Extract Text from A Specific Portion Of A YouTube Video On Linux

Sometimes you may only need text from a small part of a long YouTube video, maybe a section from a talk or a clip from a podcast. Downloading the entire video is a waste of time. But with the right tools, you can pull out only the part you want and turn it into text quickly.

In this post, I'll show you how to extract text from a specific portion of a YouTube Video using free tools such as yt-dlp, ffmpeg, and Whisper.

Everything runs right on Linux, and my setup stays clean thanks to pipx, which installs Whisper without cluttering the system.

Why I Use This Method

I like to keep control over my data. Most web services for transcription upload files somewhere I don't know. With free, open source command-line tools, I work locally and keep everything private.

This method lets me:

• Download only the audio from a YouTube video.
• Cut out just the section I care about.
• Turn that audio into text I can edit, quote, or summarize.

The whole transcription process is very simple, reliable, and works entirely offline. Let us get to work.

WARNING: Whisper and its dependencies can take up significant space (around 5–8 GB during installation). The first time you run it, it may briefly connect to download the model files. After that, it runs completely offline. Your data (audio) never leaves your computer. If you're on limited data or a slow connection, consider using lighter alternatives like faster-whisper or online transcription tools instead.

Step 1: Install the Necessary Tools

As I already mentioned, I am going to use the following open source tools to transcribing audio from a YouTube video i.e extracting text from YouTube videos:

• Yt-dlp
• FFmpeg
• Whisper
• Pipx (to install Whisper)

Here's what each tool does in our case:

Let us install these tools.

In Debian and Ubuntu-based systems, you can install these tools like below:

sudo apt install yt-dlp ffmpeg pipx
pipx ensurepath

Next install Whisper through pipx, so it doesn't scatter files across my system:

pipx install openai-whisper

pipx keeps Whisper and its dependencies inside their own isolated environment under ~/.local/pipx/. That means you can remove or upgrade it anytime without touching your main Python setup.

Now you have these commands available everywhere:

• yt-dlp
• ffmpeg
• whisper

Step 2: Download the Audio Only from a YouTube Video using Yt-dlp

To grab audio from YouTube, you can use yt-dlp:

yt-dlp -f bestaudio -x --audio-format mp3 -o "video_audio.%(ext)s" "https://www.youtube.com/watch?v=xxxx"

Replace "https://www.youtube.com/watch?v=xxxx" in the above command with your actual URL.

This downloads only the audio stream, converts it to MP3, and saves it as video_audio.mp3.

Even though you only need a part of the video, yt-dlp must download the full audio file because YouTube doesn't allow partial time-range downloads. But that's fine, because the next step trims it down in seconds.

Step 3: Trim the Exact Portion with FFmpeg

Let's say you need the section from 1:52:00 to 2:30:00 from the output audio file i.e. video_audio.mp3 in our case.

To do so, you can use ffmpeg like below:

ffmpeg -ss 01:52:00 -to 02:30:00 -i "video_audio.mp3" -c copy clip.mp3

This creates a smaller file called clip.mp3 that contains only that time range.

The -c copy option tells ffmpeg to skip re-encoding, so the process finishes almost instantly even for long recordings.

Note: If you want to transcribe the whole YouTube video, simply ignore the STEP 3.

Step 4: Convert Speech to Text using Whisper

Now use Whisper to transcribe the audio clip that we trimmed in step 3:

whisper clip.mp3 --model base --language en

I got good results with base model. You can also use some other models like tiny, small, medium, large. Check out the Whisper Github page for available models.

whisper clip.mp3 --model small --language en

Whisper produces several files:

• clip.txt: plain text transcript
• clip.srt and clip.vtt: subtitle files with timestamps

It runs entirely offline and is surprisingly accurate.

If you want GPU acceleration or faster results, you can install Faster-Whisper with pipx as well:

pipx install faster-whisper
faster-whisper --model medium clip.mp3

Both versions work great, but the regular Whisper model is often enough for short clips.

As I mentioned already, Whisper produces many files. If you want to save the text in one file, you can use this command:

whisper clip.mp3 --model base --language en --output_format txt --output_dir .

Please note the dot (.) at the end. This command creates a file named clip.txt containing the transcribed text. Here's what each part of the command does:

• --model base: uses the small, fast Whisper model.
• --language en: forces English mode for better accuracy.
• --output_format txt: generates a plain-text file.
• --output_dir .: saves it in the current folder.

Step 5: Transcribe YouTube Videos using a Script (Optional)

If you do this often, you can use a small script to handle all the steps. Save the following code in a text called yt_extract_text.sh:

#!/bin/bash
url=$1
start="01:52:00"
end="02:30:00"

yt-dlp -f bestaudio -x --audio-format mp3 -o temp.mp3 "$url"
ffmpeg -ss "$start" -to "$end" -i temp.mp3 -c copy clip.mp3
whisper clip.mp3 --model small --language en

Replace the start and end times, audio format, output file name, model and language in the script.

Then make it executable:

chmod +x yt_extract_text.sh

To use it, simply type:

./yt_extract_text.sh "https://www.youtube.com/watch?v=xxxx"

The script downloads, trims, and transcribes automatically. You will end up with a neat .txt file containing just the part you wanted.

FAQ: Extracting Text from YouTube Videos on Linux

Conclusion

I use this method often to collect quotes and notes from long YouTube videos. I no longer waste time transcribing by hand or relying on online tools that store my data. Everything happens locally, and I get high-quality text from exactly the part I need.

If you watch long lectures, podcasts, or interviews, this setup can save you hours. Try it once, it's simple, clean, and it just works.

Related Read:

• How To Extract Text From Screenshots And Images In Linux

Featured image by Pixabay.