Add pipeline code

_pipeline/script.py

@@ -0,0 +1,104 @@
+"""Stage 4, narration script. One JSON file per audio file, Audible style:
+opening-credits, chapter-NN (one chapter per file), closing-credits, retail-sample.
+A script is a list of segments {kind, added, text}. Author text is verbatim (added=false).
+Everything else is added=true. Each inserted part (preview, summary) is wrapped:
+cue chime-start, label, text, cue chime-end."""
+import json
+import os
+
+import bookio
+import llm
+import structure
+
+LABELS = {"chapter_preview": "Chapter preview", "chapter_summary": "Chapter summary", "concept_summary": "Concept summary"}
+PROMPTS = {
+    "chapter_preview": "In two sentences of plain spoken English, say what this chapter is about. Use only what the text says.",
+    "chapter_summary": "In three sentences of plain spoken English, summarize this chapter. Use only what the text says.",
+    "concept_summary": "In two sentences of plain spoken English, restate the main concept in this passage. Use only what the text says.",
+}
+
+
+def switch(cfg, book, chapter, name):
+    """Chapter switch beats book switch beats settings default."""
+    for source in (chapter.get("switches", {}), book.get("summaries", {})):
+        if source.get(name) is not None:
+            return bool(source[name])
+    return bool(cfg["summaries"][name])
+
+
+def seg(kind, text, added=True):
+    return {"kind": kind, "added": added, "text": text}
+
+
+def inserted(kind, text):
+    return [seg("cue", "chime-start"), seg("label", LABELS[kind]), seg(kind, text), seg("cue", "chime-end")]
+
+
+def spoken_header(ch):
+    n = ch["number"]
+    word = structure.WORDS[n - 1].title() if n <= len(structure.WORDS) else str(n)
+    return "Chapter %s. %s." % (word, ch["title"])
+
+
+def credits(book, cfg, closing):
+    by = "written by %s" % book["author"]
+    if cfg.get("narrator"):
+        by += ", narrated by %s" % cfg["narrator"]
+    return [seg("credits", ("This is the end of %s, %s." if closing else "This is %s, %s.") % (book["title"], by))]
+
+
+def chapter_script(cfg, book, ch, paragraphs):
+    segs = [seg("header", spoken_header(ch))]
+    full = "\n\n".join(paragraphs)
+    if switch(cfg, book, ch, "chapter_preview"):
+        segs += inserted("chapter_preview", llm.complete(cfg, PROMPTS["chapter_preview"], full))
+    mid = None
+    if switch(cfg, book, ch, "concept_summary"):
+        total, running = sum(len(p.split()) for p in paragraphs), 0
+        for i, p in enumerate(paragraphs):
+            running += len(p.split())
+            if running * 2 >= total:
+                mid = i
+                break
+    for i, p in enumerate(paragraphs):
+        segs.append(seg("author", p, added=False))
+        if i == mid:
+            segs += inserted("concept_summary", llm.complete(cfg, PROMPTS["concept_summary"], "\n\n".join(paragraphs[:i + 1])))
+    if switch(cfg, book, ch, "chapter_summary"):
+        segs += inserted("chapter_summary", llm.complete(cfg, PROMPTS["chapter_summary"], full))
+    return segs
+
+
+def retail_sample(cfg, ch, paragraphs):
+    budget, segs, words = cfg["script"]["retail_sample_words"], [seg("header", spoken_header(ch))], 0
+    for p in paragraphs:
+        n = len(p.split())
+        if segs[1:] and words + n > budget:
+            break
+        segs.append(seg("author", p, added=False))
+        words += n
+    return segs
+
+
+def write(book_dir, name, segs):
+    os.makedirs(os.path.join(book_dir, "script"), exist_ok=True)
+    with open(os.path.join(book_dir, "script", name + ".json"), "w", encoding="utf-8") as f:
+        json.dump(segs, f, indent=2, ensure_ascii=False)
+        f.write("\n")
+
+
+def run(cfg, book_dir, book, chapter_numbers):
+    """Writes script/*.json and the audio file list in book.json. Returns warnings."""
+    files = [("opening-credits", credits(book, cfg, False))]
+    chosen = [c for c in book["chapters"] if c["number"] in chapter_numbers]
+    for ch in chosen:
+        with open(os.path.join(book_dir, ch["text_file"]), encoding="utf-8") as f:
+            paragraphs = [p for p in f.read().split("\n\n") if p.strip()]
+        files.append(("chapter-%02d" % ch["number"], chapter_script(cfg, book, ch, [p.strip() for p in paragraphs])))
+        if ch is chosen[0]:
+            sample = ("retail-sample", retail_sample(cfg, ch, [p.strip() for p in paragraphs]))
+    files += [("closing-credits", credits(book, cfg, True)), sample]
+    for name, segs in files:
+        write(book_dir, name, segs)
+    book["audio_files"] = [n for n, _ in files]
+    return []