> 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/documentation/annotations/audio-annotation.md).

# Audio Annotation

Audio annotation is a temporal boundary problem. Reviewers must be able to understand exactly where an event, speaker, or acoustic state begins and ends.

### What this editor is designed to preserve

* Original waveform and playback context.
* Time-accurate segments, classifications, and structured properties.
* Overlap, silence, noise, speaker change, and uncertain-boundary decisions.

![Audio annotation Workbench with waveform](https://292810646-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FGjVLUz4wthGkGlRKM6rM%2Fuploads%2Fl6MohTpE3RhZHu2KCC3y%2Faudio-workbench.png?alt=media\&token=7cce84cb-db73-40ef-9a56-2e18fa79f352)

*The waveform, playback controls, segment regions, ontology, and workflow action should be reviewed as one time-based task.*

### Read the Workbench before annotating

The audio workspace provides:

* waveform and timeline;
* temporal event segments;
* play/pause;
* rewind and forward by 10 seconds;
* current time and duration;
* volume and mute;
* zoom;
* playback speed;
* event/class selection;
* comments;
* item properties;
* tags;
* workflow completion actions.

The audio workbench combines waveform and spectrogram context with temporal event regions. Audio can also be aligned with document text or video inside the same grouped task.

### Common audio tasks

* **Clip classification:** describe the whole recording.
* **Temporal event detection:** mark the start and end of a sound, speaker, motion, noise, or other event.
* **Segmentation:** partition a recording into meaningful regions.
* **Transcription-oriented work:** produce or review speech text.

An external audio model can return an Event result and may optionally provide a speech-recognition transcript.

### Boundary quality

Audio disagreement often comes from boundary policy rather than class identity. Teams should define whether an event begins at the first acoustic evidence, the first intelligible phoneme, or a fixed context margin; how overlap is handled; whether silence is labeled; and how background noise interacts with foreground speech.

### From assigned task to review

{% stepper %}
{% step %}

#### 1. Listen for context

Play a wider region before setting a boundary so the event is interpreted in context.
{% endstep %}

{% step %}

#### 2. Create the segment

Set the start and end time using waveform zoom and playback.
{% endstep %}

{% step %}

#### 3. Apply the class

Choose the speaker, event, state, or classification defined by the ontology.
{% endstep %}

{% step %}

#### 4. Handle overlap and silence

Follow the project policy for simultaneous speakers, background noise, pauses, and uncertain onset.
{% endstep %}

{% step %}

#### 5. Review at multiple scales

Zoom in for exact boundaries and zoom out for missed events or long-range consistency.
{% endstep %}

{% step %}

#### 6. Save and route

Complete required values and use the workflow action appropriate to the current stage.
{% endstep %}
{% endstepper %}

### Modality-specific quality controls

| Decision           | Production guidance                                                                       |
| ------------------ | ----------------------------------------------------------------------------------------- |
| Boundary tolerance | Define acceptable start/end precision for the downstream task.                            |
| Overlap            | Specify whether segments can overlap and how each source is labeled.                      |
| Silence and noise  | Separate absence of speech from annotated acoustic events according to policy.            |
| Playback           | Review at normal speed and use slower playback only where it improves a defined decision. |

### Route the result

Review rejected audio with the exact time interval and reason. Use comments for contextual guidance and Issues for recurring or owned correction.