Speech to text

AI Speech to Text turns recordings into text: meetings, interviews, podcasts, lectures, calls, and videos. The transcript shows who said what and when, and you can turn it into summaries, notes, and posts with AI Content.

​

Transcribe a file or link

Then read and work with the transcript.

​

YouTube videos

With Detect speakers on (the default), the video’s audio is transcribed like a file: speakers, times you can play from, and subtitle downloads. It takes a minute or two longer than using the captions. If the audio can’t be downloaded (a live stream, a video longer than 6 hours, or a download YouTube refuses), the transcript uses the video’s captions instead and has no speakers; the page tells you when that happens. With Detect speakers off, a video with captions is ready in seconds from its captions, as plain text without speakers or subtitle downloads. When the captions mark who is talking, the transcript shows each turn on its own line. Videos without captions are always transcribed from their audio.

​

Your transcriptions

AI Speech to Text lists your transcriptions, 20 per page, newest first. Search by name finds one. Each shows its source, type and date, and its status when it isn’t done yet: Processing, No speech, Failed, or Draft (a file or link you picked but never sent to transcribe). Click one to open it. The ⋯ menu on a row opens it in a new tab, copies its link, or deletes it. A transcription that is still processing can be deleted two hours after it started. On a Teams plan, transcriptions teammates share with the team appear here too, with the teammate’s name.

​

If a transcription fails

• Failed items are retried automatically. You can also open one and click Retry transcription; your file is kept.
• No speech means no speech was found, for example a recording with the microphone muted. Check the recording and upload it again.

​

Dictate into a text box

The microphone button in Marky, the tools, and AI Images types what you say into the text box. Dictation doesn’t use transcriptions.

​

What it costs

Each new item uses one transcription from your plan, whatever its length. An item counts when you pick a source or start a recording, even if you don’t finish it or delete it later. Retries don’t count again. Check your usage on Settings > Usage and see Audio plans and limits. AI Content and AI Chat use words.

​

For developers

Transcribe from your own code with POST /api/transcriptions, which needs a paid plan. See Create transcription.