technique

Speaker diarisation

Speaker diarisation is the technique of segmenting an audio recording by who-spoke-when — answering "who said what" rather than just "what was said" — used heavily in meeting transcription, podcasts, and call analytics.

Diarisation segments audio into per-speaker turns: "Speaker 1: hello", "Speaker 2: hi". Production STT platforms (Deepgram, AssemblyAI, Google STT) ship diarisation as a configurable option. Quality varies — clean two-speaker calls are well-handled; messy multi-speaker meetings with overlapping speech remain hard. Pair with speaker identification (matching a diarised speaker to a known voice from a sample) for full speaker labels. Used in meeting summaries (Otter, Read.ai, Granola), call centre analytics, podcast transcription, and forensic audio analysis.

When to use speaker diarisation

Meeting transcription.
Podcast / interview transcription.
Call centre analytics.

Common mistakes

Expecting accurate diarisation on heavy overlapping speech — current models struggle.
Not pairing with speaker identification — "Speaker 3" labels are useless without names.

FAQ

What is speaker diarisation?

When should I use speaker diarisation?

Meeting transcription. Podcast / interview transcription. Call centre analytics.

What are the most common mistakes with speaker diarisation?

Expecting accurate diarisation on heavy overlapping speech — current models struggle. Not pairing with speaker identification — "Speaker 3" labels are useless without names.

Voice (LLM apps) — Voice in LLM apps refers to the full speech pipeline — speech-to-text (STT), language model, text-to-speech (TTS) — that lets users converse with an AI assistant in spoken language.
Streaming STT — Streaming STT (speech-to-text) emits partial transcriptions as the user speaks — instead of waiting for end-of-utterance — enabling sub-second response from a voice assistant.

Last updated: 2026-06-01. Raw markdown: https://promtable.com/glossary/diarisation.md.