Audio Transcription Pipeline (YouTube Video to Text)

This script automates the process of downloading a video using yt-dlp, converting it to audio using ffmpeg, transcribing the audio into text with Faster-Whisper, and compiling the text into a final document.

---

Installation

Ensure you have the following installed:

1️⃣ Install Dependencies

• Install yt-dlp (for downloading YouTube videos):

  pip install yt-dlp

• Install ffmpeg (for audio conversion):
  • Windows: Download ffmpeg and add it to PATH.
  • Linux/macOS: Install via package manager:

    sudo apt install ffmpeg  # Debian-based (Ubuntu)
    brew install ffmpeg       # macOS (Homebrew)

• Install faster-whisper (for audio transcription):

  pip install faster-whisper

• Install additional dependencies:

  pip install torch numpy tqdm

---

Usage

1️⃣ Download the Video from YouTube

You can download the video using two different methods:

(A) Without Cookies (Public Videos)

If the video is public, you can download it directly without cookies:

yt-dlp -o "<OUTPUT_VIDEO_PATH>" "<YOUTUBE_VIDEO_URL>"

Example:

yt-dlp -o "C:\Users\mohan\OneDrive\Desktop\video.mp4" https://www.youtube.com/watch?v=Kbn2ab0-sGE

(B) With Cookies (Private/Unlisted Videos)

If the video requires login, you must pass your cookies file:

yt-dlp --cookies "<PATH_TO_COOKIES_FILE>" -o "<OUTPUT_VIDEO_PATH>" "<YOUTUBE_VIDEO_URL>"

Example:

yt-dlp --cookies "C:\Users\mohan\OneDrive\Desktop\cookies.txt" -o "C:\Users\mohan\OneDrive\Desktop\video.mp4" https://www.youtube.com/watch?v=Kbn2ab0-sGE

---

How to Get YouTube Cookies for Authentication

If you need to download private or unlisted videos, you must extract cookies from your browser. Follow these steps:

Method 1: Using the cookies.txt extension (Recommended)

1. Install the extension:

   • Chrome: Get cookies.txt extension
   • Firefox: Get cookies.txt extension

2. Go to YouTube and log in to your account.

3. Open the video you want to download.

4. Click on the cookies.txt extension and download the cookies file.

5. Save the cookies file, e.g., C:\Users\mohan\OneDrive\Desktop\cookies.txt.

6. Use it with yt-dlp:

   yt-dlp --cookies "C:\Users\mohan\OneDrive\Desktop\cookies.txt" -o "C:\Users\mohan\OneDrive\Desktop\video.mp4" https://www.youtube.com/watch?v=Kbn2ab0-sGE

---

2️⃣ Convert Video to Audio (MP3/WAV using FFmpeg)

Convert the downloaded file (<OUTPUT_VIDEO_PATH>) into an MP3 or WAV file.

• Convert to MP3:

  ffmpeg -i "<OUTPUT_VIDEO_PATH>" -vn -acodec libmp3lame -q:a 2 "<OUTPUT_AUDIO_PATH>.mp3"

• Convert to WAV (for better transcription accuracy):

  ffmpeg -i "<OUTPUT_VIDEO_PATH>" -vn -acodec pcm_s16le -ar 16000 -ac 1 "<OUTPUT_AUDIO_PATH>.wav"

Example:

ffmpeg -i "C:\Users\mohan\OneDrive\Desktop\video.mp4" -vn -acodec libmp3lame -q:a 2 "C:\Users\mohan\OneDrive\Desktop\audio.mp3"

---

3️⃣ Transcribe Audio to Text with Faster-Whisper

Run Faster-Whisper to transcribe the audio file.

python faster-whisper/whisper.py --model <WHISPER_MODEL_SIZE> --device <DEVICE> --output_dir "<OUTPUT_TEXT_DIR>" "<OUTPUT_AUDIO_PATH>.mp3"

• <WHISPER_MODEL_SIZE>: Choose from tiny, base, small, medium, large-v2.
• <DEVICE>: Use cuda for GPU acceleration or cpu for standard processing.

Example:

python faster-whisper/whisper.py --model small --device cuda --output_dir "C:\Users\mohan\OneDrive\Desktop\output" "C:\Users\mohan\OneDrive\Desktop\audio.mp3"

The transcription results will be saved in the output/ folder.

---

4️⃣ Combine All Transcriptions into a Single Text File

Use convert.py to merge all transcribed text into a final document.

python convert.py "<OUTPUT_TEXT_DIR>"

Example:

python convert.py "C:\Users\mohan\OneDrive\Desktop\output"

This script will:

• Read all transcribed segments from output/.
• Merge them into a single .txt file.

---

Final Output

The final transcribed text will be saved as:

<OUTPUT_TEXT_DIR>/final_transcript.txt

---

Example Workflow (All Steps Combined)

yt-dlp --cookies "C:\Users\mohan\OneDrive\Desktop\cookies.txt" -o "C:\Users\mohan\OneDrive\Desktop\video.mp4" https://www.youtube.com/watch?v=Kbn2ab0-sGE
ffmpeg -i "C:\Users\mohan\OneDrive\Desktop\video.mp4" -vn -acodec libmp3lame -q:a 2 "C:\Users\mohan\OneDrive\Desktop\audio.mp3"
python faster-whisper/whisper.py --model small --device cuda --output_dir "C:\Users\mohan\OneDrive\Desktop\output" "C:\Users\mohan\OneDrive\Desktop\audio.mp3"
python convert.py "C:\Users\mohan\OneDrive\Desktop\output"

If the video is public, remove the --cookies flag and run:

yt-dlp -o "C:\Users\mohan\OneDrive\Desktop\video.mp4" https://www.youtube.com/watch?v=Kbn2ab0-sGE

---

Notes

• Replace placeholders (<...>) with actual file paths.
• Use the cookies method only if the video is private or unlisted.
• Make sure your cookies file is up to date for YouTube authentication.
• Use a larger Whisper model (medium, large-v2) for better accuracy.
• If you don’t have a GPU, set --device cpu in the transcription step.