| """Generate ElevenLabs premium voiceover WITH word-level timestamps for the |
| DocDoe ad reel, so on-screen captions land exactly with the spoken word. |
| |
| Uses POST /v1/text-to-speech/{voice}/with-timestamps → returns audio + char |
| alignment. We request PCM 24k, wrap to WAV, and fold char timings into word |
| timings. Writes public/ad-vo/<id>.wav + manifest.json (with `words`). |
| |
| EL_KEY=sk_... backend/.venv/Scripts/python.exe backend/scripts/generate_ad_vo.py |
| """ |
| from __future__ import annotations |
|
|
| import base64 |
| import json |
| import os |
| import sys |
| import time |
| import wave |
| from pathlib import Path |
|
|
| import requests |
|
|
| ROOT = Path(__file__).resolve().parents[2] |
| OUT_DIR = ROOT / "public" / "ad-vo" |
| OUT_DIR.mkdir(parents=True, exist_ok=True) |
|
|
| VOICE_ID = "nPczCjzI2devNBz1zQrb" |
| MODEL = "eleven_multilingual_v2" |
| SAMPLE_RATE = 24000 |
| OUTPUT_FORMAT = "pcm_24000" |
| |
| VOICE_SETTINGS = {"stability": 0.6, "similarity_boost": 0.9, "style": 0.1, "use_speaker_boost": True} |
|
|
| |
| |
| SEGMENTS: list[dict[str, str]] = [ |
| {"id": "hook", "text": "Studying all night, but still scoring low?"}, |
| {"id": "problem", "text": "Your textbook is huge. The exam tests only what matters."}, |
| {"id": "intro", "text": "Meet Dock Doe. Your A.I. exam coach."}, |
| {"id": "upload", "text": "Upload a PDF. Ask any question. Get an exam ready answer."}, |
| {"id": "features", "text": "Notes, quizzes, flashcards, and mark wise answers, in seconds."}, |
| {"id": "predict", "text": "Dock Doe predicts the important questions, and writes mark wise answers."}, |
| {"id": "video", "text": "Need a teacher? Dock Doe builds a full lesson, just for you."}, |
| {"id": "proof", "text": "Free to start. Built for every board and subject."}, |
| {"id": "cta", "text": "Start free today, at Dock Doe dot in."}, |
| ] |
|
|
|
|
| def synth(text: str, key: str) -> dict: |
| url = f"https://api.elevenlabs.io/v1/text-to-speech/{VOICE_ID}/with-timestamps" |
| params = {"output_format": OUTPUT_FORMAT} |
| body = {"text": text, "model_id": MODEL, "voice_settings": VOICE_SETTINGS} |
| headers = {"xi-api-key": key, "Content-Type": "application/json"} |
| last = None |
| for attempt in range(5): |
| try: |
| resp = requests.post(url, params=params, json=body, headers=headers, timeout=120) |
| if resp.status_code != 200: |
| raise SystemExit(f"ElevenLabs {resp.status_code}: {resp.text[:300]}") |
| return resp.json() |
| except (requests.exceptions.ConnectionError, requests.exceptions.Timeout) as exc: |
| last = exc |
| time.sleep(2 * (attempt + 1)) |
| print(f" retry {attempt + 1}/5", flush=True) |
| raise SystemExit(f"ElevenLabs unreachable: {last}") |
|
|
|
|
| def write_wav(pcm: bytes, path: Path) -> float: |
| with wave.open(str(path), "wb") as w: |
| w.setnchannels(1) |
| w.setsampwidth(2) |
| w.setframerate(SAMPLE_RATE) |
| w.writeframes(pcm) |
| with wave.open(str(path), "rb") as r: |
| return round(r.getnframes() / float(r.getframerate()), 3) |
|
|
|
|
| def chars_to_words(text: str, chars: list[str], starts: list[float], ends: list[float]) -> list[dict]: |
| """Fold per-character alignment into per-word [text,start,end].""" |
| words: list[dict] = [] |
| cur = "" |
| cur_start = None |
| cur_end = 0.0 |
| for ch, st, en in zip(chars, starts, ends): |
| if ch.strip() == "": |
| if cur: |
| words.append({"w": cur, "start": round(cur_start or 0, 3), "end": round(cur_end, 3)}) |
| cur, cur_start = "", None |
| continue |
| if cur_start is None: |
| cur_start = st |
| cur += ch |
| cur_end = en |
| if cur: |
| words.append({"w": cur, "start": round(cur_start or 0, 3), "end": round(cur_end, 3)}) |
| return words |
|
|
|
|
| def main() -> int: |
| key = os.environ.get("EL_KEY", "").strip() |
| if not key: |
| raise SystemExit("Set EL_KEY env var.") |
| manifest, total = [], 0.0 |
| for seg in SEGMENTS: |
| print(f"-> {seg['id']}: {seg['text'][:46]}...", flush=True) |
| data = synth(seg["text"], key) |
| pcm = base64.b64decode(data["audio_base64"]) |
| out = OUT_DIR / f"{seg['id']}.wav" |
| dur = write_wav(pcm, out) |
| align = data.get("alignment") or data.get("normalized_alignment") or {} |
| words = chars_to_words( |
| seg["text"], |
| align.get("characters", []), |
| align.get("character_start_times_seconds", []), |
| align.get("character_end_times_seconds", []), |
| ) |
| total += dur |
| manifest.append({"id": seg["id"], "text": seg["text"], "duration": dur, "file": f"ad-vo/{seg['id']}.wav", "words": words}) |
| print(f" {dur}s, {len(words)} words", flush=True) |
| (OUT_DIR / "manifest.json").write_text(json.dumps(manifest, indent=2), encoding="utf-8") |
| print(f"\n{MODEL} | total {round(total, 2)}s / {len(manifest)} segs (word-timed)", flush=True) |
| return 0 |
|
|
|
|
| if __name__ == "__main__": |
| sys.exit(main()) |
|
|
