T Textura

Glossary term

What is diarization?

Diarization is the automatic splitting of a conversation recording into turns by speaker: who spoke and when. It answers the question "who spoke," whereas speech recognition answers "what was said."

In a phone call or meeting, voices overlap and alternate. Diarization analyzes the acoustic characteristics of speech, separates the individual speakers and labels the timeline: this segment is the agent, this one is the customer, and here they talk at the same time (overtalk).

The result is a transcript broken out by role, where you can see how much each participant spoke, who interrupted whom, where the pauses were and the longest monologue. Without diarization a transcript is a solid wall of text from which per-speaker metrics simply can't be computed.

Technically, diarization relies on **voice prints** (voice embeddings): speech is cut into short segments, a print vector is computed for each, and similar prints are then clustered into speakers. Accuracy depends on audio quality and the number of channels — multichannel recordings (each participant on their own channel) diarize more reliably than mono.

In the platform, diarization runs automatically on upload, and the separated voices can be matched against a speaker registry to track specific agents and customers across the whole archive.

How it works in the platform

The platform diarizes recordings automatically (GPU diarization on the pyannote engine) the moment they're uploaded. The speaker-separated turns feed every conversation-dynamics metric — talk share, interruptions, longest monologue — and the voice prints can be linked to a speaker registry to find specific people across the archive.

Related terms

Call transcription Call transcription is the automatic conversion of a conversation's audio recording into text with timing attached to each turn. It answers the question "what was said" and serves as the base layer on top of which any conversation analysis runs. Speech analytics Speech analytics is the automatic analysis of conversations (calls and written exchanges) to extract metrics, topics, tone and other data from what was said and how the dialogue unfolded. It turns recordings into structured data you can build dashboards on and make decisions from. STT (speech recognition) STT (speech-to-text) is the technology of automatically converting spoken speech into text. It's the base component that makes it possible to read, search and analyze conversations, as well as to understand the caller in voice agents.

Ready to start?

Turn every conversation into data, knowledge and action

Start by analyzing your conversations — no risk, no bots required. The platform turns your archive into data and a knowledge base, and voice agents plug in when you're ready.

Start for free All plans and per-operation prices

Free plan, no card required.