# SPDX-License-Identifier: Apache-2.0 # Copyright 2026 Kale """make_ids.py: the bench's text -> ids step. The 192 lines (script.jsonl) and the guard lines (config/bench.json) go through the RELEASE text front end (../../python/frontend_cli.mjs: sentences + normaliser + lexicon/g2p -> Kokoro-table ids), once per accent, and are written as ids-.json = {"guard": [{name, sents}], "bench": [{id, sents}]} (sents = one id list per sentence, no 0 pads). bench.py renders each voice with the ids of its accent (config/bench.json "accent": d03 / d13 en-gb, all others en-us). Same role as the id steps (en-us and en-gb) used for the release tables; the release tables' en-us ids came from our earlier Python copy of the same lexicon + g2p (not part of this release), so a few lines may differ by an id or two from what this script writes. usage: python3 make_ids.py [--frontend ../../python/frontend_cli.mjs] [--node node] (needs node >= 18, no npm packages)""" import argparse import json import shutil import subprocess from pathlib import Path HERE = Path(__file__).resolve().parent ap = argparse.ArgumentParser() ap.add_argument("--frontend", default=str(HERE.parent.parent / "python" / "frontend_cli.mjs")) ap.add_argument("--node", default=shutil.which("node") or "node") ap.add_argument("--config", default=str(HERE / "config" / "bench.json")) a = ap.parse_args() cfg = json.loads(Path(a.config).read_text()) lines = [r for r in map(json.loads, (HERE / "script.jsonl").read_text().splitlines()) if "id" in r] p = subprocess.Popen([a.node, a.frontend], stdin=subprocess.PIPE, stdout=subprocess.PIPE, text=True, bufsize=1) def sents(text, accent): p.stdin.write(json.dumps({"text": text, "accent": accent}) + "\n"); p.stdin.flush() r = json.loads(p.stdout.readline()) if "error" in r: raise ValueError(f"{text[:60]}: {r['error']}") return [s["ids"] for s in r["sentences"]] for accent in sorted(set(cfg["accent"].values())): out = dict(guard=[dict(name=n, sents=sents(t, accent)) for n, t in cfg["guard_lines"]], bench=[dict(id=l["id"], sents=sents(l["text"], accent)) for l in lines]) (HERE / f"ids-{accent}.json").write_text(json.dumps(out)) print(accent, len(out["guard"]), "guard lines,", len(out["bench"]), "bench lines,", sum(len(s) for l in out["bench"] for s in l["sents"]), "ids") p.terminate()