Does an iPhone transcript show who is speaking?

This usually surfaces the first time someone transcribes an interview or a meeting rather than a single voice talking to itself. The words are often close to right. The part that is just missing, with no error message and no obvious toggle to fix it, is any indication of who said which part.

Two different problems that look like one

Turning speech into words is one task: a model listens to a stretch of audio and predicts the sequence of words that produced it. Working out how many distinct voices are in a recording and which stretches belong to which voice is a different task entirely, usually called speaker diarization. It does not ask what was said — it asks who was talking, based on things like pitch, timbre and the acoustic fingerprint of a voice.

A finished transcript makes these look like one job because you want both answers at once. Apple's Speech framework, which is what on-device and server transcription on iPhone both run on, only does the first one. It gives you words. Whether that audio had one voice in it or five, the output is a plain block of text — nothing in the framework identifies or separates speakers, and no third-party app calling it can make it do something it does not do.

What this looks like when you hit it

A transcript of a two-person interview reads as a single paragraph running straight through both the questions and the answers, with no "Q:" or name to mark the switch. Overlapping speech — someone talking over someone else — comes out as whatever words the model could separate, with no note that two people were involved at all. If you were not there, or cannot remember the conversation, there is nothing in the text itself to tell you who said which line.

What actually works instead

Say names out loud, on tape

The cheapest fix happens before transcription even starts. If you restate a name or a role at the start of a turn — "so, Priya, what happened next" — that phrase ends up in the transcript as ordinary words, and it does the labelling job a diarization feature would otherwise do. It costs nothing and needs no app support at all.

Mark speaker changes while you record

Tapping a marker at the moment someone new starts talking leaves you a set of timestamps to work from afterward, the same habit that helps with a lecture or a long interview for entirely different reasons. When you go back to edit the transcript, those markers tell you exactly where to insert a name, without having to relisten to the whole recording to find the switches.

Edit the transcript once, deliberately

Because the transcript is ordinary editable text, the practical fix is a single pass afterward: play the audio back at speed, and where the markers or the content itself tell you the speaker changed, type the name in. For a short interview this is minutes of work, not hours, and it only has to happen once per recording.

Record separately when it actually matters

For anything where getting the attribution wrong would be a real problem — a legal conversation, a formal interview you plan to quote from directly — the more reliable fix sidesteps the transcript entirely: have each person record on their own device. A single phone with one microphone has no way to physically separate two voices in the room; two separate recordings do not have that ambiguity to begin with.

Where this matters most

It comes up hardest exactly where transcripts are most useful: interviews, meetings, and any conversation you want to quote from later rather than just search. None of those situations are made worse by the lack of speaker labels — they just need the small amount of manual attribution above to turn a wall of text into something you can cite with confidence.

How Voice Studio handles it

Voice Studio uses Apple's Speech framework for transcription, so what it produces is exactly what is described above: accurate words, no automatic speaker labels. The transcript is ordinary editable text, so adding names or a "Q:"/"A:" pattern after the fact takes one pass. Markers tapped during recording are there to guide that pass — a quick way to remember where the speaker changed without relistening to the whole file. Once a transcript is the way you want it, it exports as TXT or JSON, or you can hand off the audio itself as the .m4a file.

Common questions

Can any iPhone app add real speaker labels automatically?

Some apps run additional processing on top of the transcript to guess at speaker changes from voice characteristics. It is a genuinely different feature from speech-to-text itself, and it tends to struggle when voices are similar or the audio is not clean — it is a guess, not a guarantee, in a way plain word transcription is not.

Does recording in stereo or with two microphones fix this?

Not by itself. Separating voices reliably needs either genuinely separate recordings per person or a diarization step built specifically for that job — simply capturing more channels does not label who is speaking on its own.

Why does the transcript get names right when people say them out loud?

Because at that point it is just transcribing ordinary words. "This is Priya" comes out as text the same way any other sentence does — the labelling is happening in what was said, not in anything the transcription feature is doing on top of it.

Does this affect one-person recordings like a lecture or a voice journal?

No. The lack of speaker labels only matters once there is more than one voice to tell apart. A single speaker has nothing to disambiguate.