> For the complete documentation index, see [llms.txt](https://docs.unitlab.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.unitlab.ai/solutions/multimodal-patterns-guides/document-and-audio-intelligence.md).

# Document and audio intelligence

Extract entities and relations from service text, and label speakers and transcripts in recorded conversations.

Support, compliance and field-service teams use Unitlab to turn tickets, reports and recordings into structured training data for extraction, search and speech models.

## Extract entities and relations from service tickets

**Data:** text tickets · **Tools:** Entity, Relation · **Ontology:** Robot, Component, Fault code, Symptom, Repair action, Part number

<figure><img src="https://2337177515-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2Fi6jXAxtoN6UXTLAJFexq%2Fuploads%2FxdNye2POfwEB4xGDNznP%2Fsol-text-ner-relations.webp?alt=media" alt="A robot service ticket with entities such as Robot, Component, Fault code, Symptom, Repair action and Part number highlighted, and relations linking them" width="800"><figcaption><p>A service ticket with each fact highlighted and linked: which robot, which part failed, and how it was fixed.</p></figcaption></figure>

1. Select a span of text and pick its class, for example **Fault code** for `E-217`.
2. Press `R` for relation mode and link entities, such as a **Symptom** to the **Component** it affects.
3. Review the item; the export keeps character offsets, classes and relations.

## Label speakers and transcripts

**Data:** recorded conversations · **Tools:** Audio intervals, transcript · **Ontology:** one class per speaker role, such as Operator, Robot assistant and Supervisor, each with a transcript

<figure><img src="https://2337177515-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2Fi6jXAxtoN6UXTLAJFexq%2Fuploads%2FJPtfWsrxolvJBLEzkjfR%2Fsol-audio-speakers-overview.webp?alt=media" alt="A waveform split into colored speaker turns, with a table listing each turn&#x27;s speaker, transcript, start and end time" width="800"><figcaption><p>Each speaker turn is an interval with its role and transcript. The table lists all turns with their times.</p></figcaption></figure>

1. Drag across the waveform to mark one turn and pick the speaker.
2. Type or correct the transcript for that turn.
3. Check the turn table: gaps and overlaps show up as missing or crossing times.

The result trains speaker diarization and speech-to-text, and the same setup works for translation and sound-event labels.
