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

ensvoice/render.py

@@ -0,0 +1,43 @@
+"""Step 4: speak every segment in every seed of voice.seeds (work/takes/<take>_s<seed>.wav).
+
+usage: render [--weak]    --weak: only the segments pick listed as weak, in voice.extra_seeds
+Resumable: existing takes are skipped, so after a lexicon change only the changed segments are spoken.
+"""
+import json
+import sys
+
+from .config import load
+from .tts import backend
+
+
+def reference(cfg, ref_id=None):
+    ref_id = ref_id or cfg["voice"]["reference"]
+    refs = {c["id"]: c for c in json.load(open(cfg.w("refs.json"), encoding="utf-8"))}
+    if ref_id not in refs:
+        raise SystemExit(f"reference {ref_id!r} is not in work/refs.json (run refs; see README)")
+    return cfg.w("refs", f"{ref_id}.wav"), refs[ref_id]["text"]
+
+
+def main():
+    cfg = load()
+    segs = json.load(open(cfg.w("segs.json"), encoding="utf-8"))
+    seeds = cfg["voice"]["seeds"]
+    if "--weak" in sys.argv:
+        weak = set(json.load(open(cfg.w("pick_report.json")))["weak"])
+        segs = [s for s in segs if s["id"] in weak]
+        seeds = cfg["voice"].get("extra_seeds", [])
+    tts = None
+    for seed in seeds:
+        todo = [(s["acute"], str(cfg.w("takes", f"{s['take']}_s{seed}.wav"))) for s in segs
+                if not cfg.w("takes", f"{s['take']}_s{seed}.wav").exists()]
+        print(f"seed {seed}: {len(todo)} segments to speak", flush=True)
+        if not todo:
+            continue
+        if tts is None:
+            tts = backend(cfg, *reference(cfg))
+        tts.speak(todo, seed, log=lambda m: print(m, flush=True))
+    print("RENDER_DONE")
+
+
+if __name__ == "__main__":
+    main()