Guide

Can ChatGPT transcribe audio? Yes, with a free plugin

Free transcripts with speaker turns and subtitle files, in either app, with no sign-in.

The short answer

Not by itself, but yes with a plugin. ChatGPT on its own does not hand back a transcript of a recording with speaker turns and subtitle files; add the free Slice plugin to ChatGPT or Claude, then ask it to transcribe a recording. In ChatGPT you can attach the file or paste a direct link; in Claude you paste a direct link. You get the text back in the chat with speaker turns, plus a page with .txt, .srt and .vtt downloads. It is free, and the first one needs no sign-in: up to 1 hour a file, then 2 a month once you sign in to Slice. It doesn’t take YouTube or Spotify links, because those are pages rather than files.

New to the plugin? Set up Slice in ChatGPT or Claude first. It takes a couple of minutes.

Why people struggle to transcribe in ChatGPT

Two problems come up again and again on OpenAI’s forum. Long recordings have to be split, and splitting breaks the timestamps:

“I need to transcribe audio files of up to three hours in length.”
mihailpet, OpenAI Developer Community, 15 May 2024
“The question is: How to seamlessly stitch together the resulting text chunks?”
mihailpet, OpenAI Developer Community, 15 May 2024
“If I split the video into chunks, the resulting srt file might have incorrect timecodes.”
jason123, OpenAI Developer Community, 4 October 2023

And ChatGPT’s own voice recording is built for dictating a message, not for a recording you care about keeping:

“A lot of the times, my minutes of talking just get lost due to the message “network error” as soon as I hit “stop record”.”
srs.music, OpenAI Developer Community, 18 February 2025
“Just spent 8+ minutes allowing my train of thought to flow in a message I was about to send, only to be hit with “ChatGPT didn’t quite catch that.”.”
mixxmasterr, OpenAI Developer Community, 12 May 2025

The plugin takes the recording as a file or a link, transcribes all of it in one go, and keeps the result on a page you can come back to.

A real test: an 82-minute episode in one go

On 2 October 2026 we gave the plugin a direct link to the MP3 of LinuxGameCast Weekly 430, a three-host podcast published under CC BY-SA 4.0. The file was 81.9 minutes, under the free limit at the time (90 minutes; since 30 October 2026 it is 1 hour), so it went as one file with no splitting.

  • Time: under four minutes, from the link to the finished transcript.
  • Length: 14,380 words, in speaker turns.
  • Cost: nothing. The first one needs no account.
  • Files: .txt, .srt and .vtt, from a page with the full text. The .srt has 1,461 subtitles.
The Slice transcript page for LGCWEP430.mp3: 82 min, 14,380 words, buttons to copy the text and download .txt, .srt and .vtt, a box to email the files, and the transcript in speaker turns starting "Speaker A: Not the longest we've done in the past couple of weeks".
The free transcript of the same episode: 82 minutes, 14,380 words, speaker turns, and .txt, .srt and .vtt downloads. Look at the first lines: the show's name comes out as "Linux game, Linux, gamecast" and "Linux" as "Lennox". Tap to enlarge.

Did the timestamps hold across 82 minutes?

This is the part that goes wrong when a recording is split, so we checked it against an independent clock: Slice’s clip finder, which transcribed the same episode separately when we made clips from the video.

The lineIn the free .srtClip finder said
“So gyms have been closed.”1:191:20
“…how many false claims I currently get on our YouTube channel…”49:3849:37
The last subtitleends 1:21:52episode ends 1:21:55

Within about a second, an hour in, and no drift at the end. We spot-checked these lines, not all 1,461.

What it got wrong

Names. In the first minute the show’s own name came out as “Linux game, Linux, gamecast”, and “Linux” as “Lennox”. If you are publishing a transcript, search it for your show’s name and your guests’ names first; those are the words most likely to be wrong.

How do you transcribe audio in ChatGPT or Claude?

  1. Add the Slice plugin

    In ChatGPT, open Plugins, choose Add, then Create MCP App; in Claude, add a custom connector. Either way the address is https://mcp.tryslicemedia.com/mcp. Transcription needs no Slice account.

    Screens for both apps are in the setup guide.

  2. Give it the recording

    In ChatGPT, attach the audio or video file, or paste a direct link to it. In Claude, paste a direct link: a link that downloads the file itself, not a page that plays it.

    ChatGPT, with a file attached@Slice transcribe this
    ChatGPT or Claude, with a linkTranscribe this recording: https://example.com/episode-42.mp3
  3. Wait for the text

    Short files come back in the same reply. Longer ones take a minute or two: the assistant says it is still transcribing, and you can ask it to check.

    Is it done?
  4. Use the transcript

    The transcript arrives in paragraphs with speaker turns, plus a link to its own page with the full text, a copy button and .txt, .srt and .vtt downloads. Ask the assistant to summarise it, quote it or search it in the same chat.

    Summarise this in five bullet points and pull out three quotes I could post

What is different between ChatGPT and Claude?

ChatGPTClaude
Attach the file in the chatYesNo: paste a direct link
Paste a direct linkYesYes
Sign-in neededNot for the first oneNot for the first one
Limits1 hour a file, 2 a month1 hour a file, 2 a month
Results page with .txt, .srt, .vttYesYes
Email the files to youYes, signed in, to your account addressYes, signed in, to your account address

Which files and links work?

Files: MP3, M4A, WAV, FLAC, OGG, AAC, MP4, MOV, WebM and MKV, up to 1 hour long.

Links have to download the file itself. A good test: open the link in a browser tab. If it starts playing or downloading the file on its own, it works. If you see a page with a player on it, it won’t.

LinkWorks?
The MP3 address from your podcast’s RSS feedYes
A link ending in .mp3, .m4a, .wav or .mp4 that downloads the fileYes
A YouTube, Spotify or Apple Podcasts pageNo: these are pages, not files
A live stream or a call in progressNo: it transcribes recordings, not live audio

Accuracy is best in English.

What can you do with the transcript in the chat?

The point of transcribing inside an assistant is that the words are then in the conversation. A few things worth asking:

Write show notes for this episode
Find the three moments most likely to work as short clips
Email the transcript files to me@example.com

For a long episode the chat may only get the first part of the text; the results page always has all of it. If the transcript is from a video and you want the clips themselves, not just ideas for them, that is the paid half of the same plugin: making podcast clips with ChatGPT.

What if you hit a limit?

What happenedWhat to do
The file is over 1 hourSplit it into parts of 1 hour or less and send each one.
It asks you to sign inThe first one needs no account. Sign in to Slice for the rest of the month's.
You have done 2 this monthThey reset on the 1st, counted together with the Slice website. Hour packs cover more.
The link was refusedIt is probably a page, not a file. Find the direct file link, or attach the file in ChatGPT.

Prefer a browser tab? Free transcription is also on the website, as audio to text and audio to SRT.

Questions

Is the transcription really free?

Yes. The first one needs no account. After that you sign in to Slice, and you get 2 a month, shared with the Slice website. Each file can be up to 1 hour.

Can I transcribe a YouTube video or a Spotify episode?

Not from the page link. YouTube, Spotify and Apple Podcasts links are web pages, not files, and are refused. If it is your podcast, the MP3 address in your RSS feed is a direct link and works.

Why does Claude need a link when ChatGPT takes a file?

Claude passes links to connectors rather than files you attach in the chat. So in Claude, put the recording somewhere that gives a direct download link and paste that.

How accurate is it?

Usable, but read it before you publish. In our 81.9-minute test, the show's name came out as "Linux game, Linux, gamecast" and "Linux" as "Lennox" in the first minute. Fast, mumbled cross-talk came out garbled in places too. Read the names, and any line you plan to quote, before you publish.

Does it label who is speaking?

Yes. When there is more than one voice, the transcript is split into speaker turns.

Can I get subtitles for a video?

Yes. Every transcript has its own page with .srt and .vtt downloads, which video editors and web players read, and a .txt.

What happens to my recording?

Slice does not keep a copy of your audio. The transcript is deleted after 30 days, and you can delete it sooner from its page.

Get The Cut

What's working in short-form video, one short email every Tuesday.

Unsubscribe any time. Privacy

More guides

All guides

Rather do it in a browser tab?

Free for 2 recordings a month, each up to 1 hour. You get .txt, .srt and .vtt files.

Want clips too? Try Shorts