submaster/whisper_project/translate_srt_with_gemini.py

#!/usr/bin/env python3
"""translate_srt_with_gemini.py
Lee un .srt, traduce cada bloque de texto con Gemini (Google Generative API) y
escribe un nuevo .srt manteniendo índices y timestamps.

Uso:
  export GEMINI_API_KEY="..."
  .venv/bin/python whisper_project/translate_srt_with_gemini.py \
    --in whisper_project/dailyrutines.kokoro.dub.srt \
    --out whisper_project/dailyrutines.kokoro.dub.es.srt \
    --model gemini-2.5-flash

Si no pasas --gemini-api-key, se usará la variable de entorno GEMINI_API_KEY.
"""
import argparse
import json
import os
import time
from typing import List

import requests
import srt
# Intentar usar la librería oficial si está instalada (mejor compatibilidad)
try:
    import google.generativeai as genai  # type: ignore
except Exception:
    genai = None


def translate_text_google_gl(text: str, api_key: str, model: str = "gemini-2.5-flash") -> str:
    """Llamada a la API Generative Language de Google (generateContent).
    Devuelve el texto traducido (o el texto original si falla).
    """
    if not api_key:
        raise ValueError("gemini api key required")
    # Si la librería oficial está disponible, usarla (maneja internamente los endpoints)
    if genai is not None:
        try:
            genai.configure(api_key=api_key)
            model_obj = genai.GenerativeModel(model)
            # la librería acepta un prompt simple o lista; pedimos texto traducido explícitamente
            prompt = f"Traduce al español el siguiente texto y devuelve solo el texto traducido:\n\n{text}"
            resp = model_obj.generate_content(prompt, generation_config={"max_output_tokens": 1024, "temperature": 0.0})
            # resp.text está disponible en la respuesta wrapper
            if hasattr(resp, "text") and resp.text:
                return resp.text.strip()
            # fallback: revisar candidates
            if hasattr(resp, "candidates") and resp.candidates:
                c = resp.candidates[0]
                if hasattr(c, "content") and hasattr(c.content, "parts"):
                    parts = [p.text for p in c.content.parts if getattr(p, "text", None)]
                    if parts:
                        return "\n".join(parts).strip()
        except Exception as e:
            print(f"Warning: genai library translate failed: {e}")

    # Fallback HTTP (legacy/path-variant). Intentamos v1 y v1beta2 según disponibilidad.
    for prefix in ("v1", "v1beta2"):
        endpoint = (
            f"https://generativelanguage.googleapis.com/{prefix}/models/{model}:generateContent?key={api_key}"
        )
        body = {
            "prompt": {"text": f"Traduce al español el siguiente texto y devuelve solo el texto traducido:\n\n{text}"},
            "maxOutputTokens": 1024,
            "temperature": 0.0,
            "candidateCount": 1,
        }
        try:
            r = requests.post(endpoint, json=body, timeout=30)
            r.raise_for_status()
            j = r.json()
            # buscar candidatos
            if isinstance(j, dict) and "candidates" in j and isinstance(j["candidates"], list) and j["candidates"]:
                first = j["candidates"][0]
                if isinstance(first, dict):
                    if "content" in first and isinstance(first["content"], str):
                        return first["content"].strip()
                    if "output" in first and isinstance(first["output"], str):
                        return first["output"].strip()
                    if "content" in first and isinstance(first["content"], list):
                        parts = []
                        for c in first["content"]:
                            if isinstance(c, dict) and isinstance(c.get("text"), str):
                                parts.append(c.get("text"))
                        if parts:
                            return "\n".join(parts).strip()
            for key in ("output_text", "text", "response", "translated_text"):
                if key in j and isinstance(j[key], str):
                    return j[key].strip()
        except Exception as e:
            print(f"Warning: GL translate failed ({prefix}): {e}")

    return text


def translate_srt_file(in_path: str, out_path: str, api_key: str, model: str):
    with open(in_path, "r", encoding="utf-8") as fh:
        subs = list(srt.parse(fh.read()))

    for i, sub in enumerate(subs, start=1):
        text = sub.content.strip()
        if not text:
            continue
        # llamar a la API
        try:
            translated = translate_text_google_gl(text, api_key, model=model)
        except Exception as e:
            print(f"Warning: translate failed for index {sub.index}: {e}")
            translated = text
        # asignar traducido
        sub.content = translated
        # pequeño delay para no golpear la API demasiado rápido
        time.sleep(0.15)
        print(f"Translated {i}/{len(subs)}")

    out_s = srt.compose(subs)
    with open(out_path, "w", encoding="utf-8") as fh:
        fh.write(out_s)
    print(f"Wrote translated SRT to: {out_path}")


def main():
    p = argparse.ArgumentParser()
    p.add_argument("--in", dest="in_srt", required=True)
    p.add_argument("--out", dest="out_srt", required=True)
    p.add_argument("--gemini-api-key", default=None)
    p.add_argument("--model", default="gemini-2.5-flash")
    args = p.parse_args()

    key = args.gemini_api_key or os.environ.get("GEMINI_API_KEY")
    if not key:
        print("Provide --gemini-api-key or set GEMINI_API_KEY env var", flush=True)
        raise SystemExit(2)

    translate_srt_file(args.in_srt, args.out_srt, key, args.model)


if __name__ == '__main__':
    main()