@allyhub/sdk 0.1.0-alpha.7 → 0.2.0-next.0

package/node_modules/@allyhub-next/video-core/dist/timing.d.ts

@@ -0,0 +1,88 @@
+import { type LineAnchor } from "./anchors.js";
+import { type Problem } from "./problem.js";
+import type { Word } from "./media.js";
+export interface PartInput {
+    key: string;
+    /** The part's words on its speech's clock (seconds from the start of its speech). */
+    words?: readonly Word[];
+    /** Seconds of speech audio. With it the part is voiced. */
+    speechDur?: number;
+    /**
+     * The line the part says, as spoken, when the script is known (say,
+     * saying): word references are looked up in it and mapped to the words,
+     * so a misheard word still lands where it is said.
+     */
+    script?: string;
+    /**
+     * The anchors of that line (its script's @{name}), by the letters and
+     * digits before each: what "@name" and "@/name" refer to.
+     */
+    anchors?: readonly LineAnchor[];
+    /**
+     * The part's own anchors (`part({ anchors })`), in seconds from its
+     * start: "@name" and "@/name" refer to them too, speech or none.
+     */
+    placed?: readonly PlacedAnchor[];
+    /** Seconds, for a part without speech. */
+    dur: number;
+    /**
+     * The part has speech that could not be found (its own problem is
+     * reported): word references are estimated from where they sit in the
+     * script, without another problem.
+     */
+    pending?: boolean;
+    lead: number;
+    tail: number;
+    /**
+     * J and L cuts (voiced parts): seconds of the speech played under the
+     * previous part's picture (`before`) and under the next one's (`after`).
+     */
+    before?: number;
+    after?: number;
+}
+/** An anchor a part places itself: a point, or a stretch to `end`, in seconds from its start. */
+export interface PlacedAnchor {
+    name: string;
+    at: number;
+    end?: number;
+}
+export interface TimingOptions {
+    fps?: number;
+    /** Seconds per beat; parts without speech end on the beat grid. */
+    beat?: number;
+    /** Film seconds of one beat of the grid; default 0. */
+    beatPhase?: number;
+    /**
+     * Where a part without speech ends instead of the grid: the beat nearest
+     * `t` that is after `after` (the music's beats on the film, which its
+     * stretches and loops move); undefined leaves the part as long as it is.
+     */
+    snap?: (t: number, after: number) => number | undefined;
+}
+export interface TimedPart {
+    key: string;
+    start: number;
+    end: number;
+    /**
+     * Film time where the part's speech starts (voiced parts only); before
+     * `start` in a J cut.
+     */
+    speechAt?: number;
+}
+export interface Timing {
+    parts: TimedPart[];
+    duration: number;
+    problems: Problem[];
+    /** Seconds from part n's start of a word reference; problems go to `problems`. */
+    at(n: number, ref: unknown): number;
+    /** Part n's words on its own clock (seconds from the part's start). */
+    words(n: number): Word[];
+}
+/**
+ * A voiced part's J or L overlap: never negative, and at most its lead
+ * (or tail) and half its speech, so its picture keeps some of its own
+ * speech.
+ */
+export declare function overlapOf(p: PartInput, side: "before" | "after"): number;
+export declare function timeParts(given: readonly PartInput[], opts?: TimingOptions): Timing;
+//# sourceMappingURL=timing.d.ts.map

package/node_modules/@allyhub-next/video-core/dist/timing.js

@@ -0,0 +1,2 @@
+/*! AllyHub. Copyright (c) 2026 AllyHub, Inc. All rights reserved. Use is subject to LICENSE.md. */
+import{parseAnchorRef as z}from"./anchors.js";import{FPS as B}from"./formats.js";import{problem as R}from"./problem.js";import{joinWords as E,normText as W,numericRef as H,parseWordRef as I,ScriptAlignment as J,WordStream as j}from"./words.js";const g=i=>Math.round(i*1e3)/1e3;function L(i,l,a){const x=w=>Math.round(w*a)/a,p=i.map(x);for(let w=1;w<p.length;w++){const d=p[w-1]??0;(p[w]??0)<=d&&(p[w]=d+1/a)}const b=p.at(-1)??0;return{starts:p,end:Math.max(x(l),b+1/a)}}const F=i=>i.speechDur!==void 0&&i.speechDur>0;export function overlapOf(i,l){const a=i[l]??0;if(!F(i)||!Number.isFinite(a)||a<=0)return 0;const x=(i.speechDur??0)/2+(l==="before"?i.lead:i.tail);return Math.min(a,x)}export function timeParts(i,l={}){const a=i.map((t,e)=>({...t,before:e===0?0:t.before??0,after:e===i.length-1?0:t.after??0})),x=l.fps??B,p=l.beat??0,b=[];for(const t of a){const e=new Set((t.placed??[]).map(n=>n.name));for(const n of t.anchors??[])e.has(n.name)&&b.push(R("AK-T004",`The anchor ${n.name} of part ${t.key} is both in the line it says (@{${n.name}}) and in the part's own anchors; "@${n.name}" is the part's own. Rename one of them.`,{part:t.key}))}const w=[];let d=0;for(const t of a){w.push(d);let e;if(F(t))e=t.lead+(t.speechDur??0)+t.tail-overlapOf(t,"before")-overlapOf(t,"after");else{e=t.dur;const n=l.snap?.(d+e,d);if(n!==void 0)e=n-d;else if(p>0&&!l.snap){const r=l.beatPhase??0;let s=r+Math.round((d+e-r)/p)*p;s<=d+1e-9&&(s+=p),e=s-d}}d+=Math.max(e,1/x)}const{starts:G,end:K}=L(w,d,x),T=a.map((t,e)=>{const n=G[e]??0;return{key:t.key,start:n,end:G[e+1]??K,...F(t)?{speechAt:g(n+t.lead-overlapOf(t,"before"))}:{}}}),D=new Map,P=t=>{if(D.has(t))return D.get(t);const e=a[t],n=e?.script&&e.words?.length?new J(new j(e.words),[e.script]):void 0;return D.set(t,n),n};function C(t,e){const n=T[t],r=a[t];if(!n||!r)return;const s=r.anchors??[],f=r.placed??[],o=z(e,new Set([...s,...f].map(u=>u.name)));if(!o)return;const $=n.end-n.start,y=u=>(b.push(R("AK-T004",u,{part:r.key})),g($/2)),k=u=>g(Math.min($,Math.max(0,u))),m=f.find(u=>u.name===o.name);if(m)return o.end&&m.end===void 0?y(`"${e}" asks for the end of the anchor ${m.name} of part ${r.key}, which is a point. Give it as a stretch (anchors: { ${m.name}: [from, to] }), or use "@${m.name}".`):k((o.end?m.end??m.at:m.at)+o.shift);const c=s.find(u=>u.name===o.name);if(!c){const u=[...s,...f].map(q=>`@${q.name}`);return y(`The anchor "${e}" is not in the line part ${r.key} says, nor in the part's own anchors${u.length>0?` (its anchors: ${u.join(", ")})`:""}. Write @{${o.name}} before the word it is for, in the part's say in the item's script, in every language (the part must say that line, s("….say")); or give its seconds in the part: anchors: { ${o.name}: 1.2 }.`)}if(o.end&&c.end===void 0)return y(`"${e}" asks for the end of the anchor ${c.name} of part ${r.key}, which is a point. Write @{${c.name}}…@{/${c.name}} around the words in the script, or use "@${c.name}".`);const M=o.end?c.end??c.at:c.at,S=W(r.script??"");if(r.pending===!0&&n.speechAt===void 0)return S.length===0?g($/2):k(M/S.length*$+o.shift);const h=r.words??[],A=P(t);if(n.speechAt===void 0||!A||h.length===0)return y(`The anchor "${e}" is in part ${r.key}, which has no speech. Give its seconds in the part instead: anchors: { ${o.name}: 1.2 }.`);const v=h.at(-1),U=o.end?h[A.wordAtChar(0,Math.max(0,M-1))]?.end:M>=S.length?v?.end:h[A.wordAtChar(0,M)]?.start;return k(n.speechAt+(U??0)+o.shift-n.start)}function O(t,e){const n=H(e);if(n!==void 0)return n;const r=typeof e=="string"?C(t,e):void 0;if(r!==void 0)return r;const s=T[t],f=a[t];if(!s||!f)return 0;const o=s.end-s.start,$=f.words??[],{text:y,nth:k,shift:m}=I(String(e));let c=-1;$.length>0&&(c=P(t)?.wordFor(0,y,k)??-1,c<0&&(c=new j($).find(W(y),0,$.length-1,k)));const M=$[c];if(f.pending===!0&&s.speechAt===void 0){const h=W(f.script??"");let A=-1;for(let v=0;v<k;v++)A=h.indexOf(W(y),A+1);return A<0||h.length===0?g(o/2):g(Math.min(o,Math.max(0,A/h.length*o+m)))}if(c<0||!M||s.speechAt===void 0){const h=f.script??E($);return b.push(R("AK-T004",s.speechAt===void 0?`The word reference "${String(e)}" is in part ${f.key}, which has no speech. Give it in seconds from the part's start instead.`:`The word reference "${String(e)}" is not in the words of part ${f.key}${h?` (it says: "${h.length>160?`${h.slice(0,160)}…`:h}")`:""}. Use a word the part says, as "word", "word@2" or "word+0.2".`,{part:f.key})),g(o/2)}const S=s.speechAt+M.start+m;return g(Math.min(o,Math.max(0,S-s.start)))}function N(t){const e=T[t],n=a[t];if(!e||!n||e.speechAt===void 0)return[];const r=e.speechAt-e.start;return(n.words??[]).map(s=>({text:s.text,start:g(s.start+r),end:g(s.end+r)}))}return{parts:T,duration:K,problems:b,at:O,words:N}}

package/node_modules/@allyhub-next/video-core/dist/transcript.d.ts

@@ -0,0 +1,288 @@
+import type { EditedSpeech } from "./edits.js";
+import type { Word } from "./media.js";
+import { type UnheardSound, type VoiceFrames } from "./voice.js";
+/** A take's noise floor and speech level, dBFS (FilmCraft's percentiles). */
+export interface TakeLevels {
+    /** The 10th percentile of frame levels above −100 dBFS. */
+    floor: number;
+    /** The 95th percentile. */
+    speech: number;
+    /** Speech or silence: 30 % of the way from floor to speech, at least 6 dB above the floor. */
+    threshold: number;
+}
+export declare function takeLevels(frames: Pick<VoiceFrames, "db">): TakeLevels;
+/** A word's level: its loudest and its mean frame, dBFS. */
+export interface WordLevel {
+    peak: number;
+    mean: number;
+}
+export declare function wordLevels(words: readonly Word[], frames: Pick<VoiceFrames, "db" | "hop">): WordLevel[];
+/**
+ * Word bounds moved inward past frames quieter than the take's speech
+ * threshold (takeLevels), so the silence a recognizer puts inside a word
+ * goes back between words; a word keeps at least 60 ms, and a word with no
+ * frame above the threshold is left alone (FilmCraft's tighten_words).
+ * Returns new words and how many moved.
+ */
+export declare function tightenWords(words: readonly Word[], frames: Pick<VoiceFrames, "db" | "hop">): {
+    words: Word[];
+    moved: number;
+};
+/** What a word's length is compared with: a line through the take's own words. */
+export interface SpeakingRate {
+    /** Seconds a word takes whatever its length. */
+    base: number;
+    /** Seconds more for each letter. */
+    perLetter: number;
+    /** Words the line was fitted through. */
+    words: number;
+}
+/** What the measurement inside words uses. */
+export interface WordShapeOptions {
+    /** A frame counts as voiced from this voicing, 0 to 1. */
+    voiced: number;
+    /** A dip is a stretch of frames at least this many dB under the loudest frame on each side of it, inside the word. */
+    dip: number;
+}
+export declare const WORD_SHAPE_DEFAULTS: Readonly<WordShapeOptions>;
+/** A word's length against the take's rate, and the sound inside it. */
+export interface WordShape {
+    /** The word's index in the take's words. */
+    i: number;
+    /** Seconds it lasts. */
+    dur: number;
+    /** The letters its text has (0 for a word written in digits: no length follows from them). */
+    letters: number;
+    /** Seconds a word of this many letters takes by the take's rate; null without a rate or without letters. */
+    expected: number | null;
+    /** `dur` over `expected`. */
+    ratio: number | null;
+    /** `dur` minus `expected`, seconds. */
+    excess: number | null;
+    /** The stretches of voiced frames inside it, seconds of the file. */
+    voicedRuns: {
+        start: number;
+        end: number;
+    }[];
+    /** The dips in level inside it, seconds of the file, and how deep each goes, dB. */
+    dips: {
+        start: number;
+        end: number;
+        depth: number;
+    }[];
+}
+export interface WordShapes {
+    /** The take's rate; null when it has too few words to tell (`rateWhy`). */
+    rate: SpeakingRate | null;
+    rateWhy?: string;
+    options: WordShapeOptions;
+    /** One per word, in order. */
+    words: WordShape[];
+}
+/**
+ * The take's speaking rate: a line through the median length of its words
+ * of each letter count (the medians keep drawn-out words from bending it),
+ * weighted by how many words have that count. Undefined with fewer than
+ * eight words with letters, or when longer words do not last longer.
+ */
+export declare function speakingRate(words: readonly Word[]): SpeakingRate | undefined;
+/**
+ * Every word measured from inside: how long it lasts against what its
+ * letters take at the take's own rate, the stretches of voiced frames in
+ * it and the dips in level between louder frames. Nothing is judged: a
+ * word far over its expected length with two voiced stretches may hold a
+ * filler run into it, or be a word said slowly.
+ */
+export declare function wordShapes(words: readonly Word[], frames: VoiceFrames, opts?: Partial<WordShapeOptions>): WordShapes;
+/** Which words the transcript marks as long. */
+export interface LongWordOptions {
+    /** A word is listed when it lasts at least this many times its expected length… */
+    ratio: number;
+    /** …and at least this many seconds more than it. */
+    excess: number;
+}
+export declare const LONG_WORD_DEFAULTS: Readonly<LongWordOptions>;
+/** The words at least `ratio` times and `excess` seconds over their expected length. Candidates, never cuts. */
+export declare function longWords(shapes: readonly WordShape[], opts?: Partial<LongWordOptions>): WordShape[];
+/** A long word as the transcript marks it: `(long 0.62s ×2.4)`. */
+export declare function longLabel(s: WordShape): string;
+/** Words [from, to] said without a pause as long as the paragraph gap. */
+export interface Paragraph {
+    from: number;
+    to: number;
+    start: number;
+    end: number;
+    text: string;
+}
+/** A paragraph of a take ends at a pause this long, seconds. */
+export declare const PARAGRAPH_GAP = 0.5;
+/** The take's words in paragraphs: a new one after a gap of `gap` seconds or more. */
+export declare function paragraphs(words: readonly Word[], gap?: number, first?: number, last?: number): Paragraph[];
+/** Where a phrase is said: words [from, to] and their times. */
+export interface WordMatch {
+    from: number;
+    to: number;
+    start: number;
+    end: number;
+    text: string;
+}
+/**
+ * Every place the take says `phrase`: case and punctuation ignored, word
+ * by word, the last word of the phrase may be the start of a word
+ * ("manag" finds "manager").
+ */
+export declare function searchWords(words: readonly Word[], phrase: string): WordMatch[];
+/**
+ * Where the take says any of `phrases` (the agent's list: "um", "you
+ * know"), whole words, longer phrases first, never overlapping. A
+ * candidate list: nothing is removed.
+ */
+export declare function phraseMatches(words: readonly Word[], phrases: readonly string[]): (WordMatch & {
+    phrase: string;
+})[];
+/** A pause between two words, and the stretch `keep` seconds of air would remove. */
+export interface PauseCandidate {
+    /** The word before it. */
+    after: number;
+    start: number;
+    end: number;
+    gap: number;
+    /** What `maxPause: keep` cuts: the middle of the pause. */
+    remove: {
+        from: number;
+        to: number;
+        seconds: number;
+    };
+}
+/** Pauses of `min` seconds or more between words [first, last]. A candidate list. */
+export declare function pauseCandidates(words: readonly Word[], min: number, keep: number, first?: number, last?: number): PauseCandidate[];
+/** One side of a join: the kept word nearest the cut, and the sound between them. */
+export interface JoinSide {
+    /** The kept word's index in the take's words. */
+    i: number;
+    text: string;
+    /**
+     * The word's end (before the cut) or start (after it), seconds of the
+     * file; the cut itself when it is inside the word.
+     */
+    edge: number;
+    /** Seconds of the take kept between the word and the cut, by the word times (0 for a cut inside the word). */
+    gap: number;
+    /**
+     * Seconds of 10 ms frames under the quiet level reaching the cut on this
+     * side, inside the kept stretch: the silence the level shows, which may
+     * differ from `gap` when a word's bound is loose. null without levels.
+     */
+    quiet: number | null;
+    /** The loudest frame between the word and the cut, dBFS; null without levels or with no frame between. */
+    peakBetween: number | null;
+    /** The word's loudest and mean frame, dBFS; null without levels. */
+    peak: number | null;
+    mean: number | null;
+    /** The mean level of the word's `near` seconds next to the cut (its end before it, its start after it), dBFS; null without levels. */
+    near: number | null;
+}
+/** Seconds of a word next to a cut whose level `Join.step` compares, when not given. */
+export declare const JOIN_NEAR = 0.2;
+/** Where a cut take jumps from one kept stretch to the next, measured on both sides. */
+export interface Join {
+    /** Seconds of the edited take: where the join plays. */
+    at: number;
+    /** Seconds of the file where the stretch before ends, and where the next one starts. */
+    out: number;
+    in: number;
+    /** The words removed between the two kept words, if any (else a pause shortened or a sound cut). */
+    removed: {
+        from: number;
+        to: number;
+    } | null;
+    /** The cut is inside a kept word on a side (an edits file's `inWord`): no silence there by the word times. */
+    inWord: boolean;
+    before: JoinSide;
+    after: JoinSide;
+    /** `after.near` minus `before.near`, dB: the level step across the join; null without levels. */
+    step: number | null;
+}
+/**
+ * Every join of a cut take, measured: on each side, how much of the take
+ * lies between the kept word and the cut (by word times, and as frames
+ * quieter than `quietDb` by the level), the two words' levels, and the
+ * step between the `near` seconds of each word next to the cut.
+ * Numbers only: whether a join sounds natural is the agent's call.
+ */
+export declare function joinsOf(words: readonly Word[], edited: Pick<EditedSpeech, "ranges" | "kept">, frames: Pick<VoiceFrames, "db" | "hop"> | undefined, quietDb: number, near?: number): Join[];
+/** What counts as scripted, and as a line said. */
+export interface ScriptDiffOptions {
+    /** A word counts as scripted when this share of its letters matched the script exactly (0–1). */
+    match: number;
+    /** A script line counts as said at a place when this share of its letters matched there (0–1). */
+    lineMatch: number;
+}
+export declare const SCRIPT_DIFF_DEFAULTS: Readonly<ScriptDiffOptions>;
+/** A script line, measured against the take. */
+export interface ScriptLine {
+    /** Its index among the script's non-empty lines. */
+    line: number;
+    text: string;
+    /** The share of its letters the take matched exactly, aligned with the whole script (0–1). */
+    share: number;
+    /** Each place the take says it (at least `lineMatch` of its letters there). */
+    matches: WordMatch[];
+}
+/** How a take differs from the script it was read from. Data, not decisions. */
+export interface ScriptDiff {
+    /** The thresholds used. */
+    options: ScriptDiffOptions;
+    /** Stretches of words with less than `match` of their letters matched (ad-libs, fillers, false starts). */
+    unscripted: WordMatch[];
+    /** Script lines the take says more than once, each place it says them. */
+    repeated: {
+        line: number;
+        text: string;
+        matches: WordMatch[];
+    }[];
+    /** Every script line with its match share and the places it is said. */
+    lines: ScriptLine[];
+}
+export declare function scriptDiff(words: readonly Word[], lines: readonly string[], opts?: Partial<ScriptDiffOptions>): ScriptDiff;
+/** What the packed transcript marks. */
+export interface PackOptions {
+    /** A new line after a pause this long, seconds (default PARAGRAPH_GAP). */
+    gap?: number;
+    /** A paragraph goes on to a new line at the first sentence end after this many words (default 12). */
+    lineWords?: number;
+    /** Only words [from, to]. */
+    from?: number;
+    to?: number;
+    /** Sounds heard between words, shown where they are. */
+    sounds?: readonly UnheardSound[];
+    /** Word indices an edits file removes: shown struck, [-like this-]. */
+    removed?: ReadonlySet<number>;
+    /** Sound stretches an edits file removes. */
+    removedSounds?: readonly {
+        from: number;
+        to: number;
+    }[];
+    /** Stretches an edits file cuts inside kept words: shown after the word, `(cut 61.31–61.62 in word)`. */
+    inWord?: readonly {
+        from: number;
+        to: number;
+    }[];
+    /** Words to mark as long, by index: shown after the word, `(long 0.62s ×2.4)`. */
+    long?: ReadonlyMap<number, WordShape>;
+    /** Words no script matched: shown {like this}. */
+    unscripted?: ReadonlySet<number>;
+}
+/** A line of the packed transcript ends at the first sentence end after this many words. */
+export declare const LINE_WORDS = 12;
+/** A sound as the transcript shows it: `(sound 61.17–61.40 0.23s −9dB v0.87)`. */
+export declare function soundLabel(s: UnheardSound, removed?: boolean): string;
+/**
+ * The take as lines an agent plans cuts from: one per paragraph,
+ * `[012.40-015.10] #57-#71 text…`, word indices where they matter, the
+ * sounds between words inline with their stretch, level against speech
+ * and voicing; after a word, that it is long for the take's rate, and a
+ * stretch an edits file cuts inside it.
+ */
+export declare function packTranscript(words: readonly Word[], opts?: PackOptions): string[];
+//# sourceMappingURL=transcript.d.ts.map

package/node_modules/@allyhub-next/video-core/dist/transcript.js

@@ -0,0 +1,2 @@
+/*! AllyHub. Copyright (c) 2026 AllyHub, Inc. All rights reserved. Use is subject to LICENSE.md. */
+import{frameDb as E}from"./voice.js";import{findLine as z,joinWords as G,LINE_FOUND as K,normText as N,ScriptAlignment as Q,WordStream as q}from"./words.js";const b=t=>Math.round(t*1e3)/1e3,_=t=>Math.round(t*10)/10;export function takeLevels(t){const e=Array.from(t.db,n=>-n).filter(n=>n>-100).sort((n,a)=>n-a);if(e.length===0)return{floor:-100,speech:-100,threshold:-60};const r=n=>e[Math.floor((e.length-1)*n)]??-100,u=r(.1),i=r(.95);return{floor:_(u),speech:_(i),threshold:_(Math.max(u+.3*(i-u),u+6))}}function V(t,e,r){return[Math.round(e/t),Math.max(Math.round(e/t)+1,Math.round(r/t))]}export function wordLevels(t,e){return t.map(r=>{const[u,i]=V(e.hop,r.start,r.end);let n=-120,a=0,f=0;for(let l=u;l<i&&l<e.db.length;l++){const s=E(e,l);n=Math.max(n,s),a+=s,f++}return{peak:_(n),mean:_(f?a/f:-120)}})}const X=.06;export function tightenWords(t,e){const{threshold:r}=takeLevels(e),u=e.hop,i=Math.round(X/u);let n=0;return{words:t.map(f=>{const l=Math.max(0,Math.round(f.start/u)),s=Math.min(e.db.length,Math.round(f.end/u));if(s<=l+i)return{...f};const d=g=>E(e,g)>=r;let x=!1;for(let g=l;g<s;g++)d(g)&&(x=!0);if(!x)return{...f};let o=l;for(;o+i<s&&!d(o);)o++;let c=s;for(;c>o+i&&!d(c-1);)c--;const h=o>l?b(Math.max(f.start,o*u-u/2)):f.start,M=c<s?b(Math.min(f.end,Math.max(h,c*u-u/2))):f.end;return(h!==f.start||M!==f.end)&&n++,{...f,start:h,end:M}}),moved:n}}export const WORD_SHAPE_DEFAULTS={voiced:.6,dip:10};const U=8,C=2,Y=/\p{L}/gu,Z=/\p{N}/u;function H(t){const e=N(t);return Z.test(e)?0:e.match(Y)?.length??0}const w=t=>{const e=[...t].sort((u,i)=>u-i),r=e.length>>1;return e.length%2?e[r]??0:((e[r-1]??0)+(e[r]??0))/2};export function speakingRate(t){const e=new Map;let r=0;for(const x of t){const o=H(x.text),c=x.end-x.start;o===0||!(c>0)||(e.set(o,[...e.get(o)??[],c]),r++)}if(r<U)return;let u=0,i=0,n=0,a=0,f=0;for(const[x,o]of e){const c=o.length,h=w(o);u+=c,i+=c*x,n+=c*h,a+=c*x*x,f+=c*x*h}const l=u*a-i*i;let s=l>1e-9?(u*f-i*n)/l:0,d=(n-s*i)/u;if((d<0||l<=1e-9)&&(d=0,s=a>0?f/a:0),s>0)return{base:b(d),perLetter:b(s),words:r}}export function wordShapes(t,e,r={}){const u={voiced:r.voiced??WORD_SHAPE_DEFAULTS.voiced,dip:r.dip??WORD_SHAPE_DEFAULTS.dip},i=speakingRate(t),n=e.hop,a=Math.round(u.voiced*100),f=t.map((l,s)=>{const d=Math.max(0,l.end-l.start),x=H(l.text),o=i&&x>0?b(i.base+i.perLetter*x):null,c=Math.max(0,Math.round(l.start/n)),h=Math.min(e.db.length,Math.round(l.end/n)),M=[];let g=-1;for(let p=c;p<=h;p++){const $=p<h&&(e.voicing[p]??0)>=a;$&&g<0&&(g=p),!$&&g>=0&&(p-g>=C&&M.push({start:b(g*n),end:b(p*n)}),g=-1)}const m=[],L=new Float64Array(Math.max(0,h-c));let A=-120;for(let p=h-1;p>=c;p--)A=Math.max(A,E(e,p)),L[p-c]=A;let W=-120,S=-1,k=0;for(let p=c;p<=h;p++){let $=0;p<h&&(W=Math.max(W,E(e,p)),$=Math.min(W,L[p-c]??-120)-E(e,p)),$>=u.dip?(S<0&&(S=p),k=Math.max(k,$)):S>=0&&(p-S>=C&&m.push({start:b(S*n),end:b(p*n),depth:_(k)}),S=-1,k=0)}return{i:s,dur:b(d),letters:x,expected:o,ratio:o===null?null:Math.round(d/o*100)/100,excess:o===null?null:b(d-o),voicedRuns:M,dips:m}});return{rate:i??null,...i?{}:{rateWhy:t.length<U?`the take has under ${String(U)} words with letters`:"the take's longer words do not last longer than its short ones"},options:u,words:f}}export const LONG_WORD_DEFAULTS={ratio:2,excess:.15};export function longWords(t,e={}){const r=e.ratio??LONG_WORD_DEFAULTS.ratio,u=e.excess??LONG_WORD_DEFAULTS.excess;return t.filter(i=>i.ratio!==null&&i.excess!==null&&i.ratio>=r&&i.excess>=u)}export function longLabel(t){return`(long ${t.dur.toFixed(2)}s ×${(t.ratio??0).toFixed(1)})`}export const PARAGRAPH_GAP=.5;export function paragraphs(t,e=PARAGRAPH_GAP,r=0,u=t.length-1){const i=[];let n=r;for(let a=r;a<=u;a++){const f=t[a],l=t[a+1];if(!f)break;if(a===u||!l||l.start-f.end>=e){const s=t[n];s&&i.push({from:n,to:a,start:s.start,end:f.end,text:G(t.slice(n,a+1))}),n=a+1}}return i}export function searchWords(t,e){const r=e.split(/\s+/u).map(N).filter(n=>n!=="");if(r.length===0)return[];const u=t.map(n=>N(n.text)),i=[];for(let n=0;n+r.length<=t.length;n++){let a=!0;for(let f=0;f<r.length&&a;f++){const l=u[n+f]??"",s=r[f]??"";a=f===r.length-1?l.startsWith(s):l===s}a&&i.push(y(t,n,n+r.length-1))}return i}function y(t,e,r){return{from:e,to:r,start:t[e]?.start??0,end:t[r]?.end??0,text:G(t.slice(e,r+1))}}export function phraseMatches(t,e){const r=e.map(n=>({phrase:n,norm:n.split(/\s+/u).map(N).filter(a=>a!=="")})).filter(n=>n.norm.length>0).sort((n,a)=>a.norm.length-n.norm.length),u=t.map(n=>N(n.text)),i=[];for(let n=0;n<t.length;n++)for(const a of r)if(!(n+a.norm.length>t.length)&&a.norm.every((f,l)=>u[n+l]===f)){i.push({...y(t,n,n+a.norm.length-1),phrase:a.phrase}),n+=a.norm.length-1;break}return i}export function pauseCandidates(t,e,r,u=0,i=t.length-1){const n=[];for(let a=u;a<i;a++){const f=t[a],l=t[a+1];if(!f||!l)continue;const s=l.start-f.end;if(s<e)continue;const d=Math.min(r,s)/2;n.push({after:a,start:f.end,end:l.start,gap:b(s),remove:{from:b(f.end+d),to:b(l.start-d),seconds:b(Math.max(0,s-2*d))}})}return n}export const JOIN_NEAR=.2;export function joinsOf(t,e,r,u,i=JOIN_NEAR){const{ranges:n,kept:a}=e,f=r?wordLevels(t,r):void 0,l=r?.hop??.01,s=[];let d=0;for(let x=0;x+1<n.length;x++){const o=n[x],c=n[x+1];if(!o||!c)continue;d+=o.to-o.from;let h=-1,M=-1;for(const v of a){const F=t[v];F&&(F.start<o.to&&(h=v),M<0&&F.end>c.from&&(M=v))}const g=t[h],m=t[M];if(!g||!m)continue;const L=(v,F,O)=>{if(!r)return null;const I=Math.ceil(v/l-1e-9),R=Math.ceil(F/l-1e-9)-1;let D=0;for(let T=0;T<=R-I;T++){const J=O?R-T:I+T;if(E(r,J)>=u)break;D++}return b(D*l)},A=(v,F)=>{const O=Math.ceil(v/l-1e-9),I=Math.ceil(F/l-1e-9)-1,R=[];if(!r)return R;for(let D=Math.max(0,O);D<=I&&D<r.db.length;D++)R.push(E(r,D));return R},W=(v,F)=>{const O=A(v,F);return O.length?_(Math.max(...O)):null},S=(v,F)=>{const O=A(v,Math.max(F,v+l));return O.length?_(O.reduce((I,R)=>I+R,0)/O.length):null},k=(v,F,O,I,R,D,T)=>({i:v,text:F.text,edge:O,gap:b(I),quiet:R,peakBetween:D,peak:f?.[v]?.peak??null,mean:f?.[v]?.mean??null,near:T}),p=Math.min(g.end,o.to),$=Math.max(m.start,c.from),j=k(h,g,p,o.to-p,L(o.from,o.to,!0),W(p,o.to),S(Math.max(g.start,p-i),p)),P=k(M,m,$,$-c.from,L(c.from,c.to,!1),W(c.from,$),S($,Math.min(m.end,$+i)));s.push({at:b(d),out:o.to,in:c.from,removed:M>h+1?{from:h+1,to:M-1}:null,inWord:g.end>o.to+1e-9||m.start<c.from-1e-9,before:j,after:P,step:j.near===null||P.near===null?null:_(P.near-j.near)})}return s}export const SCRIPT_DIFF_DEFAULTS={match:.5,lineMatch:K};export function scriptDiff(t,e,r={}){const u={match:r.match??SCRIPT_DIFF_DEFAULTS.match,lineMatch:r.lineMatch??SCRIPT_DIFF_DEFAULTS.lineMatch},i=new q(t),n=e.filter(o=>N(o)!==""),a=new Q(i,n),f=[];let l;for(let o=0;o<=t.length;o++){let c=!1;if(o<t.length){const h=i.charOf(o),M=i.charOf(o+1);let g=0;for(let m=h;m<M;m++)g+=a.takeHits[m]??0;c=M>h&&g/(M-h)<u.match}c&&l===void 0&&(l=o),!c&&l!==void 0&&(f.push(y(t,l,o-1)),l=void 0)}const s=(o,c,h)=>{if(h<c)return[];const M=new q(t.slice(c,h+1)),g=z(M,o,0);return!g||g.score<u.lineMatch?[]:[...s(o,c,c+g.first-1),y(t,c+g.first,c+g.last),...s(o,c+g.last+1,h)]},d=n.map((o,c)=>{const h=a.lines[c]?.length??0;return{line:c,text:o,share:h?Math.round((a.matched[c]??0)/h*1e3)/1e3:0,matches:s(o,0,t.length-1)}}),x=d.filter(o=>o.matches.length>1).map(({line:o,text:c,matches:h})=>({line:o,text:c,matches:h}));return{options:u,unscripted:f,repeated:x,lines:d}}const B=t=>t.toFixed(2).padStart(6,"0");export const LINE_WORDS=12;export function soundLabel(t,e=!1){const r=t.peak>=0?`+${String(t.peak)}`:`−${String(-t.peak)}`;return`(${e?"cut ":""}sound ${t.start.toFixed(2)}–${t.end.toFixed(2)} ${(t.end-t.start).toFixed(2)}s ${r}dB v${t.voiced.toFixed(2)})`}export function packTranscript(t,e={}){const r=Math.max(0,e.from??0),u=Math.min(t.length-1,e.to??t.length-1),i=e.sounds??[],n=e.lineWords??LINE_WORDS,a=s=>(e.removedSounds??[]).some(d=>d.from<s.end&&d.to>s.start)||e.removed?.has(s.after)===!0&&e.removed.has(s.after+1),f=[],l=[];for(const s of paragraphs(t,e.gap??PARAGRAPH_GAP,r,u)){let d=s.from;for(let x=s.from;x<=s.to;x++){const o=/[.?!。？！]["')\]]*$/u.test(t[x]?.text??"");(x===s.to||o&&x-d+1>=n)&&(l.push({from:d,to:x}),d=x+1)}}for(const s of l){const d=[],x=i.filter(m=>m.after===s.from-1&&s.from===r);for(const m of x)d.push(soundLabel(m,a(m)));let o,c;const h=()=>{o&&d.push(`[-${o.join(" ")}-]`),c&&d.push(`{${c.join(" ")}}`),o=void 0,c=void 0};for(let m=s.from;m<=s.to;m++){const L=t[m]?.text??"",A=e.unscripted?.has(m)===!0;e.removed?.has(m)===!0?(c&&h(),(o??=[]).push(A?`{${L}}`:L)):A?(o&&h(),(c??=[]).push(L)):(h(),d.push(L));const W=t[m],S=e.long?.get(m);S&&e.removed?.has(m)!==!0&&(h(),d.push(longLabel(S)));for(const k of e.inWord??[]){if(!W||e.removed?.has(m)===!0)continue;const p=t[m-1];k.from<W.end&&k.to>W.start&&!(p&&m>s.from&&k.from<p.end&&k.to>p.start)&&(h(),d.push(`(cut ${k.from.toFixed(2)}–${k.to.toFixed(2)} in word)`))}(m-s.from)%8===7&&m<s.to&&(h(),d.push(`#${String(m+1)}`));for(const k of i.filter(p=>p.after===m))h(),d.push(soundLabel(k,a(k)))}h();const M=t[s.from],g=t[s.to];f.push(`[${B(M?.start??0)}-${B(g?.end??0)}] #${String(s.from)}-#${String(s.to)} ${d.join(" ")}`)}return f}