Speaker Auto Identification

Alter now uses voice fingerprinting to automatically recognize speakers.

AuthorAuthor

Samuel ROY & Olivier Legris
September 25, 2026

Hey everyone,

Alter now automatically recognizes who's speaking in your meetings, using voice fingerprinting.

No calendar event, and no "allow me to introduce myself" required. It listens to the voice and already knows it.

How it works
Get the latest version of Alter, then associate a speaker with a name, the fingerprint is created automatically.

Double-click any speaker to see the blue icon, which confirms a fingerprint exists for that voice.

Open a previous recording and click Regenerate with local to build a new fingerprint. Once Alter has recognized your voice, every future meeting is labeled correctly.

Why it stays yours
Everything runs on local transcription, so your voice and your recordings never leave your machine.

Use cases

  • Clean transcripts: correctly attributed minutes even when the meeting has no calendar link and no one introduces themselves.

  • Follow-ups & action items: know exactly who committed to what, without guessing.

  • Search across meetings: jump straight to the right speaker for "everything I said about X."

  • Privacy by design: fingerprinting, like transcription, happens locally , nothing is sent to the cloud.

Available now in Alter.

Cheers,

Sam & Olivier

Full Changelog

New Features & Enhancements

  • Local speaker identification: Nemotron diarization with Pyannote voiceprints identifies speakers entirely on-device, matching enrolled voices after a single sample.

  • Speaker name picker: Pick or confirm speaker names from contact suggestions, with persistent colors and a forget button when Alter gets it wrong.

  • Transcript redesign: Redesigned meeting transcript display with inline speaker editing, cleaner summaries, and header participants derived from the transcript text.

  • New speech models: Qwen3-ASR and Japanese Parakeet models.

  • Conversation sharing: Share discussions from the tab bar and command palette, with loading indicators in the list and tab name.

  • Arabic localization: Full Arabic UI support, including speech model and speaker review strings.

  • New models: GPT-6 Luna & Sol, Sonnet 5, Mimo 2.6 Flash & Pro, Qwen3.8 27B

Bug Fixes & Stability

  • SpeakerKit downloads: Deferred until diarization is actually needed, reducing startup overhead.

  • Buzz: Fixed live delivery across multiple channels.

  • Token limits: Raised default response token limit and capped local requests.