feat: Enhance EPUB export by grouping chunks for rendering and adding group ID support

abogen/epub3/exporter.py

@@ -24,6 +24,7 @@ class ChunkOverlay:
     speaker_id: str
     voice: Optional[str]
     level: Optional[str] = None
+    group_id: Optional[str] = None
 
 
 @dataclass(slots=True)
@@ -256,6 +257,10 @@ class EPUB3PackageBuilder:
                 normalized_id = f"{normalized_id}_dup"
             used_ids.add(normalized_id)
 
+            raw_group_key = chunk_entry.get("id") if chunk_entry else chunk_id
+            group_id = _derive_group_id(raw_group_key, level)
+            normalized_group_id = _normalize_chunk_id(group_id) if group_id else None
+
             overlays.append(
                 ChunkOverlay(
                     id=normalized_id,
@@ -266,6 +271,7 @@ class EPUB3PackageBuilder:
                     speaker_id=speaker_id,
                     voice=str(voice) if voice else None,
                     level=str(level) if level else None,
+                    group_id=normalized_group_id,
                 )
             )
 
@@ -281,7 +287,10 @@ class EPUB3PackageBuilder:
     def _render_chapter_xhtml(self, chapter: ChapterDocument) -> str:
         language = html.escape(self._language or "en")
         title = html.escape(chapter.title)
-        chunk_html = "\n".join(_render_chunk_html(chunk) for chunk in chapter.chunks)
+        grouped_chunks = _group_chunks_for_render(chapter.chunks)
+        chunk_html = "\n".join(
+            _render_chunk_group_html(group_id, items) for group_id, items in grouped_chunks
+        )
         if not chunk_html:
             chunk_html = "<p></p>"
         original_block = ""
@@ -647,7 +656,40 @@ def _normalize_chunk_id(chunk_id: Optional[Any]) -> Optional[str]:
     return safe[:120]
 
 
-def _render_chunk_html(chunk: ChunkOverlay) -> str:
+def _derive_group_id(chunk_id: Optional[Any], level: Optional[Any]) -> Optional[str]:
+    if chunk_id is None:
+        return None
+    text = str(chunk_id).strip()
+    if not text:
+        return None
+    if str(level or "").lower() == "sentence":
+        match = re.match(r"(.+?)_s\d+(?:_.*)?$", text)
+        if match:
+            return match.group(1)
+    return text
+
+
+def _group_chunks_for_render(chunks: Sequence[ChunkOverlay]) -> List[Tuple[Optional[str], List[ChunkOverlay]]]:
+    groups: List[Tuple[Optional[str], List[ChunkOverlay]]] = []
+    current_key: Optional[str] = None
+    current_items: List[ChunkOverlay] = []
+
+    for chunk in chunks:
+        key = chunk.group_id or chunk.id
+        if current_items and key != current_key:
+            groups.append((current_key, current_items))
+            current_items = []
+        if not current_items:
+            current_key = key
+        current_items.append(chunk)
+
+    if current_items:
+        groups.append((current_key, current_items))
+
+    return groups
+
+
+def _render_chunk_inline(chunk: ChunkOverlay) -> str:
     escaped_id = html.escape(chunk.id)
     speaker_attr = f" data-speaker=\"{html.escape(chunk.speaker_id)}\"" if chunk.speaker_id else ""
     voice_attr = f" data-voice=\"{html.escape(chunk.voice)}\"" if chunk.voice else ""
@@ -657,12 +699,22 @@ def _render_chunk_html(chunk: ChunkOverlay) -> str:
     if not escaped_text:
         escaped_text = "&nbsp;"
     return (
-        f"      <div class=\"chunk\" id=\"{escaped_id}\"{speaker_attr}{voice_attr}{level_attr}>"
+        f"<span class=\"chunk\" id=\"{escaped_id}\"{speaker_attr}{voice_attr}{level_attr}>"
         f"{escaped_text}"
-        "</div>"
+        "</span>"
     )
 
 
+def _render_chunk_group_html(group_id: Optional[str], chunks: Sequence[ChunkOverlay]) -> str:
+    if not chunks:
+        return ""
+    group_attr = f" data-group=\"{html.escape(group_id)}\"" if group_id else ""
+    inline_html = "".join(_render_chunk_inline(chunk) for chunk in chunks)
+    if not inline_html:
+        inline_html = "&nbsp;"
+    return f"      <p class=\"chunk-group\"{group_attr}>{inline_html}</p>"
+
+
 def _format_smil_time(value: Optional[float]) -> str:
     if value is None or value < 0:
         value = 0.0
@@ -840,20 +892,11 @@ h1 {
   margin-bottom: 0.5em;
 }
 
-div.chunk {
+.chunk-group {
-    margin: 0 0 1em 0;
+    margin: 0.5em 0;
+}
+
+.chunk-group .chunk {
     white-space: pre-wrap;
 }
-
-div.chunk[data-level="sentence"] {
-    margin-bottom: 0.5em;
-}
-
-div.chunk[data-level="sentence"] + div.chunk[data-level="sentence"] {
-    margin-top: -0.2em;
-}
-
-p {
-  margin: 0.5em 0;
-}
 """

abogen/voice_cache.py

@@ -10,6 +10,10 @@ except Exception:  # pragma: no cover - import fallback
     hf_hub_download = None  # type: ignore[assignment]
     LocalEntryNotFoundError = None  # type: ignore[assignment]
 
+if LocalEntryNotFoundError is None:  # pragma: no cover - fallback for tests
+    class LocalEntryNotFoundError(Exception):
+        pass
+
 from abogen.constants import VOICES_INTERNAL
 
 _CACHE_LOCK = threading.Lock()
@@ -116,11 +120,18 @@ def _ensure_single_voice_asset(
         raise RuntimeError("huggingface_hub is required to cache voices")
 
     filename = f"voices/{voice_id}.pt"
+    common_kwargs = {
+        "repo_id": repo_id,
+        "filename": filename,
+    }
+    if cache_dir is not None:
+        common_kwargs["cache_dir"] = cache_dir
 
-    hf_hub_download(
+    try:
-        repo_id=repo_id,
+        hf_hub_download(local_files_only=True, **common_kwargs)
-        filename=filename,
+        return False
-        cache_dir=cache_dir,
+    except LocalEntryNotFoundError:
-        resume_download=True,
+        pass
-    )
+
+    hf_hub_download(resume_download=True, **common_kwargs)
     return True

tests/test_epub_exporter.py

@@ -177,4 +177,78 @@ def test_epub3_preserves_original_whitespace(tmp_path) -> None:
     match = re.search(r"<pre class=\"chapter-original\"[^>]*>(.*?)</pre>", chapter_doc, re.DOTALL)
     assert match is not None
     original_text = html.unescape(match.group(1))
-    assert "Second line\n\nThird paragraph." in original_text
+    assert "Second line\n\nThird paragraph." in original_text
+
+
+def test_epub3_sentence_chunks_render_as_paragraphs(tmp_path) -> None:
+    extraction = ExtractionResult(
+        chapters=[
+            ExtractedChapter(
+                title="Chapter 1",
+                text="First sentence. Second sentence in same paragraph.\n\nNew paragraph starts here.",
+            )
+        ],
+        metadata={"title": "Sample", "artist": "Author", "language": "en"},
+    )
+
+    chunks = [
+        {
+            "id": "chap0000_p0000_s0000",
+            "chapter_index": 0,
+            "chunk_index": 0,
+            "text": "First sentence.",
+            "level": "sentence",
+            "speaker_id": "narrator",
+        },
+        {
+            "id": "chap0000_p0000_s0001",
+            "chapter_index": 0,
+            "chunk_index": 1,
+            "text": "Second sentence in same paragraph.",
+            "level": "sentence",
+            "speaker_id": "narrator",
+        },
+        {
+            "id": "chap0000_p0001_s0000",
+            "chapter_index": 0,
+            "chunk_index": 2,
+            "text": "New paragraph starts here.",
+            "level": "sentence",
+            "speaker_id": "narrator",
+        },
+    ]
+
+    chunk_markers = [
+        {"id": chunk["id"], "chapter_index": 0, "chunk_index": chunk["chunk_index"], "start": None, "end": None}
+        for chunk in chunks
+    ]
+
+    audio_path = tmp_path / "audio.mp3"
+    audio_path.write_bytes(b"ID3 audio")
+    output_path = tmp_path / "output.epub"
+
+    build_epub3_package(
+        output_path=output_path,
+        book_id="job-paragraphs",
+        extraction=extraction,
+        metadata_tags={"title": "Sample", "artist": "Author", "language": "en"},
+        chapter_markers=[],
+        chunk_markers=chunk_markers,
+        chunks=chunks,
+        audio_path=audio_path,
+        speaker_mode="single",
+    )
+
+    with zipfile.ZipFile(output_path) as archive:
+        chapter_doc = archive.read("OEBPS/text/chapter_0001.xhtml").decode("utf-8")
+
+    assert '<div class="chunk"' not in chapter_doc
+    assert chapter_doc.count('<p class="chunk-group"') == 2
+    assert 'First sentence.' in chapter_doc
+    assert 'Second sentence in same paragraph.' in chapter_doc
+
+    first_paragraph_start = chapter_doc.find('<p class="chunk-group"')
+    first_paragraph_end = chapter_doc.find('</p>', first_paragraph_start)
+    first_paragraph = chapter_doc[first_paragraph_start:first_paragraph_end]
+    assert "First sentence." in first_paragraph
+    assert "Second sentence in same paragraph." in first_paragraph

tests/test_voice_cache.py

@@ -4,7 +4,7 @@ from typing import cast
 import pytest
 
 from abogen.constants import VOICES_INTERNAL
-from abogen.voice_cache import _CACHED_VOICES, ensure_voice_assets
+from abogen.voice_cache import LocalEntryNotFoundError, _CACHED_VOICES, ensure_voice_assets
 from abogen.web.conversion_runner import _collect_required_voice_ids
 from abogen.web.service import Job
 
@@ -19,9 +19,18 @@ def clear_voice_cache():
 def test_ensure_voice_assets_downloads_missing(monkeypatch):
     recorded = []
 
+    cached = set()
+
     def fake_download(**kwargs):
-        recorded.append(kwargs["filename"])
+        filename = kwargs["filename"]
-        return "/tmp/fake"
+        if kwargs.get("local_files_only"):
+            if filename in cached:
+                return f"/tmp/{filename}"
+            raise LocalEntryNotFoundError(f"{filename} missing")
+
+        recorded.append(filename)
+        cached.add(filename)
+        return f"/tmp/{filename}"
 
     monkeypatch.setattr("abogen.voice_cache.hf_hub_download", fake_download)