From Voice to Text: Create a Usable Transcript with ChatGPT and other LLMs

A delicate sound wave flows from a vintage microphone into lines on a cream page framed by sage-green plants.

You have an interview on your phone, an old family video, or a voice memo you want as text. An AI assistant can turn the spoken words into a transcript. You upload the existing file and ask it to write down what is said.

Upload your recording

Here is a straightforward example using Gemini, which supports both audio and video uploads:

  1. Open Gemini and sign in.
  2. Choose Add files → Upload Files beside the message box.
  3. Select the recording already saved on your device and wait for it to attach.
  4. Paste the prompt below and send it.

Start with a short clip. For a video, use the video file itself; the prompt asks for its spoken words. Google’s upload guide explains the current file and length limits. If your recording is too long, try a shorter section first.

You can also use this prompt in ChatGPT or another assistant when it supports transcription of your uploaded file. If your current chat cannot read the recording, use the Gemini steps above.

The prompt

Please transcribe all spoken words in the attached audio or video file.

Keep the original language and wording. Add punctuation and paragraphs so the text is easy to read. Do not summarize or translate it.

Use Speaker 1, Speaker 2, and so on when you can distinguish different voices. Write [unclear] wherever you cannot make out a word instead of guessing. Preserve corrections such as "1982—no, 1983."

For a video, transcribe the speech without adding descriptions of the pictures.

Return the transcript, followed by a short list of passages I should listen to again. If you cannot access the recording, tell me instead of inventing its contents.

Check it and save it

The result appears in the chat. Play back a few passages while reading along, especially names, dates, numbers, and anything marked [unclear]. Check the beginning and end too: a fluent answer is not proof that the whole recording was covered.

Make your corrections, then copy the text into Word, Google Docs, or a plain text file. Give it the same name as the recording, such as grandmas-first-job-transcript.docx, and save it beside the original. If the assistant offers a downloadable document, download that instead.

Keep the audio or video as well: the transcript captures words, while the recording preserves the voice. Upload recordings only when you have permission to share them with the service. For a family collection, The Memory Keeper adds a short description that helps you find each recording again.

Comments

Leave a Reply

Your email address will not be published. Required fields are marked *