YouTube videos
In general, all the methods below rely on the same underlying source: YouTube's own captions either uploaded by the creator or auto-generated by YouTube. Nothing here independently "listens" to the audio; the caption text is what everything is built on. This matters because caption quality (or its absence) sets a ceiling on how good any of these results can be.

Gemini
Recommended - The only major consumer LLM that extracts and transcribes the video itself

• Paste the YouTube link directly into a chat, no manual steps needed.
• Can produce several different output formats depending on what you ask for: a verbatim transcript, a verbatim transcript with speaker labels, a literary-edited (cleaned-up) transcript, or a brief summary of general topics.
• It still relies on the underlying captions, but it can also correlate them with the video itself to add speaker labels — something the captions alone don't provide. With several speakers, this correlation can still get it wrong.

Manual extraction + other LLMs (Claude, ChatGPT, DeepSeek, Grok)

1. Open the video on YouTube.
2. Below the video, click the "...more" (or "...show more") text under the description.
3. Click "Show transcript".
4. The transcript appears in a panel next to the video, with timestamps for each line.
5. Select all the text and copy it, or use the three-dot menu on the transcript panel to toggle timestamps on/off before copying.
6. Paste the copied text into any LLM for summarizing or analysis.

Notes and limitations:

• Only works if the video has captions available at all.
• Auto-generated captions can have mistakes: no punctuation, misheard words, no indication of who's speaking. Quality varies a lot depending on audio clarity and accent.
• YouTube's captions do not distinguish between speakers — an interview with several people comes out as one continuous stream of text with no speaker labels. If you need to know who said what, you'll need a separate diarization step (see the Whisper note below) or to mark speakers manually.
• If a video has no captions at all, this method won't work — you'll need Whisper (see below) after downloading the audio.

Once you have the transcript, paste it into Claude, ChatGPT, DeepSeek, or Grok (only for English texts) to get summary or main topics of the audio.

Local files - Whisper
Whisper (OpenAI's open-source speech-to-text model) turns any local video or audio file into a text transcript, entirely offline and free — useful when there are no existing captions to fall back on.

Buzz Desktop (Free, Open Source) - Recommended
Website: Buzz website
GitHub: Buzz GitHub repository

How to use:

1. Download and install Buzz.
2. Open the app.
3. Drag and drop your audio file.
4. Choose a Whisper model (e.g. medium or large-v3 for the best accuracy).
5. Click Transcribe.
6. Export the transcript as TXT, SRT, or VTT.

Whisper Desktop
GitHub: Whisper Desktop GitHub repository

How to use:

1. Download the latest release.
2. Install and launch the application.
3. Open your audio file.
4. Select the language (or Auto Detect).
5. Click Transcribe.
6. Save the transcript.

Notes:

• For long interviews or meetings, I'd recommend Buzz. It's actively maintained, works offline, supports speaker identification, and provides a polished interface powered by Whisper.
• Model size is a trade-off: smaller models (tiny/base) are faster but less accurate; larger ones (medium/large) are slower but much more accurate.
• Whisper doesn't distinguish speakers. If you need to know who said what, that requires a separate diarization step (e.g., pyannote.audio).

OpenAI ChatGPT
On paid accounts, an option to upload a video directly is gradually starting to appear, but it's not yet widely or reliably available — access varies by plan, app, and rollout stage.

Python — works for YouTube, other platforms, and your own local audio files alike.

Core tools:

• yt-dlp — downloads audio (and metadata) from YouTube and hundreds of other sites.
• Whisper — transcribes the downloaded (or local) audio.

This is the most scriptable option — it handles single files, whole playlists, and metadata export in one run, without depending on any platform's built-in captions.

See the full working version here: How to parse YouTube videos