@pie-players/pie-assessment-toolkit 0.3.43 → 0.3.45

package/dist/services/tts/catalog-span-alignment.js

@@ -0,0 +1,300 @@
+import { extractSpokenText } from "./ssml/spoken-text.js";
+import { createSpeechAlignmentTokenPattern, normalizeTextForSpeech, } from "./text-processing.js";
+// Shared Unicode-aware tokenizer (see `createSpeechAlignmentTokenPattern`), so
+// accented Latin and non-Latin words tokenize the same way here as in the math
+// speech tokenizer. LIMITATION (i18n): the phrase table below is still English,
+// so multi-word operator phrases ("plus or minus", "divided by") only align for
+// English speech; other locales degrade to coarse region highlighting. A fresh
+// pattern per call because of the global flag's mutable `lastIndex`.
+const TOKEN_PATTERN = createSpeechAlignmentTokenPattern();
+const NUMERIC_WORDS = new Map([
+    ["zero", "0"],
+    ["one", "1"],
+    ["two", "2"],
+    ["three", "3"],
+    ["four", "4"],
+    ["five", "5"],
+    ["six", "6"],
+    ["seven", "7"],
+    ["eight", "8"],
+    ["nine", "9"],
+    ["ten", "10"],
+]);
+const SPOKEN_PHRASE_TARGETS = [
+    { phrase: ["plus", "or", "minus"], visible: "±", score: 8 },
+    { phrase: ["square", "root"], visible: "√", score: 8 },
+    { phrase: ["divided", "by"], visible: "/", score: 7 },
+    { phrase: ["all", "over"], visible: "/", score: 7 },
+    { phrase: ["equals"], visible: "=", score: 7 },
+    { phrase: ["equal"], visible: "=", score: 7 },
+    { phrase: ["times"], visible: "\u2062", score: 7 },
+    { phrase: ["multiplied", "by"], visible: "\u2062", score: 7 },
+    { phrase: ["over"], visible: "/", score: 6 },
+    { phrase: ["squared"], visible: "²", score: 6 },
+    { phrase: ["squared"], visible: "2", score: 4 },
+    { phrase: ["negative"], visible: "-", score: 5 },
+    { phrase: ["minus"], visible: "-", score: 5 },
+    { phrase: ["plus"], visible: "+", score: 5 },
+];
+const tokenize = (text) => {
+    const tokens = [];
+    for (const match of text.matchAll(TOKEN_PATTERN)) {
+        const value = match[0];
+        const start = match.index ?? 0;
+        tokens.push({
+            text: value,
+            normalized: value.toLowerCase(),
+            start,
+            end: start + value.length,
+        });
+    }
+    return tokens;
+};
+const normalizedTokenValue = (token) => NUMERIC_WORDS.get(token.normalized) || token.normalized;
+const createCandidates = (spokenTokens, visibleTokens) => {
+    const candidates = [];
+    for (let spokenIndex = 0; spokenIndex < spokenTokens.length; spokenIndex++) {
+        for (let visibleIndex = 0; visibleIndex < visibleTokens.length; visibleIndex++) {
+            const spokenValue = normalizedTokenValue(spokenTokens[spokenIndex]);
+            const visibleValue = normalizedTokenValue(visibleTokens[visibleIndex]);
+            if (spokenValue === visibleValue) {
+                candidates.push({
+                    spokenStartToken: spokenIndex,
+                    spokenEndToken: spokenIndex + 1,
+                    visibleToken: visibleIndex,
+                    score: 10,
+                });
+            }
+        }
+        for (const phraseTarget of SPOKEN_PHRASE_TARGETS) {
+            const phraseTokens = spokenTokens.slice(spokenIndex, spokenIndex + phraseTarget.phrase.length);
+            if (phraseTokens.length !== phraseTarget.phrase.length)
+                continue;
+            if (!phraseTokens.every((token, tokenIndex) => token.normalized === phraseTarget.phrase[tokenIndex])) {
+                continue;
+            }
+            for (let visibleIndex = 0; visibleIndex < visibleTokens.length; visibleIndex++) {
+                if (visibleTokens[visibleIndex].normalized !== phraseTarget.visible) {
+                    continue;
+                }
+                candidates.push({
+                    spokenStartToken: spokenIndex,
+                    spokenEndToken: spokenIndex + phraseTarget.phrase.length,
+                    visibleToken: visibleIndex,
+                    score: phraseTarget.score,
+                });
+            }
+        }
+    }
+    return candidates;
+};
+const computeAnchors = (spokenTokens, visibleTokens) => {
+    const candidates = createCandidates(spokenTokens, visibleTokens).sort((left, right) => spokenTokens[left.spokenStartToken].start -
+        spokenTokens[right.spokenStartToken].start ||
+        visibleTokens[left.visibleToken].start - visibleTokens[right.visibleToken].start);
+    const bestScore = new Array(candidates.length).fill(0);
+    const previous = new Array(candidates.length).fill(-1);
+    for (let i = 0; i < candidates.length; i++) {
+        const candidate = candidates[i];
+        bestScore[i] = candidate.score;
+        for (let j = 0; j < i; j++) {
+            const prior = candidates[j];
+            if (prior.spokenEndToken <= candidate.spokenStartToken &&
+                prior.visibleToken < candidate.visibleToken) {
+                const spokenGap = candidate.spokenStartToken - prior.spokenEndToken;
+                const visibleGap = candidate.visibleToken - prior.visibleToken - 1;
+                const gapPenalty = Math.max(0, Math.abs(spokenGap - visibleGap) - 2);
+                const score = bestScore[j] + candidate.score - gapPenalty;
+                if (score > bestScore[i]) {
+                    bestScore[i] = score;
+                    previous[i] = j;
+                }
+            }
+        }
+    }
+    let bestIndex = -1;
+    for (let i = 0; i < bestScore.length; i++) {
+        if (bestIndex < 0 || bestScore[i] > bestScore[bestIndex]) {
+            bestIndex = i;
+        }
+    }
+    if (bestIndex < 0)
+        return [];
+    const chosen = [];
+    let current = bestIndex;
+    while (current >= 0) {
+        chosen.unshift(candidates[current]);
+        current = previous[current];
+    }
+    return chosen.map((candidate) => {
+        const firstSpoken = spokenTokens[candidate.spokenStartToken];
+        const lastSpoken = spokenTokens[candidate.spokenEndToken - 1];
+        const visible = visibleTokens[candidate.visibleToken];
+        return {
+            spokenStart: firstSpoken.start,
+            spokenEnd: lastSpoken.end,
+            visibleStart: visible.start,
+            visibleEnd: visible.end,
+            score: candidate.score,
+        };
+    });
+};
+const scoreConfidence = (anchors, spokenTokens, visibleTokens) => {
+    if (anchors.length === 0)
+        return 0;
+    const tokenCoverage = anchors.length / Math.max(1, Math.min(spokenTokens.length, visibleTokens.length));
+    const averageScore = anchors.reduce((total, anchor) => total + anchor.score, 0) /
+        anchors.length /
+        10;
+    return Math.min(1, tokenCoverage * averageScore);
+};
+export const createCatalogSpanAlignment = (args) => {
+    const extracted = extractSpokenText(args.speechText);
+    const visibleText = normalizeTextForSpeech(args.visibleText);
+    const spokenText = extracted.spokenText;
+    const spokenTokens = tokenize(spokenText);
+    const visibleTokens = tokenize(visibleText);
+    const anchors = extracted.unsupportedSemantic
+        ? []
+        : computeAnchors(spokenTokens, visibleTokens);
+    const confidence = scoreConfidence(anchors, spokenTokens, visibleTokens);
+    const isExact = spokenText === visibleText && spokenText.length > 0;
+    const hasShortSingleExactAnchor = anchors.length === 1 &&
+        anchors[0].score === 10 &&
+        spokenTokens.length <= 2 &&
+        visibleTokens.length <= 2;
+    const playbackMode = extracted.unsupportedSemantic
+        ? "region-fallback"
+        : isExact
+            ? "exact-word"
+            : anchors.length >= 2 || hasShortSingleExactAnchor
+                ? confidence >= 0.12
+                    ? "anchor-span"
+                    : "region-fallback"
+                : "region-fallback";
+    const boundaryOffsetMode = extracted.unsupportedSemantic
+        ? "unsupported"
+        : extracted.hasMarkup
+            ? "raw-ssml"
+            : "plain-spoken-text";
+    return {
+        speechText: args.speechText,
+        spokenText,
+        visibleText,
+        playbackMode,
+        boundaryOffsetMode,
+        anchors,
+        confidence,
+        spokenTokens,
+        visibleTokens,
+        rawToSpokenOffsetMap: extracted.rawToSpokenOffsetMap,
+    };
+};
+const findTokenAtOffset = (tokens, offset) => tokens.find((token) => token.start <= offset && offset < token.end) || null;
+const normalizeBoundaryWord = (word) => {
+    if (!word)
+        return null;
+    if (/^<[^>]+>$/.test(word.trim()))
+        return null;
+    const token = tokenize(word)[0];
+    if (!token)
+        return null;
+    return NUMERIC_WORDS.get(token.normalized) || token.normalized;
+};
+const candidateMatchesBoundaryWord = (alignment, candidate, boundaryWord) => {
+    const normalizedBoundary = normalizeBoundaryWord(boundaryWord);
+    if (!candidate || !normalizedBoundary)
+        return false;
+    const token = findTokenAtOffset(alignment.spokenTokens, candidate.start);
+    if (!token)
+        return false;
+    return normalizedTokenValue(token) === normalizedBoundary;
+};
+const mapRawOffsetToSpokenOffset = (alignment, position) => {
+    const direct = alignment.rawToSpokenOffsetMap.get(position);
+    if (direct !== undefined)
+        return direct;
+    for (let offset = position; offset < alignment.speechText.length; offset++) {
+        const mapped = alignment.rawToSpokenOffsetMap.get(offset);
+        if (mapped !== undefined)
+            return mapped;
+        if (alignment.speechText[offset] === ">")
+            break;
+    }
+    return null;
+};
+const resolveRawBoundaryOffset = (alignment, position, safeLength) => {
+    const mappedStart = mapRawOffsetToSpokenOffset(alignment, position);
+    if (mappedStart === null)
+        return null;
+    const mappedEnd = mapRawOffsetToSpokenOffset(alignment, Math.max(position, position + safeLength - 1));
+    const token = findTokenAtOffset(alignment.spokenTokens, mappedStart);
+    const mappedLength = mappedEnd !== null && mappedEnd >= mappedStart
+        ? mappedEnd - mappedStart + 1
+        : token
+            ? token.end - mappedStart
+            : safeLength;
+    return { start: mappedStart, length: Math.max(1, mappedLength) };
+};
+const resolvePlainBoundaryOffset = (alignment, position, safeLength) => {
+    if (position < 0 || position >= alignment.spokenText.length)
+        return null;
+    const token = findTokenAtOffset(alignment.spokenTokens, position);
+    if (!token)
+        return null;
+    return { start: position, length: Math.min(safeLength, token.end - position) };
+};
+export const resolveSpokenBoundaryOffset = (alignment, position, length = 1, boundaryWord) => {
+    if (alignment.boundaryOffsetMode === "unsupported" ||
+        !Number.isFinite(position)) {
+        return null;
+    }
+    const safeLength = Math.max(1, Number.isFinite(length) ? length : 1);
+    if (boundaryWord && /^<[^>]+>$/.test(boundaryWord.trim()))
+        return null;
+    if (alignment.boundaryOffsetMode === "raw-ssml") {
+        const rawCandidate = resolveRawBoundaryOffset(alignment, position, safeLength);
+        const plainCandidate = resolvePlainBoundaryOffset(alignment, position, safeLength);
+        if (boundaryWord) {
+            if (candidateMatchesBoundaryWord(alignment, rawCandidate, boundaryWord)) {
+                return rawCandidate;
+            }
+            if (candidateMatchesBoundaryWord(alignment, plainCandidate, boundaryWord)) {
+                return plainCandidate;
+            }
+            if (normalizeBoundaryWord(boundaryWord))
+                return null;
+        }
+        return rawCandidate || plainCandidate;
+    }
+    return (resolvePlainBoundaryOffset(alignment, position, safeLength) ||
+        resolveRawBoundaryOffset(alignment, position, safeLength));
+};
+export const resolveVisibleSpanForBoundary = (alignment, spokenOffset) => {
+    if (alignment.playbackMode === "region-fallback")
+        return null;
+    const exactAnchor = alignment.anchors.find((anchor) => anchor.spokenStart <= spokenOffset && spokenOffset < anchor.spokenEnd);
+    if (exactAnchor) {
+        return { start: exactAnchor.visibleStart, end: exactAnchor.visibleEnd };
+    }
+    let previous = null;
+    let next = null;
+    for (const anchor of alignment.anchors) {
+        if (anchor.spokenEnd <= spokenOffset) {
+            previous = anchor;
+            continue;
+        }
+        if (anchor.spokenStart > spokenOffset) {
+            next = anchor;
+            break;
+        }
+    }
+    if (previous && next) {
+        return {
+            start: Math.min(previous.visibleStart, next.visibleStart),
+            end: Math.max(previous.visibleEnd, next.visibleEnd),
+        };
+    }
+    return null;
+};
+//# sourceMappingURL=catalog-span-alignment.js.map

package/dist/services/tts/catalog-span-alignment.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"catalog-span-alignment.js","sourceRoot":"","sources":["../../../src/services/tts/catalog-span-alignment.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,iBAAiB,EAAE,MAAM,uBAAuB,CAAC;AAC1D,OAAO,EACN,iCAAiC,EACjC,sBAAsB,GACtB,MAAM,sBAAsB,CAAC;AA+C9B,+EAA+E;AAC/E,+EAA+E;AAC/E,gFAAgF;AAChF,gFAAgF;AAChF,+EAA+E;AAC/E,qEAAqE;AACrE,MAAM,aAAa,GAAG,iCAAiC,EAAE,CAAC;AAE1D,MAAM,aAAa,GAAG,IAAI,GAAG,CAAiB;IAC7C,CAAC,MAAM,EAAE,GAAG,CAAC;IACb,CAAC,KAAK,EAAE,GAAG,CAAC;IACZ,CAAC,KAAK,EAAE,GAAG,CAAC;IACZ,CAAC,OAAO,EAAE,GAAG,CAAC;IACd,CAAC,MAAM,EAAE,GAAG,CAAC;IACb,CAAC,MAAM,EAAE,GAAG,CAAC;IACb,CAAC,KAAK,EAAE,GAAG,CAAC;IACZ,CAAC,OAAO,EAAE,GAAG,CAAC;IACd,CAAC,OAAO,EAAE,GAAG,CAAC;IACd,CAAC,MAAM,EAAE,GAAG,CAAC;IACb,CAAC,KAAK,EAAE,IAAI,CAAC;CACb,CAAC,CAAC;AAEH,MAAM,qBAAqB,GAItB;IACJ,EAAE,MAAM,EAAE,CAAC,MAAM,EAAE,IAAI,EAAE,OAAO,CAAC,EAAE,OAAO,EAAE,GAAG,EAAE,KAAK,EAAE,CAAC,EAAE;IAC3D,EAAE,MAAM,EAAE,CAAC,QAAQ,EAAE,MAAM,CAAC,EAAE,OAAO,EAAE,GAAG,EAAE,KAAK,EAAE,CAAC,EAAE;IACtD,EAAE,MAAM,EAAE,CAAC,SAAS,EAAE,IAAI,CAAC,EAAE,OAAO,EAAE,GAAG,EAAE,KAAK,EAAE,CAAC,EAAE;IACrD,EAAE,MAAM,EAAE,CAAC,KAAK,EAAE,MAAM,CAAC,EAAE,OAAO,EAAE,GAAG,EAAE,KAAK,EAAE,CAAC,EAAE;IACnD,EAAE,MAAM,EAAE,CAAC,QAAQ,CAAC,EAAE,OAAO,EAAE,GAAG,EAAE,KAAK,EAAE,CAAC,EAAE;IAC9C,EAAE,MAAM,EAAE,CAAC,OAAO,CAAC,EAAE,OAAO,EAAE,GAAG,EAAE,KAAK,EAAE,CAAC,EAAE;IAC7C,EAAE,MAAM,EAAE,CAAC,OAAO,CAAC,EAAE,OAAO,EAAE,QAAQ,EAAE,KAAK,EAAE,CAAC,EAAE;IAClD,EAAE,MAAM,EAAE,CAAC,YAAY,EAAE,IAAI,CAAC,EAAE,OAAO,EAAE,QAAQ,EAAE,KAAK,EAAE,CAAC,EAAE;IAC7D,EAAE,MAAM,EAAE,CAAC,MAAM,CAAC,EAAE,OAAO,EAAE,GAAG,EAAE,KAAK,EAAE,CAAC,EAAE;IAC5C,EAAE,MAAM,EAAE,CAAC,SAAS,CAAC,EAAE,OAAO,EAAE,GAAG,EAAE,KAAK,EAAE,CAAC,EAAE;IAC/C,EAAE,MAAM,EAAE,CAAC,SAAS,CAAC,EAAE,OAAO,EAAE,GAAG,EAAE,KAAK,EAAE,CAAC,EAAE;IAC/C,EAAE,MAAM,EAAE,CAAC,UAAU,CAAC,EAAE,OAAO,EAAE,GAAG,EAAE,KAAK,EAAE,CAAC,EAAE;IAChD,EAAE,MAAM,EAAE,CAAC,OAAO,CAAC,EAAE,OAAO,EAAE,GAAG,EAAE,KAAK,EAAE,CAAC,EAAE;IAC7C,EAAE,MAAM,EAAE,CAAC,MAAM,CAAC,EAAE,OAAO,EAAE,GAAG,EAAE,KAAK,EAAE,CAAC,EAAE;CAC5C,CAAC;AAEF,MAAM,QAAQ,GAAG,CAAC,IAAY,EAAsB,EAAE;IACrD,MAAM,MAAM,GAAuB,EAAE,CAAC;IACtC,KAAK,MAAM,KAAK,IAAI,IAAI,CAAC,QAAQ,CAAC,aAAa,CAAC,EAAE,CAAC;QAClD,MAAM,KAAK,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC;QACvB,MAAM,KAAK,GAAG,KAAK,CAAC,KAAK,IAAI,CAAC,CAAC;QAC/B,MAAM,CAAC,IAAI,CAAC;YACX,IAAI,EAAE,KAAK;YACX,UAAU,EAAE,KAAK,CAAC,WAAW,EAAE;YAC/B,KAAK;YACL,GAAG,EAAE,KAAK,GAAG,KAAK,CAAC,MAAM;SACzB,CAAC,CAAC;IACJ,CAAC;IACD,OAAO,MAAM,CAAC;AACf,CAAC,CAAC;AAEF,MAAM,oBAAoB,GAAG,CAAC,KAAuB,EAAU,EAAE,CAChE,aAAa,CAAC,GAAG,CAAC,KAAK,CAAC,UAAU,CAAC,IAAI,KAAK,CAAC,UAAU,CAAC;AAEzD,MAAM,gBAAgB,GAAG,CACxB,YAAgC,EAChC,aAAiC,EACd,EAAE;IACrB,MAAM,UAAU,GAAqB,EAAE,CAAC;IACxC,KAAK,IAAI,WAAW,GAAG,CAAC,EAAE,WAAW,GAAG,YAAY,CAAC,MAAM,EAAE,WAAW,EAAE,EAAE,CAAC;QAC5E,KAAK,IAAI,YAAY,GAAG,CAAC,EAAE,YAAY,GAAG,aAAa,CAAC,MAAM,EAAE,YAAY,EAAE,EAAE,CAAC;YAChF,MAAM,WAAW,GAAG,oBAAoB,CAAC,YAAY,CAAC,WAAW,CAAC,CAAC,CAAC;YACpE,MAAM,YAAY,GAAG,oBAAoB,CAAC,aAAa,CAAC,YAAY,CAAC,CAAC,CAAC;YACvE,IAAI,WAAW,KAAK,YAAY,EAAE,CAAC;gBAClC,UAAU,CAAC,IAAI,CAAC;oBACf,gBAAgB,EAAE,WAAW;oBAC7B,cAAc,EAAE,WAAW,GAAG,CAAC;oBAC/B,YAAY,EAAE,YAAY;oBAC1B,KAAK,EAAE,EAAE;iBACT,CAAC,CAAC;YACJ,CAAC;QACF,CAAC;QAED,KAAK,MAAM,YAAY,IAAI,qBAAqB,EAAE,CAAC;YAClD,MAAM,YAAY,GAAG,YAAY,CAAC,KAAK,CACtC,WAAW,EACX,WAAW,GAAG,YAAY,CAAC,MAAM,CAAC,MAAM,CACxC,CAAC;YACF,IAAI,YAAY,CAAC,MAAM,KAAK,YAAY,CAAC,MAAM,CAAC,MAAM;gBAAE,SAAS;YACjE,IACC,CAAC,YAAY,CAAC,KAAK,CAClB,CAAC,KAAK,EAAE,UAAU,EAAE,EAAE,CACrB,KAAK,CAAC,UAAU,KAAK,YAAY,CAAC,MAAM,CAAC,UAAU,CAAC,CACrD,EACA,CAAC;gBACF,SAAS;YACV,CAAC;YACD,KACC,IAAI,YAAY,GAAG,CAAC,EACpB,YAAY,GAAG,aAAa,CAAC,MAAM,EACnC,YAAY,EAAE,EACb,CAAC;gBACF,IAAI,aAAa,CAAC,YAAY,CAAC,CAAC,UAAU,KAAK,YAAY,CAAC,OAAO,EAAE,CAAC;oBACrE,SAAS;gBACV,CAAC;gBACD,UAAU,CAAC,IAAI,CAAC;oBACf,gBAAgB,EAAE,WAAW;oBAC7B,cAAc,EAAE,WAAW,GAAG,YAAY,CAAC,MAAM,CAAC,MAAM;oBACxD,YAAY,EAAE,YAAY;oBAC1B,KAAK,EAAE,YAAY,CAAC,KAAK;iBACzB,CAAC,CAAC;YACJ,CAAC;QACF,CAAC;IACF,CAAC;IACD,OAAO,UAAU,CAAC;AACnB,CAAC,CAAC;AAEF,MAAM,cAAc,GAAG,CACtB,YAAgC,EAChC,aAAiC,EACX,EAAE;IACxB,MAAM,UAAU,GAAG,gBAAgB,CAAC,YAAY,EAAE,aAAa,CAAC,CAAC,IAAI,CACpE,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CACf,YAAY,CAAC,IAAI,CAAC,gBAAgB,CAAC,CAAC,KAAK;QACxC,YAAY,CAAC,KAAK,CAAC,gBAAgB,CAAC,CAAC,KAAK;QAC3C,aAAa,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC,KAAK,GAAG,aAAa,CAAC,KAAK,CAAC,YAAY,CAAC,CAAC,KAAK,CACjF,CAAC;IACF,MAAM,SAAS,GAAG,IAAI,KAAK,CAAC,UAAU,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,CAAC,CAAa,CAAC;IACnE,MAAM,QAAQ,GAAG,IAAI,KAAK,CAAC,UAAU,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAa,CAAC;IAEnE,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,UAAU,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QAC5C,MAAM,SAAS,GAAG,UAAU,CAAC,CAAC,CAAC,CAAC;QAChC,SAAS,CAAC,CAAC,CAAC,GAAG,SAAS,CAAC,KAAK,CAAC;QAC/B,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;YAC5B,MAAM,KAAK,GAAG,UAAU,CAAC,CAAC,CAAC,CAAC;YAC5B,IACC,KAAK,CAAC,cAAc,IAAI,SAAS,CAAC,gBAAgB;gBAClD,KAAK,CAAC,YAAY,GAAG,SAAS,CAAC,YAAY,EAC1C,CAAC;gBACF,MAAM,SAAS,GAAG,SAAS,CAAC,gBAAgB,GAAG,KAAK,CAAC,cAAc,CAAC;gBACpE,MAAM,UAAU,GAAG,SAAS,CAAC,YAAY,GAAG,KAAK,CAAC,YAAY,GAAG,CAAC,CAAC;gBACnE,MAAM,UAAU,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,SAAS,GAAG,UAAU,CAAC,GAAG,CAAC,CAAC,CAAC;gBACrE,MAAM,KAAK,GAAG,SAAS,CAAC,CAAC,CAAC,GAAG,SAAS,CAAC,KAAK,GAAG,UAAU,CAAC;gBAC1D,IAAI,KAAK,GAAG,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC;oBAC1B,SAAS,CAAC,CAAC,CAAC,GAAG,KAAK,CAAC;oBACrB,QAAQ,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;gBACjB,CAAC;YACF,CAAC;QACF,CAAC;IACF,CAAC;IAED,IAAI,SAAS,GAAG,CAAC,CAAC,CAAC;IACnB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,SAAS,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QAC3C,IAAI,SAAS,GAAG,CAAC,IAAI,SAAS,CAAC,CAAC,CAAC,GAAG,SAAS,CAAC,SAAS,CAAC,EAAE,CAAC;YAC1D,SAAS,GAAG,CAAC,CAAC;QACf,CAAC;IACF,CAAC;IACD,IAAI,SAAS,GAAG,CAAC;QAAE,OAAO,EAAE,CAAC;IAE7B,MAAM,MAAM,GAAqB,EAAE,CAAC;IACpC,IAAI,OAAO,GAAG,SAAS,CAAC;IACxB,OAAO,OAAO,IAAI,CAAC,EAAE,CAAC;QACrB,MAAM,CAAC,OAAO,CAAC,UAAU,CAAC,OAAO,CAAC,CAAC,CAAC;QACpC,OAAO,GAAG,QAAQ,CAAC,OAAO,CAAC,CAAC;IAC7B,CAAC;IAED,OAAO,MAAM,CAAC,GAAG,CAAC,CAAC,SAAS,EAAE,EAAE;QAC/B,MAAM,WAAW,GAAG,YAAY,CAAC,SAAS,CAAC,gBAAgB,CAAC,CAAC;QAC7D,MAAM,UAAU,GAAG,YAAY,CAAC,SAAS,CAAC,cAAc,GAAG,CAAC,CAAC,CAAC;QAC9D,MAAM,OAAO,GAAG,aAAa,CAAC,SAAS,CAAC,YAAY,CAAC,CAAC;QACtD,OAAO;YACN,WAAW,EAAE,WAAW,CAAC,KAAK;YAC9B,SAAS,EAAE,UAAU,CAAC,GAAG;YACzB,YAAY,EAAE,OAAO,CAAC,KAAK;YAC3B,UAAU,EAAE,OAAO,CAAC,GAAG;YACvB,KAAK,EAAE,SAAS,CAAC,KAAK;SACtB,CAAC;IACH,CAAC,CAAC,CAAC;AACJ,CAAC,CAAC;AAEF,MAAM,eAAe,GAAG,CACvB,OAA4B,EAC5B,YAAgC,EAChC,aAAiC,EACxB,EAAE;IACX,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,CAAC,CAAC;IACnC,MAAM,aAAa,GAClB,OAAO,CAAC,MAAM,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,YAAY,CAAC,MAAM,EAAE,aAAa,CAAC,MAAM,CAAC,CAAC,CAAC;IACnF,MAAM,YAAY,GACjB,OAAO,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,MAAM,EAAE,EAAE,CAAC,KAAK,GAAG,MAAM,CAAC,KAAK,EAAE,CAAC,CAAC;QAC1D,OAAO,CAAC,MAAM;QACd,EAAE,CAAC;IACJ,OAAO,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,aAAa,GAAG,YAAY,CAAC,CAAC;AAClD,CAAC,CAAC;AAEF,MAAM,CAAC,MAAM,0BAA0B,GAAG,CAAC,IAG1C,EAAwB,EAAE;IAC1B,MAAM,SAAS,GAAG,iBAAiB,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC;IACrD,MAAM,WAAW,GAAG,sBAAsB,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC;IAC7D,MAAM,UAAU,GAAG,SAAS,CAAC,UAAU,CAAC;IACxC,MAAM,YAAY,GAAG,QAAQ,CAAC,UAAU,CAAC,CAAC;IAC1C,MAAM,aAAa,GAAG,QAAQ,CAAC,WAAW,CAAC,CAAC;IAC5C,MAAM,OAAO,GAAG,SAAS,CAAC,mBAAmB;QAC5C,CAAC,CAAC,EAAE;QACJ,CAAC,CAAC,cAAc,CAAC,YAAY,EAAE,aAAa,CAAC,CAAC;IAC/C,MAAM,UAAU,GAAG,eAAe,CAAC,OAAO,EAAE,YAAY,EAAE,aAAa,CAAC,CAAC;IACzE,MAAM,OAAO,GAAG,UAAU,KAAK,WAAW,IAAI,UAAU,CAAC,MAAM,GAAG,CAAC,CAAC;IACpE,MAAM,yBAAyB,GAC9B,OAAO,CAAC,MAAM,KAAK,CAAC;QACpB,OAAO,CAAC,CAAC,CAAC,CAAC,KAAK,KAAK,EAAE;QACvB,YAAY,CAAC,MAAM,IAAI,CAAC;QACxB,aAAa,CAAC,MAAM,IAAI,CAAC,CAAC;IAC3B,MAAM,YAAY,GAA6B,SAAS,CAAC,mBAAmB;QAC3E,CAAC,CAAC,iBAAiB;QACnB,CAAC,CAAC,OAAO;YACR,CAAC,CAAC,YAAY;YACd,CAAC,CAAC,OAAO,CAAC,MAAM,IAAI,CAAC,IAAI,yBAAyB;gBACjD,CAAC,CAAC,UAAU,IAAI,IAAI;oBACnB,CAAC,CAAC,aAAa;oBACf,CAAC,CAAC,iBAAiB;gBACpB,CAAC,CAAC,iBAAiB,CAAC;IACvB,MAAM,kBAAkB,GAAuB,SAAS,CAAC,mBAAmB;QAC3E,CAAC,CAAC,aAAa;QACf,CAAC,CAAC,SAAS,CAAC,SAAS;YACpB,CAAC,CAAC,UAAU;YACZ,CAAC,CAAC,mBAAmB,CAAC;IAExB,OAAO;QACN,UAAU,EAAE,IAAI,CAAC,UAAU;QAC3B,UAAU;QACV,WAAW;QACX,YAAY;QACZ,kBAAkB;QAClB,OAAO;QACP,UAAU;QACV,YAAY;QACZ,aAAa;QACb,oBAAoB,EAAE,SAAS,CAAC,oBAAoB;KACpD,CAAC;AACH,CAAC,CAAC;AAEF,MAAM,iBAAiB,GAAG,CACzB,MAA0B,EAC1B,MAAc,EACY,EAAE,CAC5B,MAAM,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,KAAK,IAAI,MAAM,IAAI,MAAM,GAAG,KAAK,CAAC,GAAG,CAAC,IAAI,IAAI,CAAC;AAE7E,MAAM,qBAAqB,GAAG,CAAC,IAAa,EAAiB,EAAE;IAC9D,IAAI,CAAC,IAAI;QAAE,OAAO,IAAI,CAAC;IACvB,IAAI,WAAW,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,EAAE,CAAC;QAAE,OAAO,IAAI,CAAC;IAC/C,MAAM,KAAK,GAAG,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC;IAChC,IAAI,CAAC,KAAK;QAAE,OAAO,IAAI,CAAC;IACxB,OAAO,aAAa,CAAC,GAAG,CAAC,KAAK,CAAC,UAAU,CAAC,IAAI,KAAK,CAAC,UAAU,CAAC;AAChE,CAAC,CAAC;AAEF,MAAM,4BAA4B,GAAG,CACpC,SAA+B,EAC/B,SAAmD,EACnD,YAAqB,EACX,EAAE;IACZ,MAAM,kBAAkB,GAAG,qBAAqB,CAAC,YAAY,CAAC,CAAC;IAC/D,IAAI,CAAC,SAAS,IAAI,CAAC,kBAAkB;QAAE,OAAO,KAAK,CAAC;IACpD,MAAM,KAAK,GAAG,iBAAiB,CAAC,SAAS,CAAC,YAAY,EAAE,SAAS,CAAC,KAAK,CAAC,CAAC;IACzE,IAAI,CAAC,KAAK;QAAE,OAAO,KAAK,CAAC;IACzB,OAAO,oBAAoB,CAAC,KAAK,CAAC,KAAK,kBAAkB,CAAC;AAC3D,CAAC,CAAC;AAEF,MAAM,0BAA0B,GAAG,CAClC,SAA+B,EAC/B,QAAgB,EACA,EAAE;IAClB,MAAM,MAAM,GAAG,SAAS,CAAC,oBAAoB,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC;IAC5D,IAAI,MAAM,KAAK,SAAS;QAAE,OAAO,MAAM,CAAC;IACxC,KAAK,IAAI,MAAM,GAAG,QAAQ,EAAE,MAAM,GAAG,SAAS,CAAC,UAAU,CAAC,MAAM,EAAE,MAAM,EAAE,EAAE,CAAC;QAC5E,MAAM,MAAM,GAAG,SAAS,CAAC,oBAAoB,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC;QAC1D,IAAI,MAAM,KAAK,SAAS;YAAE,OAAO,MAAM,CAAC;QACxC,IAAI,SAAS,CAAC,UAAU,CAAC,MAAM,CAAC,KAAK,GAAG;YAAE,MAAM;IACjD,CAAC;IACD,OAAO,IAAI,CAAC;AACb,CAAC,CAAC;AAEF,MAAM,wBAAwB,GAAG,CAChC,SAA+B,EAC/B,QAAgB,EAChB,UAAkB,EACyB,EAAE;IAC7C,MAAM,WAAW,GAAG,0BAA0B,CAAC,SAAS,EAAE,QAAQ,CAAC,CAAC;IACpE,IAAI,WAAW,KAAK,IAAI;QAAE,OAAO,IAAI,CAAC;IACtC,MAAM,SAAS,GAAG,0BAA0B,CAC3C,SAAS,EACT,IAAI,CAAC,GAAG,CAAC,QAAQ,EAAE,QAAQ,GAAG,UAAU,GAAG,CAAC,CAAC,CAC7C,CAAC;IACF,MAAM,KAAK,GAAG,iBAAiB,CAAC,SAAS,CAAC,YAAY,EAAE,WAAW,CAAC,CAAC;IACrE,MAAM,YAAY,GACjB,SAAS,KAAK,IAAI,IAAI,SAAS,IAAI,WAAW;QAC7C,CAAC,CAAC,SAAS,GAAG,WAAW,GAAG,CAAC;QAC7B,CAAC,CAAC,KAAK;YACN,CAAC,CAAC,KAAK,CAAC,GAAG,GAAG,WAAW;YACzB,CAAC,CAAC,UAAU,CAAC;IAChB,OAAO,EAAE,KAAK,EAAE,WAAW,EAAE,MAAM,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,YAAY,CAAC,EAAE,CAAC;AAClE,CAAC,CAAC;AAEF,MAAM,0BAA0B,GAAG,CAClC,SAA+B,EAC/B,QAAgB,EAChB,UAAkB,EACyB,EAAE;IAC7C,IAAI,QAAQ,GAAG,CAAC,IAAI,QAAQ,IAAI,SAAS,CAAC,UAAU,CAAC,MAAM;QAAE,OAAO,IAAI,CAAC;IACzE,MAAM,KAAK,GAAG,iBAAiB,CAAC,SAAS,CAAC,YAAY,EAAE,QAAQ,CAAC,CAAC;IAClE,IAAI,CAAC,KAAK;QAAE,OAAO,IAAI,CAAC;IACxB,OAAO,EAAE,KAAK,EAAE,QAAQ,EAAE,MAAM,EAAE,IAAI,CAAC,GAAG,CAAC,UAAU,EAAE,KAAK,CAAC,GAAG,GAAG,QAAQ,CAAC,EAAE,CAAC;AAChF,CAAC,CAAC;AAEF,MAAM,CAAC,MAAM,2BAA2B,GAAG,CAC1C,SAA+B,EAC/B,QAAgB,EAChB,MAAM,GAAG,CAAC,EACV,YAAqB,EACsB,EAAE;IAC7C,IACC,SAAS,CAAC,kBAAkB,KAAK,aAAa;QAC9C,CAAC,MAAM,CAAC,QAAQ,CAAC,QAAQ,CAAC,EACzB,CAAC;QACF,OAAO,IAAI,CAAC;IACb,CAAC;IACD,MAAM,UAAU,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IACrE,IAAI,YAAY,IAAI,WAAW,CAAC,IAAI,CAAC,YAAY,CAAC,IAAI,EAAE,CAAC;QAAE,OAAO,IAAI,CAAC;IACvE,IAAI,SAAS,CAAC,kBAAkB,KAAK,UAAU,EAAE,CAAC;QACjD,MAAM,YAAY,GAAG,wBAAwB,CAC5C,SAAS,EACT,QAAQ,EACR,UAAU,CACV,CAAC;QACF,MAAM,cAAc,GAAG,0BAA0B,CAChD,SAAS,EACT,QAAQ,EACR,UAAU,CACV,CAAC;QACF,IAAI,YAAY,EAAE,CAAC;YAClB,IACC,4BAA4B,CAAC,SAAS,EAAE,YAAY,EAAE,YAAY,CAAC,EAClE,CAAC;gBACF,OAAO,YAAY,CAAC;YACrB,CAAC;YACD,IACC,4BAA4B,CAAC,SAAS,EAAE,cAAc,EAAE,YAAY,CAAC,EACpE,CAAC;gBACF,OAAO,cAAc,CAAC;YACvB,CAAC;YACD,IAAI,qBAAqB,CAAC,YAAY,CAAC;gBAAE,OAAO,IAAI,CAAC;QACtD,CAAC;QACD,OAAO,YAAY,IAAI,cAAc,CAAC;IACvC,CAAC;IACD,OAAO,CACN,0BAA0B,CAAC,SAAS,EAAE,QAAQ,EAAE,UAAU,CAAC;QAC3D,wBAAwB,CAAC,SAAS,EAAE,QAAQ,EAAE,UAAU,CAAC,CACzD,CAAC;AACH,CAAC,CAAC;AAEF,MAAM,CAAC,MAAM,6BAA6B,GAAG,CAC5C,SAA+B,EAC/B,YAAoB,EACoB,EAAE;IAC1C,IAAI,SAAS,CAAC,YAAY,KAAK,iBAAiB;QAAE,OAAO,IAAI,CAAC;IAC9D,MAAM,WAAW,GAAG,SAAS,CAAC,OAAO,CAAC,IAAI,CACzC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,WAAW,IAAI,YAAY,IAAI,YAAY,GAAG,MAAM,CAAC,SAAS,CACjF,CAAC;IACF,IAAI,WAAW,EAAE,CAAC;QACjB,OAAO,EAAE,KAAK,EAAE,WAAW,CAAC,YAAY,EAAE,GAAG,EAAE,WAAW,CAAC,UAAU,EAAE,CAAC;IACzE,CAAC;IAED,IAAI,QAAQ,GAA6B,IAAI,CAAC;IAC9C,IAAI,IAAI,GAA6B,IAAI,CAAC;IAC1C,KAAK,MAAM,MAAM,IAAI,SAAS,CAAC,OAAO,EAAE,CAAC;QACxC,IAAI,MAAM,CAAC,SAAS,IAAI,YAAY,EAAE,CAAC;YACtC,QAAQ,GAAG,MAAM,CAAC;YAClB,SAAS;QACV,CAAC;QACD,IAAI,MAAM,CAAC,WAAW,GAAG,YAAY,EAAE,CAAC;YACvC,IAAI,GAAG,MAAM,CAAC;YACd,MAAM;QACP,CAAC;IACF,CAAC;IACD,IAAI,QAAQ,IAAI,IAAI,EAAE,CAAC;QACtB,OAAO;YACN,KAAK,EAAE,IAAI,CAAC,GAAG,CAAC,QAAQ,CAAC,YAAY,EAAE,IAAI,CAAC,YAAY,CAAC;YACzD,GAAG,EAAE,IAAI,CAAC,GAAG,CAAC,QAAQ,CAAC,UAAU,EAAE,IAAI,CAAC,UAAU,CAAC;SACnD,CAAC;IACH,CAAC;IACD,OAAO,IAAI,CAAC;AACb,CAAC,CAAC","sourcesContent":["import { extractSpokenText } from \"./ssml/spoken-text.js\";\nimport {\n\tcreateSpeechAlignmentTokenPattern,\n\tnormalizeTextForSpeech,\n} from \"./text-processing.js\";\n\nexport type CatalogChunkPlaybackMode =\n\t| \"exact-word\"\n\t| \"anchor-span\"\n\t| \"region-fallback\";\n\nexport type BoundaryOffsetMode =\n\t| \"plain-spoken-text\"\n\t| \"raw-ssml\"\n\t| \"unsupported\";\n\nexport interface CatalogTextToken {\n\ttext: string;\n\tnormalized: string;\n\tstart: number;\n\tend: number;\n}\n\nexport interface CatalogSpanAnchor {\n\tspokenStart: number;\n\tspokenEnd: number;\n\tvisibleStart: number;\n\tvisibleEnd: number;\n\tscore: number;\n}\n\nexport interface CatalogSpanAlignment {\n\tspeechText: string;\n\tspokenText: string;\n\tvisibleText: string;\n\tplaybackMode: CatalogChunkPlaybackMode;\n\tboundaryOffsetMode: BoundaryOffsetMode;\n\tanchors: CatalogSpanAnchor[];\n\tconfidence: number;\n\tspokenTokens: CatalogTextToken[];\n\tvisibleTokens: CatalogTextToken[];\n\trawToSpokenOffsetMap: Map<number, number>;\n}\n\ninterface MatchCandidate {\n\tspokenStartToken: number;\n\tspokenEndToken: number;\n\tvisibleToken: number;\n\tscore: number;\n}\n\n// Shared Unicode-aware tokenizer (see `createSpeechAlignmentTokenPattern`), so\n// accented Latin and non-Latin words tokenize the same way here as in the math\n// speech tokenizer. LIMITATION (i18n): the phrase table below is still English,\n// so multi-word operator phrases (\"plus or minus\", \"divided by\") only align for\n// English speech; other locales degrade to coarse region highlighting. A fresh\n// pattern per call because of the global flag's mutable `lastIndex`.\nconst TOKEN_PATTERN = createSpeechAlignmentTokenPattern();\n\nconst NUMERIC_WORDS = new Map<string, string>([\n\t[\"zero\", \"0\"],\n\t[\"one\", \"1\"],\n\t[\"two\", \"2\"],\n\t[\"three\", \"3\"],\n\t[\"four\", \"4\"],\n\t[\"five\", \"5\"],\n\t[\"six\", \"6\"],\n\t[\"seven\", \"7\"],\n\t[\"eight\", \"8\"],\n\t[\"nine\", \"9\"],\n\t[\"ten\", \"10\"],\n]);\n\nconst SPOKEN_PHRASE_TARGETS: Array<{\n\tphrase: string[];\n\tvisible: string;\n\tscore: number;\n}> = [\n\t{ phrase: [\"plus\", \"or\", \"minus\"], visible: \"±\", score: 8 },\n\t{ phrase: [\"square\", \"root\"], visible: \"√\", score: 8 },\n\t{ phrase: [\"divided\", \"by\"], visible: \"/\", score: 7 },\n\t{ phrase: [\"all\", \"over\"], visible: \"/\", score: 7 },\n\t{ phrase: [\"equals\"], visible: \"=\", score: 7 },\n\t{ phrase: [\"equal\"], visible: \"=\", score: 7 },\n\t{ phrase: [\"times\"], visible: \"\\u2062\", score: 7 },\n\t{ phrase: [\"multiplied\", \"by\"], visible: \"\\u2062\", score: 7 },\n\t{ phrase: [\"over\"], visible: \"/\", score: 6 },\n\t{ phrase: [\"squared\"], visible: \"²\", score: 6 },\n\t{ phrase: [\"squared\"], visible: \"2\", score: 4 },\n\t{ phrase: [\"negative\"], visible: \"-\", score: 5 },\n\t{ phrase: [\"minus\"], visible: \"-\", score: 5 },\n\t{ phrase: [\"plus\"], visible: \"+\", score: 5 },\n];\n\nconst tokenize = (text: string): CatalogTextToken[] => {\n\tconst tokens: CatalogTextToken[] = [];\n\tfor (const match of text.matchAll(TOKEN_PATTERN)) {\n\t\tconst value = match[0];\n\t\tconst start = match.index ?? 0;\n\t\ttokens.push({\n\t\t\ttext: value,\n\t\t\tnormalized: value.toLowerCase(),\n\t\t\tstart,\n\t\t\tend: start + value.length,\n\t\t});\n\t}\n\treturn tokens;\n};\n\nconst normalizedTokenValue = (token: CatalogTextToken): string =>\n\tNUMERIC_WORDS.get(token.normalized) || token.normalized;\n\nconst createCandidates = (\n\tspokenTokens: CatalogTextToken[],\n\tvisibleTokens: CatalogTextToken[],\n): MatchCandidate[] => {\n\tconst candidates: MatchCandidate[] = [];\n\tfor (let spokenIndex = 0; spokenIndex < spokenTokens.length; spokenIndex++) {\n\t\tfor (let visibleIndex = 0; visibleIndex < visibleTokens.length; visibleIndex++) {\n\t\t\tconst spokenValue = normalizedTokenValue(spokenTokens[spokenIndex]);\n\t\t\tconst visibleValue = normalizedTokenValue(visibleTokens[visibleIndex]);\n\t\t\tif (spokenValue === visibleValue) {\n\t\t\t\tcandidates.push({\n\t\t\t\t\tspokenStartToken: spokenIndex,\n\t\t\t\t\tspokenEndToken: spokenIndex + 1,\n\t\t\t\t\tvisibleToken: visibleIndex,\n\t\t\t\t\tscore: 10,\n\t\t\t\t});\n\t\t\t}\n\t\t}\n\n\t\tfor (const phraseTarget of SPOKEN_PHRASE_TARGETS) {\n\t\t\tconst phraseTokens = spokenTokens.slice(\n\t\t\t\tspokenIndex,\n\t\t\t\tspokenIndex + phraseTarget.phrase.length,\n\t\t\t);\n\t\t\tif (phraseTokens.length !== phraseTarget.phrase.length) continue;\n\t\t\tif (\n\t\t\t\t!phraseTokens.every(\n\t\t\t\t\t(token, tokenIndex) =>\n\t\t\t\t\t\ttoken.normalized === phraseTarget.phrase[tokenIndex],\n\t\t\t\t)\n\t\t\t) {\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tfor (\n\t\t\t\tlet visibleIndex = 0;\n\t\t\t\tvisibleIndex < visibleTokens.length;\n\t\t\t\tvisibleIndex++\n\t\t\t) {\n\t\t\t\tif (visibleTokens[visibleIndex].normalized !== phraseTarget.visible) {\n\t\t\t\t\tcontinue;\n\t\t\t\t}\n\t\t\t\tcandidates.push({\n\t\t\t\t\tspokenStartToken: spokenIndex,\n\t\t\t\t\tspokenEndToken: spokenIndex + phraseTarget.phrase.length,\n\t\t\t\t\tvisibleToken: visibleIndex,\n\t\t\t\t\tscore: phraseTarget.score,\n\t\t\t\t});\n\t\t\t}\n\t\t}\n\t}\n\treturn candidates;\n};\n\nconst computeAnchors = (\n\tspokenTokens: CatalogTextToken[],\n\tvisibleTokens: CatalogTextToken[],\n): CatalogSpanAnchor[] => {\n\tconst candidates = createCandidates(spokenTokens, visibleTokens).sort(\n\t\t(left, right) =>\n\t\t\tspokenTokens[left.spokenStartToken].start -\n\t\t\t\tspokenTokens[right.spokenStartToken].start ||\n\t\t\tvisibleTokens[left.visibleToken].start - visibleTokens[right.visibleToken].start,\n\t);\n\tconst bestScore = new Array(candidates.length).fill(0) as number[];\n\tconst previous = new Array(candidates.length).fill(-1) as number[];\n\n\tfor (let i = 0; i < candidates.length; i++) {\n\t\tconst candidate = candidates[i];\n\t\tbestScore[i] = candidate.score;\n\t\tfor (let j = 0; j < i; j++) {\n\t\t\tconst prior = candidates[j];\n\t\t\tif (\n\t\t\t\tprior.spokenEndToken <= candidate.spokenStartToken &&\n\t\t\t\tprior.visibleToken < candidate.visibleToken\n\t\t\t) {\n\t\t\t\tconst spokenGap = candidate.spokenStartToken - prior.spokenEndToken;\n\t\t\t\tconst visibleGap = candidate.visibleToken - prior.visibleToken - 1;\n\t\t\t\tconst gapPenalty = Math.max(0, Math.abs(spokenGap - visibleGap) - 2);\n\t\t\t\tconst score = bestScore[j] + candidate.score - gapPenalty;\n\t\t\t\tif (score > bestScore[i]) {\n\t\t\t\t\tbestScore[i] = score;\n\t\t\t\t\tprevious[i] = j;\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\n\tlet bestIndex = -1;\n\tfor (let i = 0; i < bestScore.length; i++) {\n\t\tif (bestIndex < 0 || bestScore[i] > bestScore[bestIndex]) {\n\t\t\tbestIndex = i;\n\t\t}\n\t}\n\tif (bestIndex < 0) return [];\n\n\tconst chosen: MatchCandidate[] = [];\n\tlet current = bestIndex;\n\twhile (current >= 0) {\n\t\tchosen.unshift(candidates[current]);\n\t\tcurrent = previous[current];\n\t}\n\n\treturn chosen.map((candidate) => {\n\t\tconst firstSpoken = spokenTokens[candidate.spokenStartToken];\n\t\tconst lastSpoken = spokenTokens[candidate.spokenEndToken - 1];\n\t\tconst visible = visibleTokens[candidate.visibleToken];\n\t\treturn {\n\t\t\tspokenStart: firstSpoken.start,\n\t\t\tspokenEnd: lastSpoken.end,\n\t\t\tvisibleStart: visible.start,\n\t\t\tvisibleEnd: visible.end,\n\t\t\tscore: candidate.score,\n\t\t};\n\t});\n};\n\nconst scoreConfidence = (\n\tanchors: CatalogSpanAnchor[],\n\tspokenTokens: CatalogTextToken[],\n\tvisibleTokens: CatalogTextToken[],\n): number => {\n\tif (anchors.length === 0) return 0;\n\tconst tokenCoverage =\n\t\tanchors.length / Math.max(1, Math.min(spokenTokens.length, visibleTokens.length));\n\tconst averageScore =\n\t\tanchors.reduce((total, anchor) => total + anchor.score, 0) /\n\t\tanchors.length /\n\t\t10;\n\treturn Math.min(1, tokenCoverage * averageScore);\n};\n\nexport const createCatalogSpanAlignment = (args: {\n\tspeechText: string;\n\tvisibleText: string;\n}): CatalogSpanAlignment => {\n\tconst extracted = extractSpokenText(args.speechText);\n\tconst visibleText = normalizeTextForSpeech(args.visibleText);\n\tconst spokenText = extracted.spokenText;\n\tconst spokenTokens = tokenize(spokenText);\n\tconst visibleTokens = tokenize(visibleText);\n\tconst anchors = extracted.unsupportedSemantic\n\t\t? []\n\t\t: computeAnchors(spokenTokens, visibleTokens);\n\tconst confidence = scoreConfidence(anchors, spokenTokens, visibleTokens);\n\tconst isExact = spokenText === visibleText && spokenText.length > 0;\n\tconst hasShortSingleExactAnchor =\n\t\tanchors.length === 1 &&\n\t\tanchors[0].score === 10 &&\n\t\tspokenTokens.length <= 2 &&\n\t\tvisibleTokens.length <= 2;\n\tconst playbackMode: CatalogChunkPlaybackMode = extracted.unsupportedSemantic\n\t\t? \"region-fallback\"\n\t\t: isExact\n\t\t\t? \"exact-word\"\n\t\t\t: anchors.length >= 2 || hasShortSingleExactAnchor\n\t\t\t\t? confidence >= 0.12\n\t\t\t\t\t? \"anchor-span\"\n\t\t\t\t\t: \"region-fallback\"\n\t\t\t\t: \"region-fallback\";\n\tconst boundaryOffsetMode: BoundaryOffsetMode = extracted.unsupportedSemantic\n\t\t? \"unsupported\"\n\t\t: extracted.hasMarkup\n\t\t\t? \"raw-ssml\"\n\t\t\t: \"plain-spoken-text\";\n\n\treturn {\n\t\tspeechText: args.speechText,\n\t\tspokenText,\n\t\tvisibleText,\n\t\tplaybackMode,\n\t\tboundaryOffsetMode,\n\t\tanchors,\n\t\tconfidence,\n\t\tspokenTokens,\n\t\tvisibleTokens,\n\t\trawToSpokenOffsetMap: extracted.rawToSpokenOffsetMap,\n\t};\n};\n\nconst findTokenAtOffset = (\n\ttokens: CatalogTextToken[],\n\toffset: number,\n): CatalogTextToken | null =>\n\ttokens.find((token) => token.start <= offset && offset < token.end) || null;\n\nconst normalizeBoundaryWord = (word?: string): string | null => {\n\tif (!word) return null;\n\tif (/^<[^>]+>$/.test(word.trim())) return null;\n\tconst token = tokenize(word)[0];\n\tif (!token) return null;\n\treturn NUMERIC_WORDS.get(token.normalized) || token.normalized;\n};\n\nconst candidateMatchesBoundaryWord = (\n\talignment: CatalogSpanAlignment,\n\tcandidate: { start: number; length: number } | null,\n\tboundaryWord?: string,\n): boolean => {\n\tconst normalizedBoundary = normalizeBoundaryWord(boundaryWord);\n\tif (!candidate || !normalizedBoundary) return false;\n\tconst token = findTokenAtOffset(alignment.spokenTokens, candidate.start);\n\tif (!token) return false;\n\treturn normalizedTokenValue(token) === normalizedBoundary;\n};\n\nconst mapRawOffsetToSpokenOffset = (\n\talignment: CatalogSpanAlignment,\n\tposition: number,\n): number | null => {\n\tconst direct = alignment.rawToSpokenOffsetMap.get(position);\n\tif (direct !== undefined) return direct;\n\tfor (let offset = position; offset < alignment.speechText.length; offset++) {\n\t\tconst mapped = alignment.rawToSpokenOffsetMap.get(offset);\n\t\tif (mapped !== undefined) return mapped;\n\t\tif (alignment.speechText[offset] === \">\") break;\n\t}\n\treturn null;\n};\n\nconst resolveRawBoundaryOffset = (\n\talignment: CatalogSpanAlignment,\n\tposition: number,\n\tsafeLength: number,\n): { start: number; length: number } | null => {\n\tconst mappedStart = mapRawOffsetToSpokenOffset(alignment, position);\n\tif (mappedStart === null) return null;\n\tconst mappedEnd = mapRawOffsetToSpokenOffset(\n\t\talignment,\n\t\tMath.max(position, position + safeLength - 1),\n\t);\n\tconst token = findTokenAtOffset(alignment.spokenTokens, mappedStart);\n\tconst mappedLength =\n\t\tmappedEnd !== null && mappedEnd >= mappedStart\n\t\t\t? mappedEnd - mappedStart + 1\n\t\t\t: token\n\t\t\t\t? token.end - mappedStart\n\t\t\t\t: safeLength;\n\treturn { start: mappedStart, length: Math.max(1, mappedLength) };\n};\n\nconst resolvePlainBoundaryOffset = (\n\talignment: CatalogSpanAlignment,\n\tposition: number,\n\tsafeLength: number,\n): { start: number; length: number } | null => {\n\tif (position < 0 || position >= alignment.spokenText.length) return null;\n\tconst token = findTokenAtOffset(alignment.spokenTokens, position);\n\tif (!token) return null;\n\treturn { start: position, length: Math.min(safeLength, token.end - position) };\n};\n\nexport const resolveSpokenBoundaryOffset = (\n\talignment: CatalogSpanAlignment,\n\tposition: number,\n\tlength = 1,\n\tboundaryWord?: string,\n): { start: number; length: number } | null => {\n\tif (\n\t\talignment.boundaryOffsetMode === \"unsupported\" ||\n\t\t!Number.isFinite(position)\n\t) {\n\t\treturn null;\n\t}\n\tconst safeLength = Math.max(1, Number.isFinite(length) ? length : 1);\n\tif (boundaryWord && /^<[^>]+>$/.test(boundaryWord.trim())) return null;\n\tif (alignment.boundaryOffsetMode === \"raw-ssml\") {\n\t\tconst rawCandidate = resolveRawBoundaryOffset(\n\t\t\talignment,\n\t\t\tposition,\n\t\t\tsafeLength,\n\t\t);\n\t\tconst plainCandidate = resolvePlainBoundaryOffset(\n\t\t\talignment,\n\t\t\tposition,\n\t\t\tsafeLength,\n\t\t);\n\t\tif (boundaryWord) {\n\t\t\tif (\n\t\t\t\tcandidateMatchesBoundaryWord(alignment, rawCandidate, boundaryWord)\n\t\t\t) {\n\t\t\t\treturn rawCandidate;\n\t\t\t}\n\t\t\tif (\n\t\t\t\tcandidateMatchesBoundaryWord(alignment, plainCandidate, boundaryWord)\n\t\t\t) {\n\t\t\t\treturn plainCandidate;\n\t\t\t}\n\t\t\tif (normalizeBoundaryWord(boundaryWord)) return null;\n\t\t}\n\t\treturn rawCandidate || plainCandidate;\n\t}\n\treturn (\n\t\tresolvePlainBoundaryOffset(alignment, position, safeLength) ||\n\t\tresolveRawBoundaryOffset(alignment, position, safeLength)\n\t);\n};\n\nexport const resolveVisibleSpanForBoundary = (\n\talignment: CatalogSpanAlignment,\n\tspokenOffset: number,\n): { start: number; end: number } | null => {\n\tif (alignment.playbackMode === \"region-fallback\") return null;\n\tconst exactAnchor = alignment.anchors.find(\n\t\t(anchor) => anchor.spokenStart <= spokenOffset && spokenOffset < anchor.spokenEnd,\n\t);\n\tif (exactAnchor) {\n\t\treturn { start: exactAnchor.visibleStart, end: exactAnchor.visibleEnd };\n\t}\n\n\tlet previous: CatalogSpanAnchor | null = null;\n\tlet next: CatalogSpanAnchor | null = null;\n\tfor (const anchor of alignment.anchors) {\n\t\tif (anchor.spokenEnd <= spokenOffset) {\n\t\t\tprevious = anchor;\n\t\t\tcontinue;\n\t\t}\n\t\tif (anchor.spokenStart > spokenOffset) {\n\t\t\tnext = anchor;\n\t\t\tbreak;\n\t\t}\n\t}\n\tif (previous && next) {\n\t\treturn {\n\t\t\tstart: Math.min(previous.visibleStart, next.visibleStart),\n\t\t\tend: Math.max(previous.visibleEnd, next.visibleEnd),\n\t\t};\n\t}\n\treturn null;\n};\n"]}

package/dist/services/tts/generated-speech/assemble-plan.d.ts

@@ -0,0 +1,33 @@
+import type { MathAwareSpeechChunk } from "../math-aware-text-processing.js";
+import { type ResolvedMathSpeech, type SREMathSpeechOptions } from "../math-speech.js";
+import type { AssembledSpeech } from "./types.js";
+type MathChunk = Extract<MathAwareSpeechChunk, {
+    type: "math";
+}>;
+/**
+ * Resolves spoken text for a single math chunk. Injectable so the runtime can
+ * supply a memoized resolver (deterministic caching) and tests can stub SRE.
+ * Mirrors `resolveMathSpeechFromChunks([chunk], options)`.
+ */
+export type MathSpeechResolver = (chunk: MathChunk, options: {
+    language?: string;
+    produceSsml?: boolean;
+    mathSpeech?: SREMathSpeechOptions;
+}) => Promise<ResolvedMathSpeech>;
+/**
+ * Assemble an anchor-free speech plan from already-collected math-aware chunks.
+ *
+ * PURE: no DOM walking, no app-service imports. Preserves the spoken-text and
+ * visible-span bookkeeping from the inline generated path, so the runtime output
+ * is stable once serialized to plain text.
+ */
+export declare const assembleGeneratedSpeech: (args: {
+    chunks: ReadonlyArray<MathAwareSpeechChunk>;
+    visibleText: string;
+    language?: string;
+    mathSpeech?: SREMathSpeechOptions;
+    /** Request SRE SSML per math equation (used by the SSML playback format). */
+    produceSsml?: boolean;
+    resolveMathSpeech?: MathSpeechResolver;
+}) => Promise<AssembledSpeech>;
+export {};

package/dist/services/tts/generated-speech/assemble-plan.js

@@ -0,0 +1,111 @@
+import { resolveMathSpeechFromChunks, } from "../math-speech.js";
+import { normalizeTextForSpeech } from "../text-processing.js";
+import { segmentSentences as segmentTextToSentences } from "../text-segmentation.js";
+const defaultMathSpeechResolver = (chunk, options) => resolveMathSpeechFromChunks([chunk], options);
+const normalizeLocale = (language) => (language || "en").split("-")[0].toLowerCase() || "en";
+const PRESERVE_PROSE_CHUNK_TAGS = new Set([
+    "H1",
+    "H2",
+    "H3",
+    "H4",
+    "H5",
+    "H6",
+    "LI",
+    "LABEL",
+    "BUTTON",
+]);
+const PRESERVE_PROSE_CHUNK_ROLES = new Set([
+    "heading",
+    "listitem",
+    "option",
+    "radio",
+]);
+const shouldPreserveProseChunk = (chunk) => {
+    if (chunk.type !== "text" || !chunk.sourceElement)
+        return false;
+    const tagName = chunk.sourceElement.tagName.toUpperCase();
+    const role = (chunk.sourceElement.getAttribute("role") || "").toLowerCase();
+    return (PRESERVE_PROSE_CHUNK_TAGS.has(tagName) ||
+        PRESERVE_PROSE_CHUNK_ROLES.has(role));
+};
+/**
+ * Assemble an anchor-free speech plan from already-collected math-aware chunks.
+ *
+ * PURE: no DOM walking, no app-service imports. Preserves the spoken-text and
+ * visible-span bookkeeping from the inline generated path, so the runtime output
+ * is stable once serialized to plain text.
+ */
+export const assembleGeneratedSpeech = async (args) => {
+    const resolveMathSpeech = args.resolveMathSpeech ?? defaultMathSpeechResolver;
+    const { visibleText } = args;
+    const segments = [];
+    // Cursor-based span tracking:
+    // locate each chunk's normalized text within the aggregate visible text,
+    // advancing a cursor so math fallbacks reserve their span even though only
+    // prose spans are sliced for highlighting.
+    let visibleCursor = 0;
+    const nextVisibleSpan = (text) => {
+        const normalized = normalizeTextForSpeech(text);
+        const found = visibleText.indexOf(normalized, visibleCursor);
+        const start = found >= 0 ? found : visibleCursor;
+        const end = start + normalized.length;
+        visibleCursor = end;
+        return { start, end };
+    };
+    const speechParts = [];
+    for (let index = 0; index < args.chunks.length; index++) {
+        const chunk = args.chunks[index];
+        if (chunk.type === "text") {
+            const proseSegments = shouldPreserveProseChunk(chunk)
+                ? [{ text: chunk.text, offset: 0 }]
+                : segmentTextToSentences(chunk.text, {
+                    locale: args.language,
+                });
+            for (const prose of proseSegments) {
+                const text = normalizeTextForSpeech(prose.text);
+                if (!text)
+                    continue;
+                const visibleSpan = nextVisibleSpan(text);
+                segments.push({
+                    sourceChunkIndex: index,
+                    segment: {
+                        kind: "prose",
+                        spokenText: text,
+                        visibleText: text,
+                        visibleSpan,
+                    },
+                });
+                speechParts.push(text);
+            }
+            continue;
+        }
+        const visibleSpan = nextVisibleSpan(chunk.fallbackText);
+        const generated = await resolveMathSpeech(chunk, {
+            language: args.language,
+            produceSsml: args.produceSsml,
+            mathSpeech: args.mathSpeech,
+        });
+        const usedFallback = !generated.speechText;
+        const spokenText = generated.speechText || chunk.fallbackText;
+        segments.push({
+            sourceChunkIndex: index,
+            segment: {
+                kind: "math",
+                spokenText,
+                mathml: chunk.mathml,
+                fallbackText: chunk.fallbackText,
+                visibleSpan,
+                usedFallback,
+                ssml: generated.ssml,
+            },
+        });
+        speechParts.push(spokenText);
+    }
+    return {
+        segments,
+        locale: normalizeLocale(args.language),
+        visibleText,
+        plainSpeechText: normalizeTextForSpeech(speechParts.join(" ")),
+    };
+};
+//# sourceMappingURL=assemble-plan.js.map

package/dist/services/tts/generated-speech/assemble-plan.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"assemble-plan.js","sourceRoot":"","sources":["../../../../src/services/tts/generated-speech/assemble-plan.ts"],"names":[],"mappings":"AACA,OAAO,EACN,2BAA2B,GAG3B,MAAM,mBAAmB,CAAC;AAC3B,OAAO,EAAE,sBAAsB,EAAE,MAAM,uBAAuB,CAAC;AAC/D,OAAO,EAAE,gBAAgB,IAAI,sBAAsB,EAAE,MAAM,yBAAyB,CAAC;AAmBrF,MAAM,yBAAyB,GAAuB,CAAC,KAAK,EAAE,OAAO,EAAE,EAAE,CACxE,2BAA2B,CAAC,CAAC,KAAK,CAAC,EAAE,OAAO,CAAC,CAAC;AAE/C,MAAM,eAAe,GAAG,CAAC,QAAiB,EAAU,EAAE,CACrD,CAAC,QAAQ,IAAI,IAAI,CAAC,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,WAAW,EAAE,IAAI,IAAI,CAAC;AAExD,MAAM,yBAAyB,GAAG,IAAI,GAAG,CAAC;IACzC,IAAI;IACJ,IAAI;IACJ,IAAI;IACJ,IAAI;IACJ,IAAI;IACJ,IAAI;IACJ,IAAI;IACJ,OAAO;IACP,QAAQ;CACR,CAAC,CAAC;AAEH,MAAM,0BAA0B,GAAG,IAAI,GAAG,CAAC;IAC1C,SAAS;IACT,UAAU;IACV,QAAQ;IACR,OAAO;CACP,CAAC,CAAC;AAEH,MAAM,wBAAwB,GAAG,CAChC,KAA2B,EACjB,EAAE;IACZ,IAAI,KAAK,CAAC,IAAI,KAAK,MAAM,IAAI,CAAC,KAAK,CAAC,aAAa;QAAE,OAAO,KAAK,CAAC;IAChE,MAAM,OAAO,GAAG,KAAK,CAAC,aAAa,CAAC,OAAO,CAAC,WAAW,EAAE,CAAC;IAC1D,MAAM,IAAI,GAAG,CAAC,KAAK,CAAC,aAAa,CAAC,YAAY,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC,CAAC,WAAW,EAAE,CAAC;IAC5E,OAAO,CACN,yBAAyB,CAAC,GAAG,CAAC,OAAO,CAAC;QACtC,0BAA0B,CAAC,GAAG,CAAC,IAAI,CAAC,CACpC,CAAC;AACH,CAAC,CAAC;AAEF;;;;;;GAMG;AACH,MAAM,CAAC,MAAM,uBAAuB,GAAG,KAAK,EAAE,IAQ7C,EAA4B,EAAE;IAC9B,MAAM,iBAAiB,GAAG,IAAI,CAAC,iBAAiB,IAAI,yBAAyB,CAAC;IAC9E,MAAM,EAAE,WAAW,EAAE,GAAG,IAAI,CAAC;IAC7B,MAAM,QAAQ,GAAuB,EAAE,CAAC;IAExC,8BAA8B;IAC9B,yEAAyE;IACzE,2EAA2E;IAC3E,2CAA2C;IAC3C,IAAI,aAAa,GAAG,CAAC,CAAC;IACtB,MAAM,eAAe,GAAG,CAAC,IAAY,EAAkC,EAAE;QACxE,MAAM,UAAU,GAAG,sBAAsB,CAAC,IAAI,CAAC,CAAC;QAChD,MAAM,KAAK,GAAG,WAAW,CAAC,OAAO,CAAC,UAAU,EAAE,aAAa,CAAC,CAAC;QAC7D,MAAM,KAAK,GAAG,KAAK,IAAI,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,aAAa,CAAC;QACjD,MAAM,GAAG,GAAG,KAAK,GAAG,UAAU,CAAC,MAAM,CAAC;QACtC,aAAa,GAAG,GAAG,CAAC;QACpB,OAAO,EAAE,KAAK,EAAE,GAAG,EAAE,CAAC;IACvB,CAAC,CAAC;IAEF,MAAM,WAAW,GAAa,EAAE,CAAC;IAEjC,KAAK,IAAI,KAAK,GAAG,CAAC,EAAE,KAAK,GAAG,IAAI,CAAC,MAAM,CAAC,MAAM,EAAE,KAAK,EAAE,EAAE,CAAC;QACzD,MAAM,KAAK,GAAG,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;QACjC,IAAI,KAAK,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;YAC3B,MAAM,aAAa,GAAG,wBAAwB,CAAC,KAAK,CAAC;gBACpD,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,KAAK,CAAC,IAAI,EAAE,MAAM,EAAE,CAAC,EAAE,CAAC;gBACnC,CAAC,CAAC,sBAAsB,CAAC,KAAK,CAAC,IAAI,EAAE;oBACnC,MAAM,EAAE,IAAI,CAAC,QAAQ;iBACrB,CAAC,CAAC;YACL,KAAK,MAAM,KAAK,IAAI,aAAa,EAAE,CAAC;gBACnC,MAAM,IAAI,GAAG,sBAAsB,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;gBAChD,IAAI,CAAC,IAAI;oBAAE,SAAS;gBACpB,MAAM,WAAW,GAAG,eAAe,CAAC,IAAI,CAAC,CAAC;gBAC1C,QAAQ,CAAC,IAAI,CAAC;oBACb,gBAAgB,EAAE,KAAK;oBACvB,OAAO,EAAE;wBACR,IAAI,EAAE,OAAO;wBACb,UAAU,EAAE,IAAI;wBAChB,WAAW,EAAE,IAAI;wBACjB,WAAW;qBACX;iBACD,CAAC,CAAC;gBACH,WAAW,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YACxB,CAAC;YACD,SAAS;QACV,CAAC;QACD,MAAM,WAAW,GAAG,eAAe,CAAC,KAAK,CAAC,YAAY,CAAC,CAAC;QACxD,MAAM,SAAS,GAAG,MAAM,iBAAiB,CAAC,KAAK,EAAE;YAChD,QAAQ,EAAE,IAAI,CAAC,QAAQ;YACvB,WAAW,EAAE,IAAI,CAAC,WAAW;YAC7B,UAAU,EAAE,IAAI,CAAC,UAAU;SAC3B,CAAC,CAAC;QACH,MAAM,YAAY,GAAG,CAAC,SAAS,CAAC,UAAU,CAAC;QAC3C,MAAM,UAAU,GAAG,SAAS,CAAC,UAAU,IAAI,KAAK,CAAC,YAAY,CAAC;QAC9D,QAAQ,CAAC,IAAI,CAAC;YACb,gBAAgB,EAAE,KAAK;YACvB,OAAO,EAAE;gBACR,IAAI,EAAE,MAAM;gBACZ,UAAU;gBACV,MAAM,EAAE,KAAK,CAAC,MAAM;gBACpB,YAAY,EAAE,KAAK,CAAC,YAAY;gBAChC,WAAW;gBACX,YAAY;gBACZ,IAAI,EAAE,SAAS,CAAC,IAAI;aACpB;SACD,CAAC,CAAC;QACH,WAAW,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC;IAC9B,CAAC;IAED,OAAO;QACN,QAAQ;QACR,MAAM,EAAE,eAAe,CAAC,IAAI,CAAC,QAAQ,CAAC;QACtC,WAAW;QACX,eAAe,EAAE,sBAAsB,CAAC,WAAW,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;KAC9D,CAAC;AACH,CAAC,CAAC","sourcesContent":["import type { MathAwareSpeechChunk } from \"../math-aware-text-processing.js\";\nimport {\n\tresolveMathSpeechFromChunks,\n\ttype ResolvedMathSpeech,\n\ttype SREMathSpeechOptions,\n} from \"../math-speech.js\";\nimport { normalizeTextForSpeech } from \"../text-processing.js\";\nimport { segmentSentences as segmentTextToSentences } from \"../text-segmentation.js\";\nimport type { AssembledSegment, AssembledSpeech } from \"./types.js\";\n\ntype MathChunk = Extract<MathAwareSpeechChunk, { type: \"math\" }>;\n\n/**\n * Resolves spoken text for a single math chunk. Injectable so the runtime can\n * supply a memoized resolver (deterministic caching) and tests can stub SRE.\n * Mirrors `resolveMathSpeechFromChunks([chunk], options)`.\n */\nexport type MathSpeechResolver = (\n\tchunk: MathChunk,\n\toptions: {\n\t\tlanguage?: string;\n\t\tproduceSsml?: boolean;\n\t\tmathSpeech?: SREMathSpeechOptions;\n\t},\n) => Promise<ResolvedMathSpeech>;\n\nconst defaultMathSpeechResolver: MathSpeechResolver = (chunk, options) =>\n\tresolveMathSpeechFromChunks([chunk], options);\n\nconst normalizeLocale = (language?: string): string =>\n\t(language || \"en\").split(\"-\")[0].toLowerCase() || \"en\";\n\nconst PRESERVE_PROSE_CHUNK_TAGS = new Set([\n\t\"H1\",\n\t\"H2\",\n\t\"H3\",\n\t\"H4\",\n\t\"H5\",\n\t\"H6\",\n\t\"LI\",\n\t\"LABEL\",\n\t\"BUTTON\",\n]);\n\nconst PRESERVE_PROSE_CHUNK_ROLES = new Set([\n\t\"heading\",\n\t\"listitem\",\n\t\"option\",\n\t\"radio\",\n]);\n\nconst shouldPreserveProseChunk = (\n\tchunk: MathAwareSpeechChunk,\n): boolean => {\n\tif (chunk.type !== \"text\" || !chunk.sourceElement) return false;\n\tconst tagName = chunk.sourceElement.tagName.toUpperCase();\n\tconst role = (chunk.sourceElement.getAttribute(\"role\") || \"\").toLowerCase();\n\treturn (\n\t\tPRESERVE_PROSE_CHUNK_TAGS.has(tagName) ||\n\t\tPRESERVE_PROSE_CHUNK_ROLES.has(role)\n\t);\n};\n\n/**\n * Assemble an anchor-free speech plan from already-collected math-aware chunks.\n *\n * PURE: no DOM walking, no app-service imports. Preserves the spoken-text and\n * visible-span bookkeeping from the inline generated path, so the runtime output\n * is stable once serialized to plain text.\n */\nexport const assembleGeneratedSpeech = async (args: {\n\tchunks: ReadonlyArray<MathAwareSpeechChunk>;\n\tvisibleText: string;\n\tlanguage?: string;\n\tmathSpeech?: SREMathSpeechOptions;\n\t/** Request SRE SSML per math equation (used by the SSML playback format). */\n\tproduceSsml?: boolean;\n\tresolveMathSpeech?: MathSpeechResolver;\n}): Promise<AssembledSpeech> => {\n\tconst resolveMathSpeech = args.resolveMathSpeech ?? defaultMathSpeechResolver;\n\tconst { visibleText } = args;\n\tconst segments: AssembledSegment[] = [];\n\n\t// Cursor-based span tracking:\n\t// locate each chunk's normalized text within the aggregate visible text,\n\t// advancing a cursor so math fallbacks reserve their span even though only\n\t// prose spans are sliced for highlighting.\n\tlet visibleCursor = 0;\n\tconst nextVisibleSpan = (text: string): { start: number; end: number } => {\n\t\tconst normalized = normalizeTextForSpeech(text);\n\t\tconst found = visibleText.indexOf(normalized, visibleCursor);\n\t\tconst start = found >= 0 ? found : visibleCursor;\n\t\tconst end = start + normalized.length;\n\t\tvisibleCursor = end;\n\t\treturn { start, end };\n\t};\n\n\tconst speechParts: string[] = [];\n\n\tfor (let index = 0; index < args.chunks.length; index++) {\n\t\tconst chunk = args.chunks[index];\n\t\tif (chunk.type === \"text\") {\n\t\t\tconst proseSegments = shouldPreserveProseChunk(chunk)\n\t\t\t\t? [{ text: chunk.text, offset: 0 }]\n\t\t\t\t: segmentTextToSentences(chunk.text, {\n\t\t\t\t\t\tlocale: args.language,\n\t\t\t\t\t});\n\t\t\tfor (const prose of proseSegments) {\n\t\t\t\tconst text = normalizeTextForSpeech(prose.text);\n\t\t\t\tif (!text) continue;\n\t\t\t\tconst visibleSpan = nextVisibleSpan(text);\n\t\t\t\tsegments.push({\n\t\t\t\t\tsourceChunkIndex: index,\n\t\t\t\t\tsegment: {\n\t\t\t\t\t\tkind: \"prose\",\n\t\t\t\t\t\tspokenText: text,\n\t\t\t\t\t\tvisibleText: text,\n\t\t\t\t\t\tvisibleSpan,\n\t\t\t\t\t},\n\t\t\t\t});\n\t\t\t\tspeechParts.push(text);\n\t\t\t}\n\t\t\tcontinue;\n\t\t}\n\t\tconst visibleSpan = nextVisibleSpan(chunk.fallbackText);\n\t\tconst generated = await resolveMathSpeech(chunk, {\n\t\t\tlanguage: args.language,\n\t\t\tproduceSsml: args.produceSsml,\n\t\t\tmathSpeech: args.mathSpeech,\n\t\t});\n\t\tconst usedFallback = !generated.speechText;\n\t\tconst spokenText = generated.speechText || chunk.fallbackText;\n\t\tsegments.push({\n\t\t\tsourceChunkIndex: index,\n\t\t\tsegment: {\n\t\t\t\tkind: \"math\",\n\t\t\t\tspokenText,\n\t\t\t\tmathml: chunk.mathml,\n\t\t\t\tfallbackText: chunk.fallbackText,\n\t\t\t\tvisibleSpan,\n\t\t\t\tusedFallback,\n\t\t\t\tssml: generated.ssml,\n\t\t\t},\n\t\t});\n\t\tspeechParts.push(spokenText);\n\t}\n\n\treturn {\n\t\tsegments,\n\t\tlocale: normalizeLocale(args.language),\n\t\tvisibleText,\n\t\tplainSpeechText: normalizeTextForSpeech(speechParts.join(\" \")),\n\t};\n};\n"]}

package/dist/services/tts/generated-speech/dom/build-dom-plan.d.ts

@@ -0,0 +1,25 @@
+import type { TextProcessingOptions } from "../../text-processing.js";
+import { type MathSpeechResolver } from "../assemble-plan.js";
+import type { SREMathSpeechOptions } from "../../math-speech.js";
+import type { GeneratedSpeechPlan } from "../types.js";
+import type { DomAnchor } from "./types.js";
+export interface BuildGeneratedSpeechResult {
+    plan: GeneratedSpeechPlan<DomAnchor>;
+    containsMathMarkup: boolean;
+    /** Aggregate visible text from the DOM walk (may be empty). */
+    visibleText: string;
+}
+/**
+ * DOM adapter: walk a live content root, assemble the (pure) speech plan, and
+ * bind each segment to its live-DOM anchor. This is the only generated-speech
+ * entry point that touches the DOM; assembly and serialization stay pure.
+ */
+export declare const buildGeneratedSpeechFromRoot: (args: {
+    contentRoot: Element;
+    language?: string;
+    mathSpeech?: SREMathSpeechOptions;
+    textProcessingOptions?: TextProcessingOptions;
+    /** Request SRE SSML per equation (for the SSML playback format). */
+    produceSsml?: boolean;
+    resolveMathSpeech?: MathSpeechResolver;
+}) => Promise<BuildGeneratedSpeechResult>;

package/dist/services/tts/generated-speech/dom/build-dom-plan.js

@@ -0,0 +1,66 @@
+import { resolveReadableRegion } from "../../highlight-pipeline/index.js";
+import { collectMathAwareTextAndMap } from "../../math-aware-text-processing.js";
+import { assembleGeneratedSpeech, } from "../assemble-plan.js";
+const sliceVisibleMap = (map, start, end) => {
+    const sliced = new Map();
+    for (let index = start; index < end; index++) {
+        const mapping = map.get(index);
+        if (mapping) {
+            sliced.set(index - start, mapping);
+        }
+    }
+    return sliced;
+};
+/**
+ * DOM adapter: walk a live content root, assemble the (pure) speech plan, and
+ * bind each segment to its live-DOM anchor. This is the only generated-speech
+ * entry point that touches the DOM; assembly and serialization stay pure.
+ */
+export const buildGeneratedSpeechFromRoot = async (args) => {
+    const extracted = collectMathAwareTextAndMap(args.contentRoot, args.textProcessingOptions);
+    const assembled = await assembleGeneratedSpeech({
+        chunks: extracted.chunks,
+        visibleText: extracted.visibleText,
+        language: args.language,
+        mathSpeech: args.mathSpeech,
+        produceSsml: args.produceSsml,
+        resolveMathSpeech: args.resolveMathSpeech,
+    });
+    const items = assembled.segments.map(({ segment, sourceChunkIndex }) => {
+        const sourceChunk = extracted.chunks[sourceChunkIndex];
+        if (segment.kind === "prose") {
+            const sourceElement = sourceChunk?.type === "text"
+                ? (sourceChunk.sourceElement ?? args.contentRoot)
+                : args.contentRoot;
+            return {
+                segment,
+                anchor: {
+                    sourceElement,
+                    regionElement: resolveReadableRegion(sourceElement, args.contentRoot),
+                    visibleMap: sliceVisibleMap(extracted.map, segment.visibleSpan.start, segment.visibleSpan.end),
+                },
+            };
+        }
+        const mathElement = sourceChunk?.type === "math"
+            ? (sourceChunk.sourceElement ?? null)
+            : null;
+        return {
+            segment,
+            anchor: {
+                sourceElement: mathElement,
+                regionElement: mathElement,
+            },
+        };
+    });
+    return {
+        plan: {
+            items,
+            locale: assembled.locale,
+            visibleText: assembled.visibleText,
+            plainSpeechText: assembled.plainSpeechText,
+        },
+        containsMathMarkup: extracted.containsMathMarkup,
+        visibleText: extracted.visibleText,
+    };
+};
+//# sourceMappingURL=build-dom-plan.js.map

package/dist/services/tts/generated-speech/dom/build-dom-plan.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"build-dom-plan.js","sourceRoot":"","sources":["../../../../../src/services/tts/generated-speech/dom/build-dom-plan.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,qBAAqB,EAAE,MAAM,mCAAmC,CAAC;AAC1E,OAAO,EAAE,0BAA0B,EAAE,MAAM,qCAAqC,CAAC;AAKjF,OAAO,EACN,uBAAuB,GAEvB,MAAM,qBAAqB,CAAC;AAK7B,MAAM,eAAe,GAAG,CACvB,GAAsB,EACtB,KAAa,EACb,GAAW,EACS,EAAE;IACtB,MAAM,MAAM,GAAsB,IAAI,GAAG,EAAE,CAAC;IAC5C,KAAK,IAAI,KAAK,GAAG,KAAK,EAAE,KAAK,GAAG,GAAG,EAAE,KAAK,EAAE,EAAE,CAAC;QAC9C,MAAM,OAAO,GAAG,GAAG,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;QAC/B,IAAI,OAAO,EAAE,CAAC;YACb,MAAM,CAAC,GAAG,CAAC,KAAK,GAAG,KAAK,EAAE,OAAO,CAAC,CAAC;QACpC,CAAC;IACF,CAAC;IACD,OAAO,MAAM,CAAC;AACf,CAAC,CAAC;AASF;;;;GAIG;AACH,MAAM,CAAC,MAAM,4BAA4B,GAAG,KAAK,EAAE,IAQlD,EAAuC,EAAE;IACzC,MAAM,SAAS,GAAG,0BAA0B,CAC3C,IAAI,CAAC,WAAW,EAChB,IAAI,CAAC,qBAAqB,CAC1B,CAAC;IACF,MAAM,SAAS,GAAG,MAAM,uBAAuB,CAAC;QAC/C,MAAM,EAAE,SAAS,CAAC,MAAM;QACxB,WAAW,EAAE,SAAS,CAAC,WAAW;QAClC,QAAQ,EAAE,IAAI,CAAC,QAAQ;QACvB,UAAU,EAAE,IAAI,CAAC,UAAU;QAC3B,WAAW,EAAE,IAAI,CAAC,WAAW;QAC7B,iBAAiB,EAAE,IAAI,CAAC,iBAAiB;KACzC,CAAC,CAAC;IAEH,MAAM,KAAK,GAA0B,SAAS,CAAC,QAAQ,CAAC,GAAG,CAC1D,CAAC,EAAE,OAAO,EAAE,gBAAgB,EAAE,EAAE,EAAE;QACjC,MAAM,WAAW,GAAG,SAAS,CAAC,MAAM,CAAC,gBAAgB,CAAC,CAAC;QACvD,IAAI,OAAO,CAAC,IAAI,KAAK,OAAO,EAAE,CAAC;YAC9B,MAAM,aAAa,GAClB,WAAW,EAAE,IAAI,KAAK,MAAM;gBAC3B,CAAC,CAAC,CAAC,WAAW,CAAC,aAAa,IAAI,IAAI,CAAC,WAAW,CAAC;gBACjD,CAAC,CAAC,IAAI,CAAC,WAAW,CAAC;YACrB,OAAO;gBACN,OAAO;gBACP,MAAM,EAAE;oBACP,aAAa;oBACb,aAAa,EAAE,qBAAqB,CACnC,aAAa,EACb,IAAI,CAAC,WAAW,CAChB;oBACD,UAAU,EAAE,eAAe,CAC1B,SAAS,CAAC,GAAG,EACb,OAAO,CAAC,WAAW,CAAC,KAAK,EACzB,OAAO,CAAC,WAAW,CAAC,GAAG,CACvB;iBACD;aACD,CAAC;QACH,CAAC;QACD,MAAM,WAAW,GAChB,WAAW,EAAE,IAAI,KAAK,MAAM;YAC3B,CAAC,CAAC,CAAC,WAAW,CAAC,aAAa,IAAI,IAAI,CAAC;YACrC,CAAC,CAAC,IAAI,CAAC;QACT,OAAO;YACN,OAAO;YACP,MAAM,EAAE;gBACP,aAAa,EAAE,WAAW;gBAC1B,aAAa,EAAE,WAAW;aAC1B;SACD,CAAC;IACH,CAAC,CACD,CAAC;IAEF,OAAO;QACN,IAAI,EAAE;YACL,KAAK;YACL,MAAM,EAAE,SAAS,CAAC,MAAM;YACxB,WAAW,EAAE,SAAS,CAAC,WAAW;YAClC,eAAe,EAAE,SAAS,CAAC,eAAe;SAC1C;QACD,kBAAkB,EAAE,SAAS,CAAC,kBAAkB;QAChD,WAAW,EAAE,SAAS,CAAC,WAAW;KAClC,CAAC;AACH,CAAC,CAAC","sourcesContent":["import { resolveReadableRegion } from \"../../highlight-pipeline/index.js\";\nimport { collectMathAwareTextAndMap } from \"../../math-aware-text-processing.js\";\nimport type {\n\tNormalizedTextMap,\n\tTextProcessingOptions,\n} from \"../../text-processing.js\";\nimport {\n\tassembleGeneratedSpeech,\n\ttype MathSpeechResolver,\n} from \"../assemble-plan.js\";\nimport type { SREMathSpeechOptions } from \"../../math-speech.js\";\nimport type { GeneratedSpeechPlan, PlanItem } from \"../types.js\";\nimport type { DomAnchor } from \"./types.js\";\n\nconst sliceVisibleMap = (\n\tmap: NormalizedTextMap,\n\tstart: number,\n\tend: number,\n): NormalizedTextMap => {\n\tconst sliced: NormalizedTextMap = new Map();\n\tfor (let index = start; index < end; index++) {\n\t\tconst mapping = map.get(index);\n\t\tif (mapping) {\n\t\t\tsliced.set(index - start, mapping);\n\t\t}\n\t}\n\treturn sliced;\n};\n\nexport interface BuildGeneratedSpeechResult {\n\tplan: GeneratedSpeechPlan<DomAnchor>;\n\tcontainsMathMarkup: boolean;\n\t/** Aggregate visible text from the DOM walk (may be empty). */\n\tvisibleText: string;\n}\n\n/**\n * DOM adapter: walk a live content root, assemble the (pure) speech plan, and\n * bind each segment to its live-DOM anchor. This is the only generated-speech\n * entry point that touches the DOM; assembly and serialization stay pure.\n */\nexport const buildGeneratedSpeechFromRoot = async (args: {\n\tcontentRoot: Element;\n\tlanguage?: string;\n\tmathSpeech?: SREMathSpeechOptions;\n\ttextProcessingOptions?: TextProcessingOptions;\n\t/** Request SRE SSML per equation (for the SSML playback format). */\n\tproduceSsml?: boolean;\n\tresolveMathSpeech?: MathSpeechResolver;\n}): Promise<BuildGeneratedSpeechResult> => {\n\tconst extracted = collectMathAwareTextAndMap(\n\t\targs.contentRoot,\n\t\targs.textProcessingOptions,\n\t);\n\tconst assembled = await assembleGeneratedSpeech({\n\t\tchunks: extracted.chunks,\n\t\tvisibleText: extracted.visibleText,\n\t\tlanguage: args.language,\n\t\tmathSpeech: args.mathSpeech,\n\t\tproduceSsml: args.produceSsml,\n\t\tresolveMathSpeech: args.resolveMathSpeech,\n\t});\n\n\tconst items: PlanItem<DomAnchor>[] = assembled.segments.map(\n\t\t({ segment, sourceChunkIndex }) => {\n\t\t\tconst sourceChunk = extracted.chunks[sourceChunkIndex];\n\t\t\tif (segment.kind === \"prose\") {\n\t\t\t\tconst sourceElement =\n\t\t\t\t\tsourceChunk?.type === \"text\"\n\t\t\t\t\t\t? (sourceChunk.sourceElement ?? args.contentRoot)\n\t\t\t\t\t\t: args.contentRoot;\n\t\t\t\treturn {\n\t\t\t\t\tsegment,\n\t\t\t\t\tanchor: {\n\t\t\t\t\t\tsourceElement,\n\t\t\t\t\t\tregionElement: resolveReadableRegion(\n\t\t\t\t\t\t\tsourceElement,\n\t\t\t\t\t\t\targs.contentRoot,\n\t\t\t\t\t\t),\n\t\t\t\t\t\tvisibleMap: sliceVisibleMap(\n\t\t\t\t\t\t\textracted.map,\n\t\t\t\t\t\t\tsegment.visibleSpan.start,\n\t\t\t\t\t\t\tsegment.visibleSpan.end,\n\t\t\t\t\t\t),\n\t\t\t\t\t},\n\t\t\t\t};\n\t\t\t}\n\t\t\tconst mathElement =\n\t\t\t\tsourceChunk?.type === \"math\"\n\t\t\t\t\t? (sourceChunk.sourceElement ?? null)\n\t\t\t\t\t: null;\n\t\t\treturn {\n\t\t\t\tsegment,\n\t\t\t\tanchor: {\n\t\t\t\t\tsourceElement: mathElement,\n\t\t\t\t\tregionElement: mathElement,\n\t\t\t\t},\n\t\t\t};\n\t\t},\n\t);\n\n\treturn {\n\t\tplan: {\n\t\t\titems,\n\t\t\tlocale: assembled.locale,\n\t\t\tvisibleText: assembled.visibleText,\n\t\t\tplainSpeechText: assembled.plainSpeechText,\n\t\t},\n\t\tcontainsMathMarkup: extracted.containsMathMarkup,\n\t\tvisibleText: extracted.visibleText,\n\t};\n};\n"]}

package/dist/services/tts/generated-speech/dom/to-chunk-inputs.d.ts

@@ -0,0 +1,19 @@
+import type { GeneratedSpeechPlan } from "../types.js";
+import type { DomAnchor, GeneratedSpeechChunk } from "./types.js";
+export interface PlanToChunksOptions {
+    /**
+     * Playback format:
+     *   - "plain" (default): plain spoken text for every chunk (Phase A, the
+     *     stable generated-speech behavior).
+     *   - "ssml": math chunks with SRE SSML send that SSML to the provider (with
+     *     a plain fallback attached); prose chunks always stay plain.
+     */
+    format?: "plain" | "ssml";
+}
+/**
+ * Convert an anchored speech plan into playback chunks for
+ * `TTSService.speakCatalogChunk`. The aggregate `contentToSpeak` stays plain
+ * (see `plan.plainSpeechText`), so no SSML leaks into break-detection or seek
+ * planning — SSML lives only in per-chunk `speechText`.
+ */
+export declare const planToCompositionChunkInputs: (plan: GeneratedSpeechPlan<DomAnchor>, options?: PlanToChunksOptions) => GeneratedSpeechChunk[];