Russian narration for the Enshrouded notes: mod 0.4.0, installer, generation pipeline

ensvoice/segments.py

@@ -0,0 +1,65 @@
+"""Step 3: the text for the voice, cut into segments (work/segs.json).
+
+Every note: lexicon.pre -> lines (a blank line = paragraph 'p', a single newline = 'l') -> lines over
+MAXC characters cut at sentence ends into near-equal pieces ('s') -> RUAccent '+' -> lexicon.post ->
+U+0301 text for the ru-stress model.
+A segment's takes are named <id>_<hash>_s<seed>.wav, the hash of the reference and the spoken text:
+change the lexicon or the reference and only the segments it touches are rendered again.
+"""
+import hashlib
+import json
+import math
+import re
+
+from .accent import acute_marks, plus_marks
+from .config import load
+from .lexicon import post, pre
+
+MAXC = 380
+
+
+def split_long(s):
+    if len(s) <= MAXC:
+        return [s]
+    sents = [x.strip() for x in re.findall(r".+?(?:[.!?…]+[»\"')]*(?=\s)|$)\s*", s) if x.strip()]
+    k = math.ceil(len(s) / 330)
+    target, out, cur = len(s) / k, [], ""
+    for x in sents:
+        if cur and len(cur) + len(x) + 1 > target * 1.15 and len(out) < k - 1:
+            out.append(cur); cur = x
+        else:
+            cur = (cur + " " + x).strip()
+    out.append(cur)
+    return out
+
+
+def main():
+    cfg = load()
+    ref = cfg["voice"]["reference"]
+    notes = json.load(open(cfg.w("notes.json"), encoding="utf-8"))
+    if cfg.only:  # ENSVOICE_NOTES=n016,n238: a quick experiment on a few notes
+        notes = [n for n in notes if n["k"] in cfg.only]
+    segs = []
+    for n in notes:
+        units, join = [], "p"
+        for line in pre(n["text"]).split("\n"):
+            line = line.strip()
+            if not line:
+                join = "p"; continue
+            for j, piece in enumerate(split_long(line)):
+                units.append(("s" if j else join, piece))
+            join = "l"
+        for i, (jn, u) in enumerate(units):
+            plus = post(plus_marks(u))
+            acute = acute_marks(plus)
+            sid = f"{n['k']}_{i:02d}"
+            h = hashlib.sha1(f"{ref}|{acute}".encode()).hexdigest()[:10]
+            segs.append(dict(id=sid, take=f"{sid}_{h}", k=n["k"], i=i, join=jn if i else "-",
+                             text=u, plus=plus, acute=acute))
+    json.dump(segs, open(cfg.w("segs.json"), "w", encoding="utf-8"), ensure_ascii=False, indent=0)
+    L = [len(s["text"]) for s in segs]
+    print("segments", len(segs), "notes", len(notes), "longest", max(L), "chars", sum(L))
+
+
+if __name__ == "__main__":
+    main()