Meetings & interviews

Speaker separation that stays on your machine

OpenASR Desktop separates speakers for every local file ASR model. It keeps native speaker tracks when available (currently MOSS) and uses a shared external diarization pipeline otherwise. Optional Voice ID lets recurring people appear by name. Everything runs on macOS or Windows without an account or a central cloud transcription backend.

Primary use
Multi-person meetings, interviews, and reviews that need who-spoke-when
Desktop separation
Every local file ASR model on macOS and Windows
Optional naming
Voice ID enrolls voiceprints and labels matching speakers by name
Automation
Open-core CLI --diarize and operator-only /v1/voice-id/* API

Why local diarization matters for meetings

Meeting and interview audio often cannot leave the room—client calls, internal reviews, research sessions. A local diarization path keeps both the words and the speaker structure on devices you control.

OpenASR separates speakers so transcripts are easier to skim, quote, and assign action items without pasting sensitive audio into a hosted STT product.

Desktop workflow

Desktop presents one Voice ID switch for every local file model. It preserves native tracks when available and otherwise runs FireRedVAD, segmentation-3.0, ReDimNet2-B6, and automatic clustering behind the same workflow.

Desktop is closed-source product software on top of an Apache-2.0 open core. There is no native Linux desktop app.

  • Transcribe multi-person audio locally on Mac or Windows
  • Separate speakers with any local file ASR model
  • Optionally name recurring speakers with Voice ID
  • No account, no telemetry, no central cloud STT backend

When the CLI is the better surface

Batch jobs, servers, and headless pipelines can use the open-core Rust CLI and local HTTP API. Anonymous diarization is available via --diarize; operator-only /v1/voice-id/* routes cover identity workflows outside the Desktop UI.

Models are delivered through a signed catalog. Normal transcription does not silently auto-download models, and CLI --offline fails closed when required assets are missing.

Honest scope

OpenASR does not publish benchmark league tables or claim to be the best diarization system in every condition. Voice ID is not authentication. Choose Desktop when you want a calm local app; choose CLI/API when you need automation and integration.

Frequently asked questions

What does OpenASR use for speaker separation on Desktop?

Desktop keeps native speaker tracks when a model provides them (currently MOSS) and otherwise uses a shared external FireRedVAD, segmentation-3.0, ReDimNet2-B6, and clustering pipeline. Voice ID works with every local file ASR model.

Can I run diarization without uploading audio?

Yes. Local-first is the default: Desktop and the open-core CLI/API process audio on your machine. There is no central cloud transcription backend and no account requirement.

Is there a Linux desktop diarization app?

No native Linux desktop app. On Linux, use the Apache-2.0 open-core CLI, local HTTP API, or published container images for CPU multi-arch and CUDA.

How do I name speakers instead of Speaker 1 / Speaker 2?

On Desktop, enroll a person or assign an anonymous speaker on the completed transcript; this works with every local file ASR model. For automation, use the operator-only /v1/voice-id/* API.

Does offline mode still diarize?

Diarization runs locally when the required local models and capabilities are present. CLI --offline fails closed if something needed is missing, rather than falling back to a network path.

Voice ID Enroll voiceprints so separated speakers can carry real names. Diarization docs How separation and enrollment work in the CLI and server. Offline speech to text Keep audio on the machine while you transcribe. Download Install the desktop app for macOS or Windows.