Diarization Demo

Compare diarization backends from the same recording surface.

Use Google as a cloud baseline or route audio into a local worker for WhisperX, pyannote, Parakeet, and NeMo experiments without changing the frontend flow.

Input

Microphone / WebM Opus

Provider

WhisperX + pyannote

Mode

Two-speaker diarization

Session Control

Idle and ready

Output

Word-level speaker tags with a stable comparison surface.

Constraint

Short clips only. Large uploads are rejected.

Transcript Surface

Speaker-separated output

Speaker 1Speaker 2

Awaiting Input

The transcript pane stays empty until a recording is captured and processed. Use the same clip across providers to compare diarization quality and timing behavior.

Sequence

  1. 01 / Select a provider and model.
  2. 02 / Grant microphone access.
  3. 03 / Capture a short exchange.
  4. 04 / Compare transcript output.

Evaluation

Keep the recording surface stable and swap only one variable at a time: provider, ASR model, or diarization stack.

Constraints

The local worker is the right place for heavy Python models. The Next route stays thin and synchronous by design.