asmr: renderer for the voice bake-off

tools/asmr.py

@@ -0,0 +1,50 @@
+"""Render asmr/voices.json into asmr/audio/*.mp3 through the shared inference worker, plus a rain bed.
+
+    python3 tools/asmr.py
+
+Keyless: calls inference.yolopreneur.lol as the gandora app. Skips files that already exist. Needs ffmpeg.
+"""
+import json, os, subprocess, sys, urllib.request, urllib.error
+
+ROOT = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+OUT = os.path.join(ROOT, "asmr", "audio")
+W = "https://inference.yolopreneur.lol/v1/speech"
+H = {"Content-Type": "application/json", "Origin": "https://gandora.yolopreneur.lol", "x-app": "gandora", "User-Agent": "gandora-asmr"}
+
+cfg = json.load(open(os.path.join(ROOT, "asmr", "voices.json")))
+os.makedirs(OUT, exist_ok=True)
+fail = 0
+
+bed = os.path.join(OUT, "bed-rain.mp3")
+if not os.path.exists(bed):
+    subprocess.run(["ffmpeg", "-y", "-loglevel", "error",
+        "-f", "lavfi", "-i", "anoisesrc=c=pink:a=0.5:d=60", "-f", "lavfi", "-i", "anoisesrc=c=brown:a=0.6:d=60",
+        "-filter_complex", "[0]highpass=f=900,lowpass=f=7000,tremolo=f=7:d=0.25[a];[1]lowpass=f=300[b];"
+        "[a][b]amix=inputs=2:weights=1 0.6,afade=t=in:d=2,afade=t=out:st=58:d=2,volume=0.6", "-b:a", "96k", bed], check=True)
+
+for v in cfg["voices"]:
+    dst = os.path.join(OUT, v["id"] + ".mp3")
+    if os.path.exists(dst):
+        continue
+    text = cfg["script"].replace("Hey...", "Hey... <sigh>", 1) if v.get("tags") else cfg["script"]
+    fmt = v.get("format", "mp3")
+    body = {"model": v["model"], "voice": v["voice"], "input": text, "response_format": fmt}
+    if v.get("speed"): body["speed"] = v["speed"]
+    if v.get("style"): body["prompt"] = cfg["style"]
+    try:
+        with urllib.request.urlopen(urllib.request.Request(W, json.dumps(body).encode(), H), timeout=110) as r:
+            data = r.read()
+    except urllib.error.HTTPError as e:
+        print("ERR", v["id"], e.code, e.read()[:200]); fail += 1; continue
+    except Exception as e:
+        print("ERR", v["id"], e); fail += 1; continue
+    if fmt == "mp3":
+        open(dst, "wb").write(data)
+    else:
+        tmp = dst + "." + fmt
+        open(tmp, "wb").write(data)
+        subprocess.run(["ffmpeg", "-y", "-loglevel", "error", "-i", tmp, "-b:a", "128k", dst], check=True)
+        os.remove(tmp)
+    print("ok", v["id"])
+
+sys.exit(1 if fail == len(cfg["voices"]) else 0)