Skip to content
Sonorarium
Back to blog

Can ChatGPT Transcribe an Audio File? Gemini and Copilot Too

What ChatGPT, Gemini and Copilot do with an audio file, per their official help pages (Nov 4, 2026): plans, limits, data, and when a transcriber fits better.

Sonorarium Team 6 min read

Short answer: sometimes, depending on your plan. Per their official help pages, ChatGPT takes audio up to 512 MB on paid plans only, Gemini up to 10 minutes (3 hours with Google AI Pro or Ultra), and Copilot lists no audio formats. For short audio, try them first. For a long file, SRT or an editor, Sonorarium gives 15 free minutes when you create your account, with no card. We don't separate speakers, translate, or accept links.

These features change fast. What follows is what each service's help pages said on November 4, 2026; before you upload anything important, check the current page.

What each assistant's official help says

Assistant Takes audio? Limits, per its help page Your data
ChatGPT Yes, on paid plans. Not on Free "at this time" 512 MB per file. Long recordings may be processed in sections, and very long ones may time out On Free, Plus and Pro it may be used for training unless you turn that off in data controls. On Business, Enterprise and Edu, not by default
Gemini Yes: its help sets an audio limit 10 minutes of audio in total (3 hours with Google AI Pro or Ultra). General file-upload limits: 100 MB per file and 10 files per prompt People review a share of conversations; Google asks you not to enter confidential information
Copilot Not promised: the help lists documents, images and text 50 MB per file, 20 files per conversation Stored for up to 18 months and not used for training

ChatGPT. Its help page is the most explicit. It says you can ask ChatGPT to transcribe a recording and then ask follow-up questions about it. It lists WAV, MP3, OGG, FLAC, AAC, M4A and PCM, plus WebM or MP4 only when they are audio-only (files identified as video are not supported). It warns that transcripts may contain errors, that speaker identification may be unreliable, and that performance varies by language.

Gemini. The upload help page gives audio limits but doesn't describe transcription as a feature or say how literal the result is. It is the only one of the three whose audio limit doesn't require a paid plan: 10 minutes in total. If your recording is shorter, it's a reasonable first attempt.

Copilot. The two Microsoft pages we read (file upload and supported formats) don't include audio. That doesn't mean it never will, but today they don't promise it. Microsoft does describe two other routes for audio: Transcribe in Word, with 300 minutes a month of uploaded audio on a Microsoft 365 subscription, and Record in the Copilot mobile app, for work accounts with a Microsoft 365 Copilot license, up to 120 minutes per recording.

Risks of asking an assistant for a word-for-word transcript

  • It may not be verbatim. An assistant is a language model built to hold a conversation, and when you ask it to "transcribe" it may end up summarizing, reordering or tidying sentences. We haven't measured this and no help page claims it, so treat it as a precaution: ask explicitly for "a verbatim transcript, without summarizing or correcting," and compare a stretch with the audio before you trust it.
  • Errors and speakers. OpenAI recommends checking important details against the recording and warns that who said what may be wrong.
  • Long files. ChatGPT works in sections and may time out; Gemini's 10-minute cap means splitting the file.
  • Formats. The help pages don't mention timestamps or SRT. If you need them, ask for them and check they match the audio.
  • Your data. The audio goes to the assistant's provider: see the last column of the table. If you later paste the text into an assistant, the same rules apply to the text.

When a transcriber fits better

If your recording is a few minutes long, your assistant takes it and you don't need SRT, stay with the assistant. A transcriber pays off when:

  • The file is long. On Sonorarium you upload one file of up to 5 hours and 2 GB, without splitting it. You see the first 15 minutes free, when you create an account and with no card, and unlock the rest by paying only for what's missing.
  • You want timestamps and to edit alongside the audio. Each segment carries its minute; click it and that stretch plays, so you fix the text next to the audio.
  • You need to export. Word, TXT, SRT and VTT.
  • You care about your data. Audio is deleted after 7 days, the transcript stays in your account until you delete it, and it is never used to train models (privacy).

A worked example: a voice note of a few minutes fits inside the 15 free minutes when you create an account, with no card. A 60-minute meeting on a new account needs 45 more: the 120-minute pack costs $6 (packs start at $2 for 30 minutes; all of them on the pricing page).

The workflow that works best

  1. Upload the audio to the transcriber: in a few minutes you have the text with timestamps. If you're starting from scratch, see how to convert audio to text, and for long recordings, how to transcribe a long audio recording.
  2. Check what matters (names, figures, dates) by clicking the timestamp of each sentence.
  3. If you want a summary, ask for it here. The "Summary and outline" panel gives a short summary and an outline of topics with their minute. It is done by Claude, from Anthropic, only when you ask: it receives the text, never the audio or your email. You get 1 free summary per account, for a complete transcript of 3 minutes or more; with any pack, the summary is included on all your transcripts. It can contain mistakes, so check it against the text. Details in how to summarize a recorded meeting.
  4. Or paste the text into your assistant and ask for whatever you need: an email, a table, answers to questions. It then works on text you've already checked, not on the audio.

If you already pay for Microsoft 365, we walk through the Word route step by step in how to transcribe audio in Word or Google Docs. And if you'd rather the audio never left your computer, see how to use Whisper for free on your computer.

FAQ

Is ChatGPT free for transcription? According to its help page, audio uploads aren't available on the Free plan "at this time." With Gemini, the limit is 10 minutes of audio in total.

Which one is more accurate? We haven't measured it and the help pages don't compare. Try two minutes of your own audio and compare with what you hear. If you're after another free tool, here are the limits of each free transcription tool.

Does Sonorarium translate or separate speakers? No. It transcribes in the language of the audio and doesn't label who is speaking; you add that while you review.

Sources and date

All consulted on November 4, 2026: OpenAI help on uploading files and audio and on how your data is used; Gemini help on uploading files and its privacy hub; Microsoft help on file upload in Copilot, supported formats, Record in Copilot and Transcribe in Word. This information may have changed since that date.