Help Center

Voice ID and Speaker Names

Name who said what once, and Wave recognizes those voices automatically in every recording after that.

Last updated:

Introducing Voice ID: Wave Learns Who’s Talking

Wave AI splits every recording into separate speakers, but it doesn't know their names, so you get "Speaker 1" and "Speaker 2". Name them once, save them as Voice IDs, and Wave labels those people automatically in every recording after that.

Three ways a speaker gets a name

  • You type it. Renaming a speaker replaces that label everywhere in the transcript.
  • Wave guesses it. Wave can infer names from the conversation itself ("Thanks, Sarah"). The button is Guess speaker names on iPhone and Android, Suggest names on the web, and Identify speakers on desktop.
  • Wave recognizes the voice. With a saved Voice ID, Wave matches the voice while the recording processes and applies the name before you open the session. Those speakers are marked Matched. On iPhone and Android you also see how close the match was: Very high match, High match, or Moderate match.

Matching runs on its own for new recordings while automatic voice matching is on. It's deliberately conservative: a speaker needs at least 10 seconds of audio, and Wave leaves someone unlabeled rather than commit to a weak match.

Naming speakers in a session

Open the session and tap Speakers in the row of action buttons (it's also in the ⋯ menu). Tap the play button beside a speaker to hear them, then type into the Enter name field, or tap Guess speaker names and let Wave suggest them. Turn on Update summary with speaker names to have the summary rewritten with the real names, then tap Save.

When some speakers are already identified, the guess button reads Guess 2 remaining; that number is how many still have generic labels. Speakers Wave named from the conversation show an Identified badge.

You can also tap a speaker's name on any line of the transcript, or press and hold a line and choose Who is this?. A Who is this? sheet opens with Play and Next clip so you can hear them, plus Rename, Merge into… and Not the same person. If they have a Voice ID, Open their page takes you to them in People.

Saving voices as Voice IDs

Naming a speaker fixes one session. Saving a Voice ID fixes every session after it.

A Save voice IDs banner appears on the session: "Name speakers and Wave will recognize them in future recordings." Tap Name speakers (or Enroll voices) to open Speakers.

Once you save names, a Save Voice IDs? sheet appears. Tick the speakers to keep, mark yourself with This is me, and tap Save Voice IDs (or Later). Anyone already saved shows "Already saved. This sample improves recognition." Saving them again strengthens their profile.

Wave needs at least 10 seconds of a person's speech to build a profile, and uses up to 30 seconds of their clearest segments. Speakers with less audio are skipped.

Merging duplicate speakers

Long or noisy recordings sometimes split one person across two labels. Merging folds one speaker's lines into the other.

  • iPhone and Android: in Speakers, tap the merge icon on a speaker's row, pick who to merge into, and confirm with Combine. Or tap a speaker's name in the transcript and choose Merge into….
  • Web: in the Speakers dialog, click the merge icon on a row and choose Merge into a name, then confirm with Combine. Or click a speaker's name in the transcript and choose Merge into….
  • Desktop: in the Speakers dialog, open a speaker's menu and choose Merge into. Or click a speaker's name in the transcript and use Merge into on the Who is this? card.

Changed your mind? Undo last change in Speakers (the Speakers dialog on the web and desktop) reverses the most recent rename, merge, split or speaker change. On iPhone and Android, each of those edits also shows a message with Undo for a few seconds.

To move individual lines to a different person instead, see editing summaries and transcripts.

Splitting one label into two people

The opposite problem: two people ended up under a single name, usually because they sound alike or spoke over each other. Pick out the lines that are really the second person and split them into their own speaker. The rest of that label's lines stay where they are.

Open the session's Transcript tab, press and hold a line that belongs to the other person, and choose Select lines. Tap every other line that belongs to them (press and hold a second line to select everything in between). Tap New speaker, type the name, and tap Save. If that person is already in the session, use Move to… instead. Done leaves select mode.

For a single line, press and hold it and pick Speaker: name or Split into new speaker…, or tap the speaker's name and choose Not the same person.

Everything you select has to sit under the same speaker to begin with; that's the label being split. Once the second person has their own name, save them as a Voice ID and Wave keeps the two of them apart in future recordings instead of you fixing it each time.

Managing your Voice IDs

Your saved voices live in People, along with every recording each person is in. Open someone there to rename them, play their sample, merge duplicates, mark yourself with This is me, or Forget voice. The switch that turns matching off stays in Settings.

Tap the People chip on Home, or go to Settings → Transcription → People. Voice ID settings, just below it, holds the Automatic voice matching switch and Tidy up.

Training tells you how well a profile knows someone's voice. A voice that's still learning has limited audio and is recognized inconsistently; a trained voice holds up in most conditions. Profiles improve on their own: when Wave recognizes someone with high confidence, that recording's audio is folded into their profile, weighted so recent audio still counts. Turn off Learn from new recordings on a person's page to stop that for them.

Voice IDs live on your account and sync everywhere, so a name you save on your phone is recognized on desktop too.

Voice profiles are stored securely, used only to identify speakers in your recordings, and you can delete them at any time. A profile holds a numeric voice fingerprint plus a short audio clip; the clip is what the play sample button plays back.

FAQ

Was this article helpful?

Still need help?

Email support

On this page