ffmpeg -i video.mp4 -vn -acodec pcm_s16le -ar 16000 audio.wav

import requests, json, pathlib

API_URL = "https://api.openai.com/v1/audio/transcriptions"
HEADERS = {"Authorization": f"Bearer {os.getenv('OPENAI_API_KEY')}"}
files = {"file": open("audio.wav", "rb")}
data = {"model": "whisper-1", "language": "pt"}

resp = requests.post(API_URL, headers=HEADERS, files=files, data=data)
transcript = resp.json()["text"]

# Salvar em SRT
with open("legenda.srt", "w", encoding="utf-8") as f:
    # aqui usamos a lib pysrt para formatar timestamps
    import pysrt
    subs = pysrt.from_string(transcript)
    f.write(str(subs))

# 1️⃣ Instalar llama.cpp com suporte a Whisper
git clone https://github.com/ggerganov/whisper.cpp
cd whisper.cpp && make

# 2️⃣ Baixar modelo quantizado 4‑bit (≈ 1 GB)
wget https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-medium.bin

# 3️⃣ Transcrever
./main -m ggml-medium.bin -f audio.wav -otxt -lang pt -output_dir ./output

import re, pathlib

def txt_to_srt(txt_path, srt_path):
    lines = pathlib.Path(txt_path).read_text().splitlines()
    srt = []
    for i, line in enumerate(lines, 1):
        start = i * 2.0               # exemplo simplificado: 2 s por linha
        end   = start + 1.8
        srt.append(f"{i}\n{format_ts(start)} --> {format_ts(end)}\n{line}\n")
    pathlib.Path(srt_path).write_text("\n".join(srt), encoding="utf-8")

def format_ts(seconds):
    h = int(seconds // 3600)
    m = int((seconds % 3600) // 60)
    s = int(seconds % 60)
    ms = int((seconds - int(seconds)) * 1000)
    return f"{h:02}:{m:02}:{s:02},{ms:03}"

txt_to_srt("output/audio.txt", "legenda.srt")

name: gerar-legendas
on:
  push:
    paths:
      - "videos/**.mp4"

jobs:
  subtitles:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v3
      - name: Instalar ffmpeg e whisper.cpp
        run: |
          sudo apt-get install -y ffmpeg
          git clone https://github.com/ggerganov/whisper.cpp
          cd whisper.cpp && make
      - name: Extrair áudio e gerar SRT
        run: |
          ffmpeg -i videos/${{ github.event.head_commit.message }} -vn -acodec pcm_s16le -ar 16000 audio.wav
          ./whisper.cpp/main -m whisper.cpp/ggml-medium.bin -f audio.wav -otxt -lang pt -output_dir ./out
          python txt_to_srt.py out/audio.txt out/legenda.srt
      - name: Upload artefato
        uses: actions/upload-artifact@v3
        with:
          name: legenda
          path: out/legenda.srt