@storyteller-platform/align 0.1.26 → 0.1.27

package/dist/align/getSentenceRanges.cjs

@@ -26,6 +26,7 @@ __export(getSentenceRanges_exports, {
   mapTranscriptionTimeline: () => mapTranscriptionTimeline
 });
 module.exports = __toCommonJS(getSentenceRanges_exports);
+var import_fastest_levenshtein = require("fastest-levenshtein");
 var import_itertools = require("itertools");
 var import_runes2 = require("runes2");
 var import_ffmpeg = require("../common/ffmpeg.cjs");
@@ -66,6 +67,9 @@ function getAlignmentsForSentence(sentence, alignments) {
     if (alignment.opType === "DELETE" || alignment.opType === "INSERT" && sentenceIndex > 0) {
       score -= (alignment.ref ?? alignment.hyp).length + 1;
     }
+    if (alignment.opType === "SUBSTITUTE") {
+      score -= (0, import_fastest_levenshtein.distance)(alignment.ref, alignment.hyp);
+    }
     result.push(alignment);
   }
   return {
@@ -210,6 +214,8 @@ async function getSentenceRanges(transcriptionText, mappedTimeline, sentences, c
     }
     let alignmentIndex = 0;
     let currentTranscriptWindowIndex = 0;
+    let lastGoodSentenceIndex = slice[0] - 1;
+    let lastGoodTranscriptWindowIndex = -1;
     for (const [j, slugifiedSentence] of (0, import_itertools.enumerate)(
       slugifiedChapterSentenceWindowList
     )) {
@@ -220,6 +226,7 @@ async function getSentenceRanges(transcriptionText, mappedTimeline, sentences, c
       );
       const sentenceLengthInSlugifiedTranscript = sentenceAlignments.filter((a) => a.opType !== "DELETE").map((a) => a.hyp).join("-").length;
       if (score > 0) {
+        lastGoodSentenceIndex = j + slice[0];
         const start = findStartTimestamp(
           chapterOffset + slugifiedChapterTranscriptWindowStartIndex + currentTranscriptWindowIndex,
           mappedTimeline
@@ -295,9 +302,21 @@ async function getSentenceRanges(transcriptionText, mappedTimeline, sentences, c
       if (slugifiedChapterTranscriptWindow[currentTranscriptWindowIndex] === "-") {
         currentTranscriptWindowIndex++;
       }
+      if (score > 0) {
+        lastGoodTranscriptWindowIndex = currentTranscriptWindowIndex;
+      }
+    }
+    if (lastGoodSentenceIndex === -1) {
+      return {
+        sentenceRanges,
+        wordRanges,
+        transcriptionOffset: chapterTranscriptEndIndex,
+        firstFoundSentence,
+        lastFoundSentence: chapterSentenceIndex - 1
+      };
     }
-    chapterSentenceIndex = i;
-    slugifiedChapterTranscriptWindowStartIndex += currentTranscriptWindowIndex;
+    chapterSentenceIndex += lastGoodSentenceIndex + 1;
+    slugifiedChapterTranscriptWindowStartIndex += lastGoodTranscriptWindowIndex;
     if (slugifiedChapterTranscript[slugifiedChapterTranscriptWindowStartIndex] === "-") {
       slugifiedChapterTranscriptWindowStartIndex++;
     }

package/dist/align/getSentenceRanges.js

@@ -1,4 +1,5 @@
 import "../chunk-BIEQXUOY.js";
+import { distance } from "fastest-levenshtein";
 import { enumerate } from "itertools";
 import { runes } from "runes2";
 import { getTrackDuration } from "../common/ffmpeg.js";
@@ -39,6 +40,9 @@ function getAlignmentsForSentence(sentence, alignments) {
     if (alignment.opType === "DELETE" || alignment.opType === "INSERT" && sentenceIndex > 0) {
       score -= (alignment.ref ?? alignment.hyp).length + 1;
     }
+    if (alignment.opType === "SUBSTITUTE") {
+      score -= distance(alignment.ref, alignment.hyp);
+    }
     result.push(alignment);
   }
   return {
@@ -183,6 +187,8 @@ async function getSentenceRanges(transcriptionText, mappedTimeline, sentences, c
     }
     let alignmentIndex = 0;
     let currentTranscriptWindowIndex = 0;
+    let lastGoodSentenceIndex = slice[0] - 1;
+    let lastGoodTranscriptWindowIndex = -1;
     for (const [j, slugifiedSentence] of enumerate(
       slugifiedChapterSentenceWindowList
     )) {
@@ -193,6 +199,7 @@ async function getSentenceRanges(transcriptionText, mappedTimeline, sentences, c
       );
       const sentenceLengthInSlugifiedTranscript = sentenceAlignments.filter((a) => a.opType !== "DELETE").map((a) => a.hyp).join("-").length;
       if (score > 0) {
+        lastGoodSentenceIndex = j + slice[0];
         const start = findStartTimestamp(
           chapterOffset + slugifiedChapterTranscriptWindowStartIndex + currentTranscriptWindowIndex,
           mappedTimeline
@@ -268,9 +275,21 @@ async function getSentenceRanges(transcriptionText, mappedTimeline, sentences, c
       if (slugifiedChapterTranscriptWindow[currentTranscriptWindowIndex] === "-") {
         currentTranscriptWindowIndex++;
       }
+      if (score > 0) {
+        lastGoodTranscriptWindowIndex = currentTranscriptWindowIndex;
+      }
+    }
+    if (lastGoodSentenceIndex === -1) {
+      return {
+        sentenceRanges,
+        wordRanges,
+        transcriptionOffset: chapterTranscriptEndIndex,
+        firstFoundSentence,
+        lastFoundSentence: chapterSentenceIndex - 1
+      };
     }
-    chapterSentenceIndex = i;
-    slugifiedChapterTranscriptWindowStartIndex += currentTranscriptWindowIndex;
+    chapterSentenceIndex += lastGoodSentenceIndex + 1;
+    slugifiedChapterTranscriptWindowStartIndex += lastGoodTranscriptWindowIndex;
     if (slugifiedChapterTranscript[slugifiedChapterTranscriptWindowStartIndex] === "-") {
       slugifiedChapterTranscriptWindowStartIndex++;
     }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@storyteller-platform/align",
-  "version": "0.1.26",
+  "version": "0.1.27",
   "description": "A library and CLI for automatically aligning audiobooks and EPUBs to produce Media Overlays",
   "author": "Shane Friedman",
   "license": "MIT",
@@ -66,6 +66,7 @@
     "chalk": "^5.4.1",
     "cli-progress": "^3.12.0",
     "esbuild": "^0.27.3",
+    "fastest-levenshtein": "^1.0.16",
     "itertools": "^2.6.0",
     "locale-currency": "^1.0.0",
     "memoize": "^10.2.0",