Identify what the script does and who it is for.
Document prerequisites and installation/setup steps.
Explain configuration, API key handling, and environment variables.
Provide clear usage examples.
Describe outputs, workflow, troubleshooting, and security notes.
A Bash script that downloads audio from a YouTube URL, splits it into manageable chunks, transcribes the audio with OpenAI Whisper, and summarizes the transcription using the OpenAI Responses API.
- Downloads audio from YouTube using
yt-dlp - Converts/downloads audio as MP3
- Splits long audio files into 30-minute chunks using
ffmpeg - Transcribes each chunk using Whisper
- Combines the transcription into a single text file
- Sends the transcript to OpenAI for summarization
- Supports summary output in a target language
- Saves the final summary to a timestamped text file
The script requires the following tools:
bashffmpegjqcurlyt-dlp- Python 3
- Python package:
whisper - An OpenAI API key
git clone https://github.com/YOUR_USERNAME/YOUR_REPO_NAME.git
cd YOUR_REPO_NAME
2. Install system dependencies
On Debian/Ubuntu:
sudo apt update
sudo apt install -y ffmpeg jq curl python3 python3-venv
3. Install yt-dlp
You can install yt-dlp from your package manager:
sudo apt install -y yt-dlp
Or download the standalone binary:
curl -L https://github.com/yt-dlp/yt-dlp/releases/latest/download/yt-dlp_linux -o ~/yt-dlp_linux
chmod +x ~/yt-dlp_linux
By default, the script expects yt-dlp at:
/home/$USERNAME/yt-dlp_linux
You can override this with the YTDLP_BIN environment variable.
4. Set up Python and Whisper
The script expects a Python executable at:
$WORKDIR/bin/python
By default:
/home/$USERNAME/whisper/bin/python
Create the environment:
mkdir -p ~/whisper
python3 -m venv ~/whisper
~/whisper/bin/pip install -U pip
~/whisper/bin/pip install openai-whisper
Depending on your system, you may also need PyTorch installed separately. See the official PyTorch installation guide:
https://pytorch.org/get-started/locally/
OpenAI API Key
The script needs an OpenAI API key.
You can provide it in one of two ways.
Option 1: Environment variable
export OPENAI_API_KEY="your_api_key_here"
Option 2: Key file
By default, the script looks for a key file at:
$WORKDIR/.openai_key
For example:
echo "your_api_key_here" > ~/whisper/.openai_key
chmod 600 ~/whisper/.openai_key
Usage
./script.sh <youtube_url> [target_lang]
Example:
./script.sh "https://www.youtube.com/watch?v=VIDEO_ID"
Summarize in another language:
./script.sh "https://www.youtube.com/watch?v=VIDEO_ID" spanish
./script.sh "https://www.youtube.com/watch?v=VIDEO_ID" french
./script.sh "https://www.youtube.com/watch?v=VIDEO_ID" german
If no target language is provided, the script defaults to:
english
Configuration
The script supports the following optional environment variables:
Variable Description Default
USERNAME User used to build default paths Invoking user or current user
MODEL OpenAI model used for summarization gpt-5.4-mini
WORKDIR Working directory for audio, transcript, and output files /home/$USERNAME/whisper
KEYFILE File containing your OpenAI API key $WORKDIR/.openai_key
YTDLP_BIN Path to the yt-dlp executable /home/$USERNAME/yt-dlp_linux
OPENAI_API_KEY OpenAI API key. If set, this overrides KEYFILE unset
Example with custom paths:
WORKDIR="$HOME/whisper" \
YTDLP_BIN="$(which yt-dlp)" \
MODEL="gpt-4.1-mini" \
./script.sh "https://www.youtube.com/watch?v=VIDEO_ID" english
Output Files
The script creates and uses the following files inside WORKDIR:
File Description
input.mp3 Downloaded YouTube audio
output000.mp3, output001.mp3, etc. 30-minute audio chunks
transcription.txt Full Whisper transcription
summary-YY-MM-DD-HH-MM-SS.txt Final OpenAI-generated summary
Example output file:
summary-25-01-18-14-30-02.txt
How It Works
The script performs four main steps:
1. Download audio
Uses yt-dlp to download the audio from the provided YouTube URL and save it as input.mp3.
yt-dlp -x --audio-format mp3
2. Split audio
Uses ffmpeg to split the MP3 into 30-minute chunks.
ffmpeg -f segment -segment_time 1800
3. Transcribe audio
Uses Whisper with the turbo model to transcribe each audio chunk.
model = whisper.load_model("turbo")
The combined transcription is written to:
transcription.txt
4. Summarize transcription
Sends the transcript to the OpenAI Responses API and asks for a summary in the selected target language.
The summary is printed to the terminal and saved to a timestamped file.
Example
export OPENAI_API_KEY="your_api_key_here"
WORKDIR="$HOME/whisper" \
YTDLP_BIN="$(which yt-dlp)" \
./script.sh "https://www.youtube.com/watch?v=dQw4w9WgXcQ" english
Expected terminal output:
[INFO] Downloading audio...
[INFO] Splitting audio...
[INFO] Running Whisper transcription...
[DONE]
[INFO] Summarizing transcription with OpenAI gpt-5.4-mini...
The summary will be saved as something like:
summary-25-01-18-14-30-02.txt
Notes
The script deletes previous input.mp3, transcription.txt, and output*.mp3 files in the working directory before each run.
Audio is split into 1800-second, or 30-minute, segments.
Very long transcripts may exceed the input limit of the selected OpenAI model.
The default Whisper model is hardcoded as turbo inside the Python block.
The default OpenAI summarization model is set with the MODEL environment variable.
Troubleshooting
No OpenAI API key found
Set your API key:
export OPENAI_API_KEY="your_api_key_here"
Or create the key file:
echo "your_api_key_here" > ~/whisper/.openai_key
chmod 600 ~/whisper/.openai_key
yt-dlp not found
If you installed yt-dlp globally, run the script with:
YTDLP_BIN="$(which yt-dlp)" ./script.sh "https://www.youtube.com/watch?v=VIDEO_ID"
Whisper import error
Make sure Whisper is installed in the Python environment used by the script:
~/whisper/bin/pip install openai-whisper
ffmpeg not found
Install ffmpeg:
sudo apt install -y ffmpeg
OpenAI API error
If the script prints something like:
OpenAI API HTTP 401
Your API key may be missing, invalid, or expired.
If it prints:
OpenAI API HTTP 400
The transcript may be too large for the selected model, or the request payload may be invalid.
Security
Do not commit your OpenAI API key to GitHub.
Recommended:
echo ".openai_key" >> .gitignore
echo "input.mp3" >> .gitignore
echo "output*.mp3" >> .gitignore
echo "transcription.txt" >> .gitignore
echo "summary-*.txt" >> .gitignore
Example .gitignore:
.openai_key
input.mp3
output*.mp3
transcription.txt
summary-*.txt
/tmp/openai_body.json
Legal Notice
Make sure you have the right to download, process, transcribe, and summarize any content you use with this script. Follow YouTube's Terms of Service and all applicable laws.