Skip to content

armgit / hypit

Public Imported from GitHub
0

Clone any viral video with AI agents. Not just a script, the whole workflow: swap the face, the words, the B-roll, ship 100 variants in one command, and get your 100M views.

refactor: let the recording be reported, not judged

kashorin7b8a7a13699b
diff --git a/examples/echo-pro-aroll/svml.packages.lock b/examples/echo-pro-aroll/svml.packages.lock
index b3ae3583..450e14b9 100644
--- a/examples/echo-pro-aroll/svml.packages.lock
+++ b/examples/echo-pro-aroll/svml.packages.lock
@@ -129,7 +129,7 @@
{
"name": "@narratage/semantic-map",
"version": "0.0.0-dev",
- "digest": "sha256:418cd97bbc9be56a677da7ea78c97174127cf9c8858f7577eb86e13d743e890d"
+ "digest": "sha256:4293518a23ead1320f81d5d38e60bef96ea7f27149495bc890b9e4c275572864"
},
{
"name": "@narratage/source",
@@ -139,7 +139,7 @@
{
"name": "@narratage/speech-alignment",
"version": "0.0.0-dev",
- "digest": "sha256:e0345266c4ddefa0aed80619d315d74d1d7e8a44936e2bc6e30698bf5da4a103"
+ "digest": "sha256:98c3158c130b90b443283f19c770f358ed81caa915a8e741a699f8ffe3ea9f6d"
},
{
"name": "@narratage/speech-basis",
@@ -204,7 +204,7 @@
"version": "0.0.0-dev"
},
"facetsDigest": "sha256:86b692dc498c7d51fb87aab99ceb88464f2e3a86ac48a8abf64349dcc38fb7d0",
- "closureDigest": "sha256:85e86fbcc8fde5dc64808e42fac53c83ede6a671a6a3fd5a80dcdb507529d865"
+ "closureDigest": "sha256:0895c2f9a768c88b5ade0b3fa2a8e55b542c56f7badc2250ec196353d0aba40e"
},
{
"specifier": "@narratage/caption-gemini",
@@ -213,7 +213,7 @@
"version": "0.0.0-dev"
},
"facetsDigest": "sha256:bee8a4fce6b8213bd1ca1a94811cdcf261b71ecc39ecbf1593afa67f203b7cd5",
- "closureDigest": "sha256:2c3351b8f7e91c1a2cc59fc67a8f1f06cce8e7308aa93d11c1050884071cd735"
+ "closureDigest": "sha256:92ac37f4f303c1b9f5046f132f660e709181e292365f43b52e9b45cd1fb9dcd0"
},
{
"specifier": "@narratage/composition",
@@ -357,7 +357,7 @@
"version": "0.0.0-dev"
},
"facetsDigest": "sha256:4f041c5844a3eb9a70e8472d4cf4f534de960a77745819366dac0061ec600e76",
- "closureDigest": "sha256:8d526afa5f3356e58a5ff4b6d912b3bd9466c1214aea0caa85ee6cb7502df5a4"
+ "closureDigest": "sha256:07e886c67958b327fe79726a2f6ac967cb79088fa4ca33882196922ce740532a"
},
{
"specifier": "@narratage/speech",
@@ -375,7 +375,7 @@
"version": "0.0.0-dev"
},
"facetsDigest": "sha256:b3f48d736480bd0e1911a9195d9143ffb5c38a91bdcc55584845d6488cccc14f",
- "closureDigest": "sha256:70610ff9a7bf41b764f878d2c8f551f6369d737be616564db84fcaf2ab7093e5"
+ "closureDigest": "sha256:b7604040e35473309ca5843cbcbe3d0f1a453fa1f8353da1bdc4361ceec14519"
},
{
"specifier": "@narratage/speech-basis",
@@ -429,8 +429,8 @@
"version": "0.0.0-dev"
},
"facetsDigest": "sha256:0eb8b7e5605851c873e0d37385577242a761e23d6eae9b0a495a2667ba71a082",
- "closureDigest": "sha256:a38f468ae458b66da1340cc6fe88ef695ccff51c6a82ca3fbd4284879a44581f"
+ "closureDigest": "sha256:d342b6dd4602b1cea54a1e8bdb221d80f6c9d32c3a7b74d4a522415dbfba5d0f"
}
],
- "digest": "sha256:ad5fb2012a395b6a84fa9f36dde0d1897bec70cdd93a115b38d74448763878c9"
+ "digest": "sha256:a0b95c9ea95d2b93fb7abf49d09fc9e09a4a92abbe82f86062dfcf81a5b048eb"
}
diff --git a/examples/echo-pro-aroll/svml.runtime-packages.lock b/examples/echo-pro-aroll/svml.runtime-packages.lock
index 548aae00..9829214b 100644
--- a/examples/echo-pro-aroll/svml.runtime-packages.lock
+++ b/examples/echo-pro-aroll/svml.runtime-packages.lock
@@ -129,7 +129,7 @@
{
"name": "@narratage/provider-whisperx-local",
"version": "0.0.0-dev",
- "digest": "sha256:e0cf081b8713a757f71d96aa6e7587bdeeb1513ba5d06f747f69c8d82c732a35"
+ "digest": "sha256:ec4f25269c4aed169781ec0364e9fbcaa932eb816402856f06c54b746bf277ec"
},
{
"name": "@narratage/render-hyperframes",
@@ -154,7 +154,7 @@
{
"name": "@narratage/semantic-map",
"version": "0.0.0-dev",
- "digest": "sha256:418cd97bbc9be56a677da7ea78c97174127cf9c8858f7577eb86e13d743e890d"
+ "digest": "sha256:4293518a23ead1320f81d5d38e60bef96ea7f27149495bc890b9e4c275572864"
},
{
"name": "@narratage/source",
@@ -164,7 +164,7 @@
{
"name": "@narratage/speech-alignment",
"version": "0.0.0-dev",
- "digest": "sha256:e0345266c4ddefa0aed80619d315d74d1d7e8a44936e2bc6e30698bf5da4a103"
+ "digest": "sha256:98c3158c130b90b443283f19c770f358ed81caa915a8e741a699f8ffe3ea9f6d"
},
{
"name": "@narratage/speech-evidence",
@@ -850,7 +850,7 @@
"version": "0.0.0-dev"
},
"facetsDigest": "sha256:b2636492ab0b0b500d674899ab2ff14fc61c7f2b22ef44eb8f816f141b21ec4d",
- "closureDigest": "sha256:5fd699a33ac7215075c4ffbd023ed59b80730356f1bf480017be8f9feb774090"
+ "closureDigest": "sha256:ff3178552036273ca7397ed62e58ebe5184fd0e324f97c2540b7cce0f0503c13"
},
{
"specifier": "@narratage/provider-hyperframes-local",
@@ -886,8 +886,8 @@
"version": "0.0.0-dev"
},
"facetsDigest": "sha256:3686b84c42cc2b1bcaccf57cbdcda25a15c7d9d3778fb92d25437f9880ed8f5d",
- "closureDigest": "sha256:14db2383738515add6a4d4d8c91e345e2e8b67d47cce43810f7a732016d8f3e4"
+ "closureDigest": "sha256:45db58b6f207565bde42852a4f14e31f84d40b74d99158c8f03e18e6163ec2b7"
}
],
- "digest": "sha256:9dcca00dcf02deb59c1679e20921e668ed05d997515259c0d0235c33b5657226"
+ "digest": "sha256:e2ac372472113278dc44b30afed0033b3d6bb9c30e26f718748cb4b935156b56"
}
diff --git a/examples/talking-film-graph-check/svml.packages.lock b/examples/talking-film-graph-check/svml.packages.lock
index 7e1aa2bc..f7efcdcf 100644
--- a/examples/talking-film-graph-check/svml.packages.lock
+++ b/examples/talking-film-graph-check/svml.packages.lock
@@ -119,7 +119,7 @@
{
"name": "@narratage/semantic-map",
"version": "0.0.0-dev",
- "digest": "sha256:418cd97bbc9be56a677da7ea78c97174127cf9c8858f7577eb86e13d743e890d"
+ "digest": "sha256:4293518a23ead1320f81d5d38e60bef96ea7f27149495bc890b9e4c275572864"
},
{
"name": "@narratage/source",
@@ -129,7 +129,7 @@
{
"name": "@narratage/speech-alignment",
"version": "0.0.0-dev",
- "digest": "sha256:e0345266c4ddefa0aed80619d315d74d1d7e8a44936e2bc6e30698bf5da4a103"
+ "digest": "sha256:98c3158c130b90b443283f19c770f358ed81caa915a8e741a699f8ffe3ea9f6d"
},
{
"name": "@narratage/speech-basis",
@@ -199,7 +199,7 @@
"version": "0.0.0-dev"
},
"facetsDigest": "sha256:a30c9a993e7f24f39a78b80f8262022ca8d6250b4d454fc8de91c8922bf8d45a",
- "closureDigest": "sha256:fda5a82565390c91678b06c8143c16e90fe6edaeed0a15bc046e2f6f27ff6821"
+ "closureDigest": "sha256:7b547f5f2b14bec1640f1c3d683f02899ef394265018b213058c103b9f9bca1b"
},
{
"specifier": "@narratage/caption",
@@ -208,7 +208,7 @@
"version": "0.0.0-dev"
},
"facetsDigest": "sha256:86b692dc498c7d51fb87aab99ceb88464f2e3a86ac48a8abf64349dcc38fb7d0",
- "closureDigest": "sha256:85e86fbcc8fde5dc64808e42fac53c83ede6a671a6a3fd5a80dcdb507529d865"
+ "closureDigest": "sha256:0895c2f9a768c88b5ade0b3fa2a8e55b542c56f7badc2250ec196353d0aba40e"
},
{
"specifier": "@narratage/caption-gemini",
@@ -217,7 +217,7 @@
"version": "0.0.0-dev"
},
"facetsDigest": "sha256:bee8a4fce6b8213bd1ca1a94811cdcf261b71ecc39ecbf1593afa67f203b7cd5",
- "closureDigest": "sha256:2c3351b8f7e91c1a2cc59fc67a8f1f06cce8e7308aa93d11c1050884071cd735"
+ "closureDigest": "sha256:92ac37f4f303c1b9f5046f132f660e709181e292365f43b52e9b45cd1fb9dcd0"
},
{
"specifier": "@narratage/composition",
@@ -334,7 +334,7 @@
"version": "0.0.0-dev"
},
"facetsDigest": "sha256:4f041c5844a3eb9a70e8472d4cf4f534de960a77745819366dac0061ec600e76",
- "closureDigest": "sha256:8d526afa5f3356e58a5ff4b6d912b3bd9466c1214aea0caa85ee6cb7502df5a4"
+ "closureDigest": "sha256:07e886c67958b327fe79726a2f6ac967cb79088fa4ca33882196922ce740532a"
},
{
"specifier": "@narratage/speech",
@@ -352,7 +352,7 @@
"version": "0.0.0-dev"
},
"facetsDigest": "sha256:b3f48d736480bd0e1911a9195d9143ffb5c38a91bdcc55584845d6488cccc14f",
- "closureDigest": "sha256:70610ff9a7bf41b764f878d2c8f551f6369d737be616564db84fcaf2ab7093e5"
+ "closureDigest": "sha256:b7604040e35473309ca5843cbcbe3d0f1a453fa1f8353da1bdc4361ceec14519"
},
{
"specifier": "@narratage/speech-basis",
@@ -397,7 +397,7 @@
"version": "0.0.0-dev"
},
"facetsDigest": "sha256:9384a365dc85a38dcbd32fa316b9ecca1f226587494bb958a285d797f7babac6",
- "closureDigest": "sha256:f93d5bb0370eaac00e7510d621e302e55a1f69543584174ec89795564f28242c"
+ "closureDigest": "sha256:e2d3491d96b7cbe13e9355a12a5002d79aaf4b26afe2ce919a2aff48d4055767"
},
{
"specifier": "@narratage/visual-ir",
@@ -415,8 +415,8 @@
"version": "0.0.0-dev"
},
"facetsDigest": "sha256:0eb8b7e5605851c873e0d37385577242a761e23d6eae9b0a495a2667ba71a082",
- "closureDigest": "sha256:a38f468ae458b66da1340cc6fe88ef695ccff51c6a82ca3fbd4284879a44581f"
+ "closureDigest": "sha256:d342b6dd4602b1cea54a1e8bdb221d80f6c9d32c3a7b74d4a522415dbfba5d0f"
}
],
- "digest": "sha256:8433f850ff933acd18e99373d35f41be655f966d7a4e511e91d2a13a9e78b121"
+ "digest": "sha256:b0a665c5a0367630faf1283727be48682e016d30517878da3d18ae6314c4252d"
}
diff --git a/packages/provider-whisperx-local/src/provider.ts b/packages/provider-whisperx-local/src/provider.ts
index dc11e82e..a00f07ea 100644
--- a/packages/provider-whisperx-local/src/provider.ts
+++ b/packages/provider-whisperx-local/src/provider.ts
@@ -178,14 +178,14 @@ export function interpretWhisperXResponse(
if (text.length === 0) continue;
const start = finite(rawWord.start) ? rawWord.start : undefined;
const end = finite(rawWord.end) ? rawWord.end : undefined;
- assert((start === undefined) === (end === undefined), `WhisperX Word ${text} has partial timing`);
- if (start !== undefined && end !== undefined) {
- assert(start >= 0 && end >= start && end <= durationSec + 1e-3,
- `WhisperX Word ${text} lies outside the evidence audio`);
- }
const source = sourceSegmentFor(sourceSegments, start, end, segmentStart, segmentEnd);
const prior = previousEnd.get(source.segmentId)!;
+ // A word whose timing cannot be proved keeps its text and loses its clock,
+ // exactly as the sidecar already does for words its aligner could not place.
const timingIsProvable = start !== undefined && end !== undefined
+ && start >= 0
+ && end >= start
+ && end <= durationSec + 1e-3
&& start >= source.startSec - 1e-6
&& end <= source.endSec + 1e-6
&& start >= prior - 1e-6;
diff --git a/packages/semantic-map/src/location.ts b/packages/semantic-map/src/location.ts
index cf481fbd..3f6314d0 100644
--- a/packages/semantic-map/src/location.ts
+++ b/packages/semantic-map/src/location.ts
@@ -56,6 +56,10 @@ function frameFor(frames: ReadonlyMap<string, number>, anchorId: string, owner:
* order, and occurrences are located independently: they are never sorted,
* merged, clipped against one another or otherwise reconciled. A caller that
* needs time order sorts them itself.
+ *
+ * An occurrence whose two anchors meet or cross yields an empty or backwards
+ * span, returned as located. What a Selection covering no frames should show is
+ * a question about material, answered where that material is projected.
*/
export function selectionFrameSpans(
map: CompleteSemanticMap,
@@ -69,9 +73,6 @@ export function selectionFrameSpans(
return selection.occurrences.map((occurrence) => {
const startFrame = frameFor(frames, occurrence.open.boundary.anchorId, `NarrativeSelection ${selection.id}`);
const endFrameExclusive = frameFor(frames, occurrence.close.boundary.anchorId, `NarrativeSelection ${selection.id}`);
- if (endFrameExclusive <= startFrame) {
- throw new Error(`NarrativeSelection ${selection.id} contains an empty located occurrence`);
- }
return { startFrame, endFrameExclusive };
});
}
diff --git a/packages/speech-alignment/src/locate.ts b/packages/speech-alignment/src/locate.ts
index bb865804..3d77f22f 100644
--- a/packages/speech-alignment/src/locate.ts
+++ b/packages/speech-alignment/src/locate.ts
@@ -32,9 +32,14 @@ function finite(value: number | undefined): value is number {
return value !== undefined && Number.isFinite(value);
}
-function validateWindow(start: number, end: number, limit: number, label: string): void {
- if (!Number.isFinite(start) || !Number.isFinite(end) || start < 0 || end < start || end > limit + EPSILON) {
- fail("SPEECH_WINDOW", `${label} has invalid time window ${start}..${end}.`);
+/**
+ * A time is rejected only when no frame can be derived from it. Whether a window
+ * runs backwards, leaves its Segment or reaches past the programme is a fact
+ * about the recording, reported as measured.
+ */
+function validateWindow(start: number, end: number, _limit: number, label: string): void {
+ if (!Number.isFinite(start) || !Number.isFinite(end) || start < 0 || end < 0) {
+ fail("SPEECH_WINDOW", `${label} has an unusable time ${start}..${end}.`);
}
}
@@ -57,17 +62,12 @@ function validateBasis(narrative: Narrative, basis: SpeechAudioBasis): void {
if (basis.segments.length !== narrative.segments.length) {
fail("SPEECH_BASIS_SEGMENTS", "SpeechAudioBasis must cover every Narrative Segment exactly once.");
}
- let previousEnd = 0;
for (const [index, segment] of basis.segments.entries()) {
const expected = narrative.segments[index]!;
if (segment.segmentId !== expected.id) {
fail("SPEECH_BASIS_SEGMENTS", `SpeechAudioBasis Segment ${segment.segmentId} does not match ${expected.id}.`);
}
validateWindow(segment.startSec, segment.endSec, basis.programSpace.durationSec, `Basis Segment ${segment.segmentId}`);
- if (segment.startSec < previousEnd - EPSILON) {
- fail("SPEECH_BASIS_SEGMENTS", `SpeechAudioBasis Segment ${segment.segmentId} overlaps its predecessor.`);
- }
- previousEnd = segment.endSec;
}
}
@@ -96,86 +96,38 @@ function validateEvidence(
}
seen.add(segment.sourceSegmentId);
validateWindow(segment.startSec, segment.endSec, evidence.durationSec, `Segment ${segment.sourceSegmentId}`);
- let previousEnd = segment.startSec;
for (const [index, word] of segment.words.entries()) {
if (typeof word.text !== "string") fail("SPEECH_WORD_TEXT", `Word ${index + 1} has no text.`);
- if (finite(word.startSec) !== finite(word.endSec)) {
- fail("SPEECH_WORD_PARTIAL_TIME", `Word ${index + 1} must provide both start and end or neither.`);
- }
if (finite(word.startSec) && finite(word.endSec)) {
validateWindow(word.startSec, word.endSec, evidence.durationSec, `Word ${index + 1}`);
- if (
- word.startSec < segment.startSec - EPSILON
- || word.endSec > segment.endSec + EPSILON
- || word.startSec < previousEnd - EPSILON
- ) {
- fail("SPEECH_WORD_ORDER", `Word ${index + 1} is outside or overlaps its Segment window.`);
- }
- previousEnd = word.endSec;
- }
- if (word.score !== undefined && (!Number.isFinite(word.score) || word.score < 0 || word.score > 1)) {
- fail("SPEECH_SCORE", `Word ${index + 1} score must be between zero and one.`);
}
}
for (const [index, char] of segment.chars.entries()) {
if (!Number.isInteger(char.wordIndex) || char.wordIndex < 0 || char.wordIndex >= segment.words.length) {
fail("SPEECH_CHAR_WORD", `Character ${index + 1} has an invalid wordIndex.`);
}
- if (finite(char.startSec) !== finite(char.endSec)) {
- fail("SPEECH_CHAR_PARTIAL_TIME", `Character ${index + 1} must provide both start and end or neither.`);
- }
if (finite(char.startSec) && finite(char.endSec)) {
validateWindow(char.startSec, char.endSec, evidence.durationSec, `Character ${index + 1}`);
- const word = segment.words[char.wordIndex]!;
- if (
- char.startSec < segment.startSec - EPSILON
- || char.endSec > segment.endSec + EPSILON
- || (finite(word.startSec) && char.startSec < word.startSec - EPSILON)
- || (finite(word.endSec) && char.endSec > word.endSec + EPSILON)
- ) {
- fail("SPEECH_CHAR_WINDOW", `Character ${index + 1} falls outside its Word or Segment window.`);
- }
- }
- if (char.score !== undefined && (!Number.isFinite(char.score) || char.score < 0 || char.score > 1)) {
- fail("SPEECH_SCORE", `Character ${index + 1} score must be between zero and one.`);
}
}
- let previousVadEnd = segment.startSec;
for (const [index, span] of (segment.speechActivity ?? []).entries()) {
validateWindow(span.startSec, span.endSec, evidence.durationSec, `VAD span ${index + 1}`);
- if (
- span.startSec < segment.startSec - EPSILON
- || span.endSec > segment.endSec + EPSILON
- || span.startSec < previousVadEnd - EPSILON
- ) {
- fail("SPEECH_ACTIVITY_WINDOW", `Speech activity ${index + 1} falls outside Segment ${segment.sourceSegmentId}.`);
- }
- previousVadEnd = span.endSec;
}
}
for (const id of expected) {
if (!seen.has(id)) fail("SPEECH_SEGMENT_MISSING", `Aligned transcript is missing Segment ${id}.`);
}
const segmentsById = new Map(evidence.segments.map((segment) => [segment.sourceSegmentId, segment]));
- let previousSegmentStart = -Infinity;
- let previousSegmentEnd = -Infinity;
for (const sourceSegment of narrative.segments) {
const segment = segmentsById.get(sourceSegment.id)!;
const basisSegment = basis.segments.find((item) => item.segmentId === sourceSegment.id)!;
+ // The transcript must answer the segmentation it was asked about.
if (
Math.abs(segment.startSec - basisSegment.startSec) > EPSILON
|| Math.abs(segment.endSec - basisSegment.endSec) > EPSILON
) {
fail("SPEECH_EVIDENCE_SEGMENT_AFFINITY", `Segment ${segment.sourceSegmentId} differs from SpeechBasis.`);
}
- if (
- segment.startSec < previousSegmentStart - EPSILON
- || segment.endSec < previousSegmentEnd - EPSILON
- ) {
- fail("SPEECH_SEGMENT_ORDER", `Segment ${segment.sourceSegmentId} violates Script source order.`);
- }
- previousSegmentStart = segment.startSec;
- previousSegmentEnd = segment.endSec;
}
}
@@ -366,6 +318,21 @@ function secondsFor(basis: SpeechAudioBasis, frame: number): number {
return frame * denominator / numerator;
}
+/**
+ * Locate every Script token against one recording.
+ *
+ * The result is total: every token of every Segment carries a window, whether it
+ * was measured, derived from a neighbouring character run, or interpolated
+ * because the transcript never reached it. Whether a window runs backwards,
+ * overlaps its neighbour or leaves its Segment is reported as measured — those
+ * are facts about the recording, and deciding what they mean belongs to whoever
+ * projects them onto a timeline.
+ *
+ * Locating fails only when the Script, the audio and the transcript are not the
+ * same three things: a missing or repeated Segment, a Segment the Script never
+ * declared, a transcript answering a different segmentation, or a duration that
+ * belongs to other audio. It never fails because of a timestamp.
+ */
export function locateSpeechTiming(
narrative: Narrative,
basis: SpeechAudioBasis,
diff --git a/packages/speech-alignment/test/speech-align.test.ts b/packages/speech-alignment/test/speech-align.test.ts
index 84285b0f..97bb7027 100644
--- a/packages/speech-alignment/test/speech-align.test.ts
+++ b/packages/speech-alignment/test/speech-align.test.ts
@@ -305,17 +305,34 @@ test("multiple Script Segments stay independent even when evidence records arriv
]);
});
-test("invalid overlapping evidence word windows fail instead of producing a reversed map", () => {
- const narrative = parseScript("bad.svml", "<line>one two</line>");
- assert.throws(
- () => locate(narrative, {
- words: [
- { text: "one", startSec: 0.1, endSec: 0.5 },
- { text: "two", startSec: 0.4, endSec: 0.8 },
- ],
- }),
- (error: unknown) => error instanceof SpeechAlignmentError && error.code === "SPEECH_WORD_ORDER",
- );
+test("overlapping evidence word windows reach the map overlapping", () => {
+ // Two spoken words whose measured windows overlap. Nothing here knows whether
+ // that is a real overlap or a wobble, and no consumer of the map is bound to
+ // treat it as either, so it is reported as measured.
+ const narrative = parseScript("overlap.svml", "<line>one two</line>");
+ const map = locate(narrative, {
+ words: [
+ { text: "one", startSec: 0.1, endSec: 0.5 },
+ { text: "two", startSec: 0.4, endSec: 0.8 },
+ ],
+ });
+ assert.deepEqual(map.tokens.map((token) => [token.startSec, token.endSec]), [[0.1, 0.5], [0.4, 0.8]]);
+ assert.equal(map.tokens[1]!.startSec < map.tokens[0]!.endSec, true, "the overlap survived");
+});
+
+test("locating is total: every Script token carries a window", () => {
+ const narrative = parseScript("total.svml", "<line>alpha beta gamma delta</line>");
+ // Nothing the aligner could place: every window below is interpolated.
+ const blind = locate(narrative, {
+ words: [{ text: "alpha" }, { text: "beta" }, { text: "gamma" }, { text: "delta" }],
+ });
+ assert.equal(blind.tokens.length, narrative.tokens.length);
+ assert.equal(blind.tokens.every((token) => Number.isFinite(token.startSec) && Number.isFinite(token.endSec)), true);
+
+ // The speaker said something else entirely; the Script is still fully located.
+ const diverged = locate(narrative, { words: [{ text: "zzz", startSec: 0.2, endSec: 0.8 }] });
+ assert.equal(diverged.tokens.length, narrative.tokens.length);
+ assert.equal(diverged.tokens.every((token) => Number.isFinite(token.startSec)), true);
});
test("Evidence is interpreted only through the explicitly connected SpeechAudioBasis", () => {
diff --git a/spec/script-surface.md b/spec/script-surface.md
index 9289fefc..b01407c6 100644
--- a/spec/script-surface.md
+++ b/spec/script-surface.md
@@ -344,6 +344,12 @@ SemanticAnchorIdentity → ProgramPoint(ProgramSpace)
切点在此完全平等:段首标记左吸附取该 Segment 自己的起始切点,绝不越到前一个
Segment 的词尾。下游拿到的是 anchor 身份,不再自行做下标推算。
+定位是全覆盖的:每个 Segment 的每个 token 都带有窗口,无论它被测得、由相邻
+字符推得,还是因为转写从未触及而由插值补出。窗口倒序、越出所属 Segment 或
+与邻居重叠,都按测得原样报出;那是关于录音的事实,如何解释属于把它投影到
+时间轴的那一方。定位只在 Script、音频与转写不是同一组三件事时失败,绝不因为
+一个时间戳而失败。
+
每个点都必须存在。消费者不得补点、移动点或从相邻 occurrence 借点。每个 Segment
内部必须非降序: