@storyteller-platform/align 0.1.13 → 0.1.15

package/dist/align/align.cjs

@@ -81,16 +81,14 @@ module.exports = __toCommonJS(align_exports);
 var import_promises = require("node:fs/promises");
 var import_node_path = require("node:path");
 var import_posix = require("node:path/posix");
-var import_itertools = require("itertools");
 var import_memoize = __toESM(require("memoize"), 1);
-var import_runes2 = require("runes2");
 var import_audiobook = require("@storyteller-platform/audiobook");
 var import_epub = require("@storyteller-platform/epub");
 var import_ghost_story = require("@storyteller-platform/ghost-story");
 var import_ffmpeg = require("../common/ffmpeg.cjs");
 var import_segmentation = require("../markup/segmentation.cjs");
-var import_fuzzy = require("./fuzzy.cjs");
 var import_getSentenceRanges = require("./getSentenceRanges.cjs");
+var import_search = require("./search.cjs");
 var import_slugify = require("./slugify.cjs");
 async function align(input, output, transcriptionsDir, audiobookDir, options) {
   var _stack = [];
@@ -167,83 +165,6 @@ class Aligner {
   report = {
     chapters: []
   };
-  findBestOffset(epubSentences, transcriptionText, lastMatchOffset, dir = 1) {
-    const reverse = dir < 0;
-    if (dir < 0) {
-      epubSentences = epubSentences.toReversed().map((s) => (0, import_runes2.runes)(s).toReversed().join(""));
-      transcriptionText = (0, import_runes2.runes)(transcriptionText).toReversed().join("");
-      lastMatchOffset = transcriptionText.length - lastMatchOffset;
-    }
-    const flatSliceIndices = [
-      0,
-      ...this.alignedChapters.toSorted(
-        (a, b) => reverse ? transcriptionText.length - a.endOffset - (transcriptionText.length - b.endOffset) : a.startOffset - b.startOffset
-      ).flatMap((aligned) => [
-        reverse ? transcriptionText.length - aligned.endOffset : aligned.startOffset,
-        reverse ? transcriptionText.length - aligned.startOffset : aligned.endOffset
-      ]),
-      transcriptionText.length
-    ];
-    const sliceIndices = [];
-    for (let i = 0; i < flatSliceIndices.length - 1; i += 2) {
-      sliceIndices.push([flatSliceIndices[i], flatSliceIndices[i + 1]]);
-    }
-    const allSlices = [];
-    let startSlice = 0;
-    for (const [i, [start, end]] of (0, import_itertools.enumerate)(sliceIndices)) {
-      if (lastMatchOffset >= start && lastMatchOffset < end) {
-        if (!reverse) {
-          startSlice = i + 1;
-          allSlices.push({
-            start,
-            text: transcriptionText.slice(start, lastMatchOffset)
-          });
-        }
-        allSlices.push({
-          start: lastMatchOffset,
-          text: transcriptionText.slice(lastMatchOffset, end)
-        });
-      } else if (!reverse) {
-        allSlices.push({ start, text: transcriptionText.slice(start, end) });
-      }
-    }
-    const slices = allSlices.filter((slice) => slice.text.length);
-    if (reverse && !slices.length) {
-      const indices = sliceIndices.find(([start]) => start > lastMatchOffset);
-      if (indices) {
-        slices.push({
-          start: indices[0],
-          text: transcriptionText.slice(...indices)
-        });
-      }
-    }
-    for (const slice of slices.slice(startSlice).concat(slices.slice(0, startSlice))) {
-      let startSentence = 0;
-      while (startSentence < epubSentences.length) {
-        const needle = epubSentences.slice(startSentence, startSentence + 6).join("-");
-        const firstMatch = (0, import_fuzzy.findNearestMatch)(
-          needle,
-          slice.text,
-          Math.max(Math.floor(0.1 * needle.length), 1)
-        );
-        if (firstMatch) {
-          const start = reverse ? transcriptionText.length - (slice.start + firstMatch.index) : slice.start + firstMatch.index;
-          return {
-            startSentence: reverse ? epubSentences.length - startSentence : startSentence,
-            transcriptionOffset: start
-          };
-        }
-        startSentence += 3;
-      }
-    }
-    if (reverse) {
-      return {
-        startSentence: epubSentences.length,
-        transcriptionOffset: slices[0] ? transcriptionText.length - slices[0].start : null
-      };
-    }
-    return { startSentence: 0, transcriptionOffset: null };
-  }
   async getChapterSentences(chapterId) {
     const chapterXml = await this.epub.readXhtmlItemContents(chapterId);
     const { result: segmentation } = await (0, import_segmentation.getXhtmlSegmentation)(
@@ -306,7 +227,7 @@ class Aligner {
       value: import_epub.Epub.formatSmilDuration(chapterDuration)
     });
   }
-  addChapterReport(chapter, chapterSentences, sentenceRanges, startSentence, transcriptionOffset) {
+  addChapterReport(chapter, chapterSentences, sentenceRanges, startSentence, endSentence, transcriptionOffset) {
     this.report.chapters.push({
       href: chapter.href,
       transcriptionOffset,
@@ -330,6 +251,14 @@ class Aligner {
         matchedSentence: chapterSentences[startSentence],
         nextSentence: chapterSentences[startSentence + 1] ?? null
       },
+      lastMatchedSentenceId: endSentence,
+      lastMatchedSentenceContext: {
+        prevSentence: chapterSentences[endSentence - 1] ?? null,
+        // eslint-disable-next-line @typescript-eslint/no-non-null-assertion
+        matchedSentence: chapterSentences[endSentence],
+        nextSentence: chapterSentences[endSentence + 1] ?? null
+      },
+      chapterSentenceCount: sentenceRanges.length,
       audioFiles: sentenceRanges.reduce((acc, range) => {
         const existing = acc.find(
           (context) => context.filepath === range.audiofile
@@ -347,7 +276,7 @@ class Aligner {
       }, [])
     });
   }
-  async alignChapter(startSentence, endSentence, chapterId, transcriptionOffset, transcriptionEndOffset, locale, mapping) {
+  async alignChapter(chapterId, transcriptionOffset, transcriptionEndOffset, locale, mapping) {
     const timing = (0, import_ghost_story.createTiming)();
     timing.start("read contents");
     const manifest = await this.epub.getManifest();
@@ -362,9 +291,12 @@ class Aligner {
     const chapterSentences = await this.getChapterSentences(chapterId);
     timing.end("split to sentences");
     timing.start("align sentences");
-    const { sentenceRanges, transcriptionOffset: endTranscriptionOffset } = await (0, import_getSentenceRanges.getSentenceRanges)(
-      startSentence,
-      endSentence,
+    const {
+      sentenceRanges,
+      transcriptionOffset: endTranscriptionOffset,
+      firstFoundSentence,
+      lastFoundSentence
+    } = await (0, import_getSentenceRanges.getSentenceRanges)(
       this.transcription,
       chapterSentences,
       transcriptionOffset,
@@ -392,7 +324,8 @@ class Aligner {
       chapter,
       chapterSentences,
       sentenceRanges,
-      startSentence,
+      firstFoundSentence,
+      lastFoundSentence,
       transcriptionOffset
     );
     return {
@@ -401,8 +334,20 @@ class Aligner {
       timing
     };
   }
+  narrowToAvailableBoundary(boundary) {
+    const narrowed = { ...boundary };
+    for (const chapter of this.alignedChapters) {
+      if (chapter.startOffset > narrowed.start && chapter.startOffset <= narrowed.end) {
+        narrowed.end = chapter.startOffset - 1;
+      }
+      if (chapter.endOffset < narrowed.end && chapter.endOffset >= narrowed.start) {
+        narrowed.start = chapter.endOffset + 1;
+      }
+    }
+    return narrowed;
+  }
   async alignBook(onProgress) {
-    var _a, _b, _c, _d, _e, _f, _g, _h;
+    var _a, _b, _c, _d, _e, _f;
     const locale = this.languageOverride ?? await this.epub.getLanguage() ?? new Intl.Locale("en-US");
     this.timing.setMetadata("language", locale.toString());
     this.timing.setMetadata("granularity", this.granularity);
@@ -412,7 +357,6 @@ class Aligner {
       this.transcription.transcript,
       locale
     );
-    let lastTranscriptionOffset = 0;
     for (let index = 0; index < spine.length; index++) {
       onProgress == null ? void 0 : onProgress(index / spine.length);
       const spineItem = spine[index];
@@ -441,48 +385,29 @@ class Aligner {
         );
         continue;
       }
-      const { startSentence, transcriptionOffset: slugifiedOffset } = this.findBestOffset(
-        slugifiedChapterSentences,
-        transcriptionText,
-        mapping.map(lastTranscriptionOffset, -1)
+      const boundaries = (0, import_search.findBoundaries)(
+        slugifiedChapterSentences.join("-"),
+        transcriptionText
       );
-      if (slugifiedOffset === null) {
+      if (!boundaries) {
         (_f = this.logger) == null ? void 0 : _f.info(
-          `Couldn't find matching transcription for chapter #${index}`
+          `Could not find chapter #${index} in the transcripton`
         );
         continue;
       }
-      const transcriptionOffset = mapping.invert().map(slugifiedOffset, -1);
-      const {
-        startSentence: startEndSentence,
-        transcriptionOffset: slugifiedEndOffset
-      } = this.findBestOffset(
-        slugifiedChapterSentences,
-        transcriptionText,
-        Math.min(
-          transcriptionText.length,
-          slugifiedOffset + Math.round(slugifiedChapterSentences.join("-").length * 1.2)
-        ),
-        -1
-      );
-      const endSentence = startEndSentence;
-      const endOffset = slugifiedEndOffset === null ? this.transcription.transcript.length : mapping.invert().map(slugifiedEndOffset, 1);
-      if (endSentence - startSentence < slugifiedChapterSentences.length / 2) {
-        (_g = this.logger) == null ? void 0 : _g.info(`Found less than half of chapter #${index}, skipping`);
+      const { start, end } = this.narrowToAvailableBoundary(boundaries);
+      if (start === end) {
+        continue;
       }
-      (_h = this.logger) == null ? void 0 : _h.info(
-        `Chapter #${index} best matches transcription from ${transcriptionOffset} to ${endOffset}, from sentence ${startSentence} to ${endSentence} (of ${slugifiedChapterSentences.length}) in the book`
-      );
+      const transcriptionOffset = mapping.invert().map(Math.max(start, 0), -1);
+      const endOffset = mapping.invert().map(Math.min(end, transcriptionText.length), 1);
       const result = await this.alignChapter(
-        startSentence,
-        endSentence,
         chapterId,
         transcriptionOffset,
         endOffset,
         locale,
         mapping
       );
-      lastTranscriptionOffset = result.endTranscriptionOffset;
       this.timing.add(result.timing.summary());
     }
     const audioOrderedChapters = this.alignedChapters.toSorted((a, b) => {

package/dist/align/align.d.cts

@@ -22,6 +22,13 @@ interface ChapterReport {
         matchedSentence: string;
         nextSentence: string | null;
     };
+    lastMatchedSentenceId: number;
+    lastMatchedSentenceContext: {
+        prevSentence: string | null;
+        matchedSentence: string;
+        nextSentence: string | null;
+    };
+    chapterSentenceCount: number;
     audioFiles: AudioFileContext[];
 }
 interface Report {
@@ -47,11 +54,17 @@ declare class Aligner {
     private granularity;
     report: Report;
     constructor(epub: Epub, audiofiles: string[], transcriptions: Pick<RecognitionResult, "transcript" | "timeline">[], granularity: "sentence" | "word" | null | undefined, languageOverride?: (Intl.Locale | null) | undefined, logger?: (Logger | null) | undefined);
-    private findBestOffset;
     private getChapterSentences;
     private writeAlignedChapter;
     private addChapterReport;
     private alignChapter;
+    narrowToAvailableBoundary(boundary: {
+        start: number;
+        end: number;
+    }): {
+        start: number;
+        end: number;
+    };
     alignBook(onProgress?: ((progress: number) => void) | null): Promise<_storyteller_platform_ghost_story.TimingAggregator>;
 }
 declare function concatTranscriptions(transcriptions: Pick<RecognitionResult, "transcript" | "timeline">[], audiofiles: string[]): StorytellerTranscription;

package/dist/align/align.d.ts

@@ -22,6 +22,13 @@ interface ChapterReport {
         matchedSentence: string;
         nextSentence: string | null;
     };
+    lastMatchedSentenceId: number;
+    lastMatchedSentenceContext: {
+        prevSentence: string | null;
+        matchedSentence: string;
+        nextSentence: string | null;
+    };
+    chapterSentenceCount: number;
     audioFiles: AudioFileContext[];
 }
 interface Report {
@@ -47,11 +54,17 @@ declare class Aligner {
     private granularity;
     report: Report;
     constructor(epub: Epub, audiofiles: string[], transcriptions: Pick<RecognitionResult, "transcript" | "timeline">[], granularity: "sentence" | "word" | null | undefined, languageOverride?: (Intl.Locale | null) | undefined, logger?: (Logger | null) | undefined);
-    private findBestOffset;
     private getChapterSentences;
     private writeAlignedChapter;
     private addChapterReport;
     private alignChapter;
+    narrowToAvailableBoundary(boundary: {
+        start: number;
+        end: number;
+    }): {
+        start: number;
+        end: number;
+    };
     alignBook(onProgress?: ((progress: number) => void) | null): Promise<_storyteller_platform_ghost_story.TimingAggregator>;
 }
 declare function concatTranscriptions(transcriptions: Pick<RecognitionResult, "transcript" | "timeline">[], audiofiles: string[]): StorytellerTranscription;

package/dist/align/align.js

@@ -5,9 +5,7 @@ import {
 import { copyFile, mkdir, readFile, readdir, writeFile } from "node:fs/promises";
 import { dirname as autoDirname, join as autoJoin } from "node:path";
 import { basename, dirname, parse, relative } from "node:path/posix";
-import { enumerate } from "itertools";
 import memoize from "memoize";
-import { runes } from "runes2";
 import { isAudioFile, lookupAudioMime } from "@storyteller-platform/audiobook";
 import {
   Epub
@@ -18,13 +16,13 @@ import {
 } from "@storyteller-platform/ghost-story";
 import { getTrackDuration } from "../common/ffmpeg.js";
 import { getXhtmlSegmentation } from "../markup/segmentation.js";
-import { findNearestMatch } from "./fuzzy.js";
 import {
   expandEmptySentenceRanges,
   getChapterDuration,
   getSentenceRanges,
   interpolateSentenceRanges
 } from "./getSentenceRanges.js";
+import { findBoundaries } from "./search.js";
 import { slugify } from "./slugify.js";
 async function align(input, output, transcriptionsDir, audiobookDir, options) {
   var _stack = [];
@@ -101,83 +99,6 @@ class Aligner {
   report = {
     chapters: []
   };
-  findBestOffset(epubSentences, transcriptionText, lastMatchOffset, dir = 1) {
-    const reverse = dir < 0;
-    if (dir < 0) {
-      epubSentences = epubSentences.toReversed().map((s) => runes(s).toReversed().join(""));
-      transcriptionText = runes(transcriptionText).toReversed().join("");
-      lastMatchOffset = transcriptionText.length - lastMatchOffset;
-    }
-    const flatSliceIndices = [
-      0,
-      ...this.alignedChapters.toSorted(
-        (a, b) => reverse ? transcriptionText.length - a.endOffset - (transcriptionText.length - b.endOffset) : a.startOffset - b.startOffset
-      ).flatMap((aligned) => [
-        reverse ? transcriptionText.length - aligned.endOffset : aligned.startOffset,
-        reverse ? transcriptionText.length - aligned.startOffset : aligned.endOffset
-      ]),
-      transcriptionText.length
-    ];
-    const sliceIndices = [];
-    for (let i = 0; i < flatSliceIndices.length - 1; i += 2) {
-      sliceIndices.push([flatSliceIndices[i], flatSliceIndices[i + 1]]);
-    }
-    const allSlices = [];
-    let startSlice = 0;
-    for (const [i, [start, end]] of enumerate(sliceIndices)) {
-      if (lastMatchOffset >= start && lastMatchOffset < end) {
-        if (!reverse) {
-          startSlice = i + 1;
-          allSlices.push({
-            start,
-            text: transcriptionText.slice(start, lastMatchOffset)
-          });
-        }
-        allSlices.push({
-          start: lastMatchOffset,
-          text: transcriptionText.slice(lastMatchOffset, end)
-        });
-      } else if (!reverse) {
-        allSlices.push({ start, text: transcriptionText.slice(start, end) });
-      }
-    }
-    const slices = allSlices.filter((slice) => slice.text.length);
-    if (reverse && !slices.length) {
-      const indices = sliceIndices.find(([start]) => start > lastMatchOffset);
-      if (indices) {
-        slices.push({
-          start: indices[0],
-          text: transcriptionText.slice(...indices)
-        });
-      }
-    }
-    for (const slice of slices.slice(startSlice).concat(slices.slice(0, startSlice))) {
-      let startSentence = 0;
-      while (startSentence < epubSentences.length) {
-        const needle = epubSentences.slice(startSentence, startSentence + 6).join("-");
-        const firstMatch = findNearestMatch(
-          needle,
-          slice.text,
-          Math.max(Math.floor(0.1 * needle.length), 1)
-        );
-        if (firstMatch) {
-          const start = reverse ? transcriptionText.length - (slice.start + firstMatch.index) : slice.start + firstMatch.index;
-          return {
-            startSentence: reverse ? epubSentences.length - startSentence : startSentence,
-            transcriptionOffset: start
-          };
-        }
-        startSentence += 3;
-      }
-    }
-    if (reverse) {
-      return {
-        startSentence: epubSentences.length,
-        transcriptionOffset: slices[0] ? transcriptionText.length - slices[0].start : null
-      };
-    }
-    return { startSentence: 0, transcriptionOffset: null };
-  }
   async getChapterSentences(chapterId) {
     const chapterXml = await this.epub.readXhtmlItemContents(chapterId);
     const { result: segmentation } = await getXhtmlSegmentation(
@@ -240,7 +161,7 @@ class Aligner {
       value: Epub.formatSmilDuration(chapterDuration)
     });
   }
-  addChapterReport(chapter, chapterSentences, sentenceRanges, startSentence, transcriptionOffset) {
+  addChapterReport(chapter, chapterSentences, sentenceRanges, startSentence, endSentence, transcriptionOffset) {
     this.report.chapters.push({
       href: chapter.href,
       transcriptionOffset,
@@ -264,6 +185,14 @@ class Aligner {
         matchedSentence: chapterSentences[startSentence],
         nextSentence: chapterSentences[startSentence + 1] ?? null
       },
+      lastMatchedSentenceId: endSentence,
+      lastMatchedSentenceContext: {
+        prevSentence: chapterSentences[endSentence - 1] ?? null,
+        // eslint-disable-next-line @typescript-eslint/no-non-null-assertion
+        matchedSentence: chapterSentences[endSentence],
+        nextSentence: chapterSentences[endSentence + 1] ?? null
+      },
+      chapterSentenceCount: sentenceRanges.length,
       audioFiles: sentenceRanges.reduce((acc, range) => {
         const existing = acc.find(
           (context) => context.filepath === range.audiofile
@@ -281,7 +210,7 @@ class Aligner {
       }, [])
     });
   }
-  async alignChapter(startSentence, endSentence, chapterId, transcriptionOffset, transcriptionEndOffset, locale, mapping) {
+  async alignChapter(chapterId, transcriptionOffset, transcriptionEndOffset, locale, mapping) {
     const timing = createTiming();
     timing.start("read contents");
     const manifest = await this.epub.getManifest();
@@ -296,9 +225,12 @@ class Aligner {
     const chapterSentences = await this.getChapterSentences(chapterId);
     timing.end("split to sentences");
     timing.start("align sentences");
-    const { sentenceRanges, transcriptionOffset: endTranscriptionOffset } = await getSentenceRanges(
-      startSentence,
-      endSentence,
+    const {
+      sentenceRanges,
+      transcriptionOffset: endTranscriptionOffset,
+      firstFoundSentence,
+      lastFoundSentence
+    } = await getSentenceRanges(
       this.transcription,
       chapterSentences,
       transcriptionOffset,
@@ -326,7 +258,8 @@ class Aligner {
       chapter,
       chapterSentences,
       sentenceRanges,
-      startSentence,
+      firstFoundSentence,
+      lastFoundSentence,
       transcriptionOffset
     );
     return {
@@ -335,8 +268,20 @@ class Aligner {
       timing
     };
   }
+  narrowToAvailableBoundary(boundary) {
+    const narrowed = { ...boundary };
+    for (const chapter of this.alignedChapters) {
+      if (chapter.startOffset > narrowed.start && chapter.startOffset <= narrowed.end) {
+        narrowed.end = chapter.startOffset - 1;
+      }
+      if (chapter.endOffset < narrowed.end && chapter.endOffset >= narrowed.start) {
+        narrowed.start = chapter.endOffset + 1;
+      }
+    }
+    return narrowed;
+  }
   async alignBook(onProgress) {
-    var _a, _b, _c, _d, _e, _f, _g, _h;
+    var _a, _b, _c, _d, _e, _f;
     const locale = this.languageOverride ?? await this.epub.getLanguage() ?? new Intl.Locale("en-US");
     this.timing.setMetadata("language", locale.toString());
     this.timing.setMetadata("granularity", this.granularity);
@@ -346,7 +291,6 @@ class Aligner {
       this.transcription.transcript,
       locale
     );
-    let lastTranscriptionOffset = 0;
     for (let index = 0; index < spine.length; index++) {
       onProgress == null ? void 0 : onProgress(index / spine.length);
       const spineItem = spine[index];
@@ -375,48 +319,29 @@ class Aligner {
         );
         continue;
       }
-      const { startSentence, transcriptionOffset: slugifiedOffset } = this.findBestOffset(
-        slugifiedChapterSentences,
-        transcriptionText,
-        mapping.map(lastTranscriptionOffset, -1)
+      const boundaries = findBoundaries(
+        slugifiedChapterSentences.join("-"),
+        transcriptionText
       );
-      if (slugifiedOffset === null) {
+      if (!boundaries) {
         (_f = this.logger) == null ? void 0 : _f.info(
-          `Couldn't find matching transcription for chapter #${index}`
+          `Could not find chapter #${index} in the transcripton`
         );
         continue;
       }
-      const transcriptionOffset = mapping.invert().map(slugifiedOffset, -1);
-      const {
-        startSentence: startEndSentence,
-        transcriptionOffset: slugifiedEndOffset
-      } = this.findBestOffset(
-        slugifiedChapterSentences,
-        transcriptionText,
-        Math.min(
-          transcriptionText.length,
-          slugifiedOffset + Math.round(slugifiedChapterSentences.join("-").length * 1.2)
-        ),
-        -1
-      );
-      const endSentence = startEndSentence;
-      const endOffset = slugifiedEndOffset === null ? this.transcription.transcript.length : mapping.invert().map(slugifiedEndOffset, 1);
-      if (endSentence - startSentence < slugifiedChapterSentences.length / 2) {
-        (_g = this.logger) == null ? void 0 : _g.info(`Found less than half of chapter #${index}, skipping`);
+      const { start, end } = this.narrowToAvailableBoundary(boundaries);
+      if (start === end) {
+        continue;
       }
-      (_h = this.logger) == null ? void 0 : _h.info(
-        `Chapter #${index} best matches transcription from ${transcriptionOffset} to ${endOffset}, from sentence ${startSentence} to ${endSentence} (of ${slugifiedChapterSentences.length}) in the book`
-      );
+      const transcriptionOffset = mapping.invert().map(Math.max(start, 0), -1);
+      const endOffset = mapping.invert().map(Math.min(end, transcriptionText.length), 1);
       const result = await this.alignChapter(
-        startSentence,
-        endSentence,
         chapterId,
         transcriptionOffset,
         endOffset,
         locale,
         mapping
       );
-      lastTranscriptionOffset = result.endTranscriptionOffset;
       this.timing.add(result.timing.summary());
     }
     const audioOrderedChapters = this.alignedChapters.toSorted((a, b) => {