@storyteller-platform/align 0.1.48 → 0.1.49

package/dist/align/align.cjs

@@ -689,17 +689,22 @@ class Aligner {
     });
     const sentenceRanges = [];
     const chapterSentenceCounts = {};
+    const sentenceLengths = {};
     for (const alignedChapter of audioOrderedChapters) {
       sentenceRanges.push(...alignedChapter.sentenceRanges);
       const sentences = await this.getChapterSentences(
         alignedChapter.chapter.id
       );
       chapterSentenceCounts[alignedChapter.chapter.id] = sentences.length;
+      for (const [id, sentence] of (0, import_itertools.enumerate)(sentences)) {
+        sentenceLengths[(0, import_interpolateSentenceRanges.slotKey)({ chapterId: alignedChapter.chapter.id, id })] = sentence.text.length;
+      }
     }
     const interpolated = (0, import_interpolateSentenceRanges.interpolateSentenceRanges)(
       sentenceRanges,
       chapterSentenceCounts,
-      this.audioFileDurations
+      this.audioFileDurations,
+      sentenceLengths
     );
     const expanded = (0, import_getSentenceRanges.expandEmptySentenceRanges)(interpolated);
     const collapsed = await (0, import_getSentenceRanges.collapseSentenceRangeGaps)(expanded);

package/dist/align/align.js

@@ -43,7 +43,10 @@ import {
   getSentenceRanges,
   mapTranscriptionTimeline
 } from "./getSentenceRanges.js";
-import { interpolateSentenceRanges } from "./interpolateSentenceRanges.js";
+import {
+  interpolateSentenceRanges,
+  slotKey
+} from "./interpolateSentenceRanges.js";
 import { findBoundaries } from "./search.js";
 import { slugify } from "./slugify.js";
 import { TextFragmentFactory } from "./textFragments.js";
@@ -635,17 +638,22 @@ class Aligner {
     });
     const sentenceRanges = [];
     const chapterSentenceCounts = {};
+    const sentenceLengths = {};
     for (const alignedChapter of audioOrderedChapters) {
       sentenceRanges.push(...alignedChapter.sentenceRanges);
       const sentences = await this.getChapterSentences(
         alignedChapter.chapter.id
       );
       chapterSentenceCounts[alignedChapter.chapter.id] = sentences.length;
+      for (const [id, sentence] of enumerate(sentences)) {
+        sentenceLengths[slotKey({ chapterId: alignedChapter.chapter.id, id })] = sentence.text.length;
+      }
     }
     const interpolated = interpolateSentenceRanges(
       sentenceRanges,
       chapterSentenceCounts,
-      this.audioFileDurations
+      this.audioFileDurations,
+      sentenceLengths
     );
     const expanded = expandEmptySentenceRanges(interpolated);
     const collapsed = await collapseSentenceRangeGaps(expanded);

package/dist/align/interpolateSentenceRanges.cjs

@@ -18,55 +18,76 @@ var __copyProps = (to, from, except, desc) => {
 var __toCommonJS = (mod) => __copyProps(__defProp({}, "__esModule", { value: true }), mod);
 var interpolateSentenceRanges_exports = {};
 __export(interpolateSentenceRanges_exports, {
-  interpolateSentenceRanges: () => interpolateSentenceRanges
+  interpolateSentenceRanges: () => interpolateSentenceRanges,
+  slotKey: () => slotKey
 });
 module.exports = __toCommonJS(interpolateSentenceRanges_exports);
-function buildGapRanges(slots, left, right, audioFileDurations) {
+function slotKey(slot) {
+  return `${slot.chapterId}\0${slot.id}`;
+}
+function slotWeight(slot, sentenceLengths) {
+  const length = sentenceLengths[slotKey(slot)];
+  return length && length > 0 ? length : 1;
+}
+function buildGapRanges(slots, left, right, audioFileDurations, sentenceLengths) {
   const n = slots.length;
   if (n === 0) return [];
+  const weights = slots.map((slot) => slotWeight(slot, sentenceLengths));
   if (left.audiofile === right.audiofile) {
     const span = right.time - left.time;
-    return slots.map((slot, i) => ({
-      ...slot,
-      audiofile: left.audiofile,
-      start: left.time + span * i / n,
-      end: left.time + span * (i + 1) / n
-    }));
+    const totalWeight2 = weights.reduce((a, b) => a + b, 0);
+    const result2 = [];
+    let cursor = left.time;
+    for (const [i, slot] of slots.entries()) {
+      const start = cursor;
+      const end = start + span * weights[i] / totalWeight2;
+      result2.push({ ...slot, audiofile: left.audiofile, start, end });
+      cursor = end;
+    }
+    return result2;
   }
   const leftDuration = audioFileDurations[left.audiofile] ?? left.time;
   const leftAvail = leftDuration - left.time;
   const rightAvail = right.time;
   const total = leftAvail + rightAvail;
-  let n1 = total > 0 ? Math.round(n * (leftAvail / total)) : n;
-  let n2 = n - n1;
-  n1 = Math.max(0, n1);
-  n2 = n - n1;
+  const totalWeight = weights.reduce((a, b) => a + b, 0);
+  const leftWeightShare = total > 0 ? totalWeight * (leftAvail / total) : totalWeight;
+  let n1 = 0;
+  let accumulatedWeight = 0;
+  while (n1 < n && // eslint-disable-next-line @typescript-eslint/no-non-null-assertion
+  accumulatedWeight + weights[n1] / 2 <= leftWeightShare) {
+    accumulatedWeight += weights[n1];
+    n1++;
+  }
+  const n2 = n - n1;
   const result = [];
   if (n1 > 0) {
-    for (let i = 0; i < n1; i++) {
-      result.push({
-        // eslint-disable-next-line @typescript-eslint/no-non-null-assertion
-        ...slots[i],
-        audiofile: left.audiofile,
-        start: left.time + leftAvail * i / n1,
-        end: left.time + leftAvail * (i + 1) / n1
-      });
+    const leftSlots = slots.slice(0, n1);
+    const leftWeights = weights.slice(0, n1);
+    const leftTotalWeight = leftWeights.reduce((a, b) => a + b, 0);
+    let cursor = left.time;
+    for (const [i, slot] of leftSlots.entries()) {
+      const start = cursor;
+      const end = start + leftAvail * leftWeights[i] / leftTotalWeight;
+      result.push({ ...slot, audiofile: left.audiofile, start, end });
+      cursor = end;
     }
   }
   if (n2 > 0) {
-    for (let i = 0; i < n2; i++) {
-      result.push({
-        // eslint-disable-next-line @typescript-eslint/no-non-null-assertion
-        ...slots[n1 + i],
-        audiofile: right.audiofile,
-        start: rightAvail * i / n2,
-        end: rightAvail * (i + 1) / n2
-      });
+    const rightSlots = slots.slice(n1);
+    const rightWeights = weights.slice(n1);
+    const rightTotalWeight = rightWeights.reduce((a, b) => a + b, 0);
+    let cursor = 0;
+    for (const [i, slot] of rightSlots.entries()) {
+      const start = cursor;
+      const end = start + rightAvail * rightWeights[i] / rightTotalWeight;
+      result.push({ ...slot, audiofile: right.audiofile, start, end });
+      cursor = end;
     }
   }
   return result;
 }
-function interpolateSentenceRanges(sentenceRanges, chapterSentenceCounts, audioFileDurations) {
+function interpolateSentenceRanges(sentenceRanges, chapterSentenceCounts, audioFileDurations, sentenceLengths = {}) {
   if (sentenceRanges.length === 0) return [];
   const result = [];
   const first = sentenceRanges[0];
@@ -77,7 +98,15 @@ function interpolateSentenceRanges(sentenceRanges, chapterSentenceCounts, audioF
     }));
     const left = { time: 0, audiofile: first.audiofile };
     const right = { time: first.start, audiofile: first.audiofile };
-    result.push(...buildGapRanges(slots, left, right, audioFileDurations));
+    result.push(
+      ...buildGapRanges(
+        slots,
+        left,
+        right,
+        audioFileDurations,
+        sentenceLengths
+      )
+    );
   }
   result.push(first);
   for (let idx = 1; idx < sentenceRanges.length; idx++) {
@@ -100,7 +129,15 @@ function interpolateSentenceRanges(sentenceRanges, chapterSentenceCounts, audioF
       }
     }
     if (gapSlots.length > 0) {
-      result.push(...buildGapRanges(gapSlots, left, right, audioFileDurations));
+      result.push(
+        ...buildGapRanges(
+          gapSlots,
+          left,
+          right,
+          audioFileDurations,
+          sentenceLengths
+        )
+      );
     }
     result.push(curr);
   }
@@ -114,11 +151,20 @@ function interpolateSentenceRanges(sentenceRanges, chapterSentenceCounts, audioF
     const fileEnd = audioFileDurations[last.audiofile] ?? last.end;
     const left = { time: last.end, audiofile: last.audiofile };
     const right = { time: fileEnd, audiofile: last.audiofile };
-    result.push(...buildGapRanges(slots, left, right, audioFileDurations));
+    result.push(
+      ...buildGapRanges(
+        slots,
+        left,
+        right,
+        audioFileDurations,
+        sentenceLengths
+      )
+    );
   }
   return result;
 }
 // Annotate the CommonJS export names for ESM import in node:
 0 && (module.exports = {
-  interpolateSentenceRanges
+  interpolateSentenceRanges,
+  slotKey
 });

package/dist/align/interpolateSentenceRanges.d.cts

@@ -3,6 +3,19 @@ import '@storyteller-platform/ghost-story';
 import '@echogarden/text-segmentation';
 import '@storyteller-platform/transliteration';
 
+/** A "slot" is a sentence that needs to be synthesised. */
+type Slot = {
+    chapterId: string;
+    id: number;
+};
+/**
+ * Maps a slot to the character length of its sentence text, used to weight
+ * how much of a gap's time it should occupy. Keyed by `${chapterId}\0${id}`.
+ * Slots missing from this map (or with a non-positive length) fall back to
+ * a weight of 1, i.e. even division, same as before this map existed.
+ */
+type SentenceLengths = Record<string, number>;
+declare function slotKey(slot: Slot): string;
 /**
  * Given a sequence of sentence ranges from an entire book,
  * ordered by occurrence in audio, interpolates sentence ranges
@@ -18,6 +31,6 @@ import '@storyteller-platform/transliteration';
  *     e.g. chapter001#325 -> chapter002#0, where
  *     chapterSentenceCounts["chapter001"] === 330
  */
-declare function interpolateSentenceRanges(sentenceRanges: SentenceRange[], chapterSentenceCounts: Record<string, number>, audioFileDurations: Record<string, number>): SentenceRange[];
+declare function interpolateSentenceRanges(sentenceRanges: SentenceRange[], chapterSentenceCounts: Record<string, number>, audioFileDurations: Record<string, number>, sentenceLengths?: SentenceLengths): SentenceRange[];
 
-export { interpolateSentenceRanges };
+export { type SentenceLengths, interpolateSentenceRanges, slotKey };

package/dist/align/interpolateSentenceRanges.d.ts

@@ -3,6 +3,19 @@ import '@storyteller-platform/ghost-story';
 import '@echogarden/text-segmentation';
 import '@storyteller-platform/transliteration';
 
+/** A "slot" is a sentence that needs to be synthesised. */
+type Slot = {
+    chapterId: string;
+    id: number;
+};
+/**
+ * Maps a slot to the character length of its sentence text, used to weight
+ * how much of a gap's time it should occupy. Keyed by `${chapterId}\0${id}`.
+ * Slots missing from this map (or with a non-positive length) fall back to
+ * a weight of 1, i.e. even division, same as before this map existed.
+ */
+type SentenceLengths = Record<string, number>;
+declare function slotKey(slot: Slot): string;
 /**
  * Given a sequence of sentence ranges from an entire book,
  * ordered by occurrence in audio, interpolates sentence ranges
@@ -18,6 +31,6 @@ import '@storyteller-platform/transliteration';
  *     e.g. chapter001#325 -> chapter002#0, where
  *     chapterSentenceCounts["chapter001"] === 330
  */
-declare function interpolateSentenceRanges(sentenceRanges: SentenceRange[], chapterSentenceCounts: Record<string, number>, audioFileDurations: Record<string, number>): SentenceRange[];
+declare function interpolateSentenceRanges(sentenceRanges: SentenceRange[], chapterSentenceCounts: Record<string, number>, audioFileDurations: Record<string, number>, sentenceLengths?: SentenceLengths): SentenceRange[];
 
-export { interpolateSentenceRanges };
+export { type SentenceLengths, interpolateSentenceRanges, slotKey };

package/dist/align/interpolateSentenceRanges.js

@@ -1,50 +1,70 @@
 import "../chunk-BIEQXUOY.js";
-function buildGapRanges(slots, left, right, audioFileDurations) {
+function slotKey(slot) {
+  return `${slot.chapterId}\0${slot.id}`;
+}
+function slotWeight(slot, sentenceLengths) {
+  const length = sentenceLengths[slotKey(slot)];
+  return length && length > 0 ? length : 1;
+}
+function buildGapRanges(slots, left, right, audioFileDurations, sentenceLengths) {
   const n = slots.length;
   if (n === 0) return [];
+  const weights = slots.map((slot) => slotWeight(slot, sentenceLengths));
   if (left.audiofile === right.audiofile) {
     const span = right.time - left.time;
-    return slots.map((slot, i) => ({
-      ...slot,
-      audiofile: left.audiofile,
-      start: left.time + span * i / n,
-      end: left.time + span * (i + 1) / n
-    }));
+    const totalWeight2 = weights.reduce((a, b) => a + b, 0);
+    const result2 = [];
+    let cursor = left.time;
+    for (const [i, slot] of slots.entries()) {
+      const start = cursor;
+      const end = start + span * weights[i] / totalWeight2;
+      result2.push({ ...slot, audiofile: left.audiofile, start, end });
+      cursor = end;
+    }
+    return result2;
   }
   const leftDuration = audioFileDurations[left.audiofile] ?? left.time;
   const leftAvail = leftDuration - left.time;
   const rightAvail = right.time;
   const total = leftAvail + rightAvail;
-  let n1 = total > 0 ? Math.round(n * (leftAvail / total)) : n;
-  let n2 = n - n1;
-  n1 = Math.max(0, n1);
-  n2 = n - n1;
+  const totalWeight = weights.reduce((a, b) => a + b, 0);
+  const leftWeightShare = total > 0 ? totalWeight * (leftAvail / total) : totalWeight;
+  let n1 = 0;
+  let accumulatedWeight = 0;
+  while (n1 < n && // eslint-disable-next-line @typescript-eslint/no-non-null-assertion
+  accumulatedWeight + weights[n1] / 2 <= leftWeightShare) {
+    accumulatedWeight += weights[n1];
+    n1++;
+  }
+  const n2 = n - n1;
   const result = [];
   if (n1 > 0) {
-    for (let i = 0; i < n1; i++) {
-      result.push({
-        // eslint-disable-next-line @typescript-eslint/no-non-null-assertion
-        ...slots[i],
-        audiofile: left.audiofile,
-        start: left.time + leftAvail * i / n1,
-        end: left.time + leftAvail * (i + 1) / n1
-      });
+    const leftSlots = slots.slice(0, n1);
+    const leftWeights = weights.slice(0, n1);
+    const leftTotalWeight = leftWeights.reduce((a, b) => a + b, 0);
+    let cursor = left.time;
+    for (const [i, slot] of leftSlots.entries()) {
+      const start = cursor;
+      const end = start + leftAvail * leftWeights[i] / leftTotalWeight;
+      result.push({ ...slot, audiofile: left.audiofile, start, end });
+      cursor = end;
     }
   }
   if (n2 > 0) {
-    for (let i = 0; i < n2; i++) {
-      result.push({
-        // eslint-disable-next-line @typescript-eslint/no-non-null-assertion
-        ...slots[n1 + i],
-        audiofile: right.audiofile,
-        start: rightAvail * i / n2,
-        end: rightAvail * (i + 1) / n2
-      });
+    const rightSlots = slots.slice(n1);
+    const rightWeights = weights.slice(n1);
+    const rightTotalWeight = rightWeights.reduce((a, b) => a + b, 0);
+    let cursor = 0;
+    for (const [i, slot] of rightSlots.entries()) {
+      const start = cursor;
+      const end = start + rightAvail * rightWeights[i] / rightTotalWeight;
+      result.push({ ...slot, audiofile: right.audiofile, start, end });
+      cursor = end;
     }
   }
   return result;
 }
-function interpolateSentenceRanges(sentenceRanges, chapterSentenceCounts, audioFileDurations) {
+function interpolateSentenceRanges(sentenceRanges, chapterSentenceCounts, audioFileDurations, sentenceLengths = {}) {
   if (sentenceRanges.length === 0) return [];
   const result = [];
   const first = sentenceRanges[0];
@@ -55,7 +75,15 @@ function interpolateSentenceRanges(sentenceRanges, chapterSentenceCounts, audioF
     }));
     const left = { time: 0, audiofile: first.audiofile };
     const right = { time: first.start, audiofile: first.audiofile };
-    result.push(...buildGapRanges(slots, left, right, audioFileDurations));
+    result.push(
+      ...buildGapRanges(
+        slots,
+        left,
+        right,
+        audioFileDurations,
+        sentenceLengths
+      )
+    );
   }
   result.push(first);
   for (let idx = 1; idx < sentenceRanges.length; idx++) {
@@ -78,7 +106,15 @@ function interpolateSentenceRanges(sentenceRanges, chapterSentenceCounts, audioF
       }
     }
     if (gapSlots.length > 0) {
-      result.push(...buildGapRanges(gapSlots, left, right, audioFileDurations));
+      result.push(
+        ...buildGapRanges(
+          gapSlots,
+          left,
+          right,
+          audioFileDurations,
+          sentenceLengths
+        )
+      );
     }
     result.push(curr);
   }
@@ -92,10 +128,19 @@ function interpolateSentenceRanges(sentenceRanges, chapterSentenceCounts, audioF
     const fileEnd = audioFileDurations[last.audiofile] ?? last.end;
     const left = { time: last.end, audiofile: last.audiofile };
     const right = { time: fileEnd, audiofile: last.audiofile };
-    result.push(...buildGapRanges(slots, left, right, audioFileDurations));
+    result.push(
+      ...buildGapRanges(
+        slots,
+        left,
+        right,
+        audioFileDurations,
+        sentenceLengths
+      )
+    );
   }
   return result;
 }
 export {
-  interpolateSentenceRanges
+  interpolateSentenceRanges,
+  slotKey
 };

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@storyteller-platform/align",
-  "version": "0.1.48",
+  "version": "0.1.49",
   "description": "A library and CLI for automatically aligning audiobooks and EPUBs to produce Media Overlays",
   "author": "Shane Friedman",
   "license": "MIT",