frag = document.createDocumentFragment();
let i = 0;
while (i < text.length) {
const si = sentAt(off + i);
let j = i + 1;
while (j < text.length && sentAt(off + j) === si) j++;
const span = document.createElement('span');
span.className = 'tts-sentence';
span.dataset.si = si;
span.textContent = text.slice(i, j);
frag.appendChild(span);
i = j;
}
off += text.length;
node.parentNode.replaceChild(frag, node);
});
container.dataset.glowWrapped = '1';
container._glowSentenceCount = parts.length;
container._glowWeights = weights;
}
function attachSentenceGlow(audio, playButton) {
if (!audio || !playButton) return;
const wrapper = playButton.closest('.message-wrapper');
const container = wrapper && wrapper.querySelector('.message-content');
if (!container) return;
wrapSentencesForGlow(container);
const spans = container.querySelectorAll('.tts-sentence');
const count = container._glowSentenceCount || 0;
if (!spans.length || count <= 0) return;
let active = -1;
function setActive(idx) {
if (idx === active) return;
active = idx;
spans.forEach(function (s) {
s.classList.toggle('tts-reading', parseInt(s.dataset.si, 10) === idx);
});
}
function clear() { spans.forEach(function (s) { s.classList.remove('tts-reading'); }); }
setActive(0);
// Char-proportional fallback boundaries.
const weights = container._glowWeights || [];
const total = weights.reduce(function (a, b) { return a + b; }, 0) || 1;
const cum = [];
let acc = 0;
for (let i = 0; i < weights.length; i++) { acc += weights[i]; cum.push(acc / total); }
// Web Audio analyser — only route through it when the context is running,
// so we never mute playback on a suspended (autoplay-blocked) context.
let analyser = null;
let data = null;
const ctx = _getGlowCtx();
if (ctx && ctx.state === 'running' && !audio._glowSourced) {
try {
const src = ctx.createMediaElementSource(audio);
analyser = ctx.createAnalyser();
analyser.fftSize = window.GLOW.fftSize;
src.connect(analyser);
analyser.connect(ctx.destination);
data = new Float32Array(analyser.fftSize);
audio._glowSourced = true;
} catch (e) { analyser = null; }
}
let inSilence = false;
let silenceStart = 0;
function rms() {
analyser.getFloatTimeDomainData(data);
let sum = 0;
for (let k = 0; k < data.length; k++) sum += data[k] * data[k];
return Math.sqrt(sum / data.length);
}
function tick() {
if (audio.paused || audio.ended) return;
const ct = audio.currentTime;
if (analyser) {
const e = rms();
if (e < window.GLOW.silenceRms) {
if (!inSilence) { inSilence = true; silenceStart = ct; }
} else if (inSilence) {
const gapMs = (ct - silenceStart) * 1000;
inSilence = false;
if (gapMs >= window.GLOW.sentencePauseMs) {
setActive(Math.min(active + 1, count - 1));
}
}
} else if (isFinite(audio.duration) && audio.duration > 0) {
const frac = ct / audio.duration;
let idx = 0;
while (idx < cum.length - 1 && frac > cum[idx]) idx++;
setActive(idx);
}
requestAnimationFrame(tick);
}
audio.addEventListener('play', function () { requestAnimationFrame(tick); });
audio.addEventListener('ended', clear);
audio.addEventListener('pause', function () { if (audio.ended) clear(); });
if (!audio.paused) requestAnimationFrame(tick);
}
// Function to read text using TTS (for manual button clicks) - now with streaming
async function speakText(text, playButton, messageId) {
console.log("speakText called with:", {text: text, messageId, autoPlay: autoPlayTTS});
@ -1038,6 +1186,7 @@ async function speakTextQueued(text, playButton, messageId) {
const audio = new Audio(audioUrl);
audio.playbackRate = 0.9;
enableDownloadButton(messageId, playButton, audioUrl, voice);
attachSentenceGlow(audio, playButton);
currentQueuedAudio = audio;
audio.onended = () => {
@ -1058,6 +1207,7 @@ async function speakTextQueued(text, playButton, messageId) {
// Use streaming TTS for faster playback start
const result = await fetchTTSStreaming(cleanText, model, voice);
const audio = result.audio;
attachSentenceGlow(audio, playButton);
currentQueuedAudio = audio;
audio.onended = () => {