Transcribe Voice Memos with speaker labels
Voice Memos is where most conversations on an iPhone get recorded: an interview, a meeting, a lecture, a call on speaker. Its own transcript gives you the words, but not who said them, so a two-person interview comes back as one block of text. Cadence takes the memo as it is and gives it back as turns, Speaker 1, Speaker 2, each with its time, so you can tell the questions from the answers.
Everything below runs on your own device. The memo is not uploaded, there is no account, and speaker labels are in the free app.
On iPhone or iPad
- Install Cadence from the App Store. It needs iOS or iPadOS 26 or later.
- Share the memo from Voice Memos. Open the recording in Voice Memos, tap the ⋯ button, choose Share, and pick Cadence in the share sheet. Cadence keeps its own copy of the file, so the memo in Voice Memos is left alone. Open Cadence and the memo is there, opened and ready.
- Choose how to transcribe it. Tap Transcribe… and pick the model, the language and the number of speakers. Two models are free: Apple Speech, the default, and Parakeet v3 compact, which covers 25 European languages. Leave speakers on Auto, or set the exact number if you know it: an exact count is the best help when two voices sound alike.
- Read it back by speaker. The transcript is a list of turns with timestamps. Tap a turn to play the memo from there; the transcript follows along as it plays.
- Put names on the speakers. Tap a speaker in the row above the transcript and type a name, so Speaker 1 becomes Interviewer. Renaming is free and applies to that recording only.
- Export it. Export, beside the title, writes plain text or SRT subtitles, or copies the transcript, for free. WebVTT, Markdown, JSON and CSV come with Cadence Pro.
To skip step 3 next time, turn on Transcribe shared recordings in Cadence's Settings, under Sharing (off by default). A memo you share then opens already transcribing, whenever your default model is downloaded and nothing needs choosing.
On a Mac
Drag a recording straight out of Voice Memos onto Cadence's window, or use Voice Memos' Share menu. Cadence names it after the memo, not after the file Voice Memos exports, and the rest works as on iPhone: pick a model, set the speakers, read the turns and rename them. The Mac needs macOS 26 or later on Apple Silicon.
What to expect from the speaker labels
- Cadence tells voices apart; it does not know who they belong to. Speakers come out as Speaker 1, Speaker 2 until you name them.
- Clear audio with people taking turns works best. Two similar voices can be merged into one speaker; setting the exact number of speakers is the fix that helps most.
- For recordings where people talk over each other, Cadence Pro adds NVIDIA's Nemotron speaker model, which follows overlapping speech for up to eight speakers.
- With Cadence Pro, the speaker library can suggest a name for a voice you have named in another recording. It only suggests: nobody is named until you confirm it.
Why not just use the Voice Memos transcript?
If you recorded yourself, it is all you need. Once a second person is on the recording, the question that matters is who said what, and that is the part Cadence adds. It also gives you a choice of models, timestamps for every turn, and subtitles you can export.
Is it really free?
Yes. Transcribing whole memos, speaker labels, renaming, playback, search and plain-text and SRT export are free, with no time limit and no cap on minutes. Cadence Pro is a one-time purchase that adds the larger models, the Nemotron speaker model, briefs, the speaker library and more; see pricing.