Data Schemas

Welcome to the data schemas tour with the soundevent package! In this overview, we'll break down the various data schemas provided by the package into the following sections:

Describing the Data

soundevent provides tools to attach essential information to various objects in bioacoustic analysis:

• Users: Keeping reference of everyone's contribution.
• Terms: Standardized vocabularies ensure consistent language.
• Tags: Attaching semantic context to objects.
• Features: Numerical descriptors capturing continuously varying attributes.
• Notes: User-written free-text annotations.

Audio Content

At the core of acoustic analysis, we have schemas for:

• Recordings: Complete audio files.
• Dataset: A collection of recordings from a common source.

Acoustic Objects

Identifying distinctive sound elements within audio content:

• Geometric Objects: Defining Regions of Interest (RoI) in the temporal-frequency plane.
• Sound Events: Individual sonic occurrences.
• Sequences: Patterns of connected sound events.
• Clips: Fragments extracted from recordings.

Annotation

soundevent places emphasis on human annotation processes, covering:

• Sound Event Annotations: Expert-created markers for relevant sound events.
• Sequence Annotations: User provided annotations of sequences of sound events.
• Clip Annotations: Annotations and notes at the clip level.
• Annotation Task: Descriptions of tasks and the status of annotation.
• Annotation Project: The collective description of tasks and annotations.

Prediction

Automated processing methods also play a role, generating:

• Sound Event Predictions: Predictions made during automated processing.
• Sequence Predictions: Predictions of sequences of sound events.
• Clip Predictions: Collections of predictions and additional information at the clip level.
• Model Runs: Sets of clip predictions generated in a single run by a specific model.

Evaluation

Assessing the accuracy of predictions is crucial, and soundevent provides schemas for:

• Matches: Predicted sound events overlapping with ground truth.
• Clip Evaluation: Information about matches and performance metrics at the clip level.
• Evaluation: Comprehensive details on model performance across the entire evaluation set.
• Evaluation Set: Human annotations serving as ground truth.

Want to know more? Dive in for a closer look at each of these schemas.

Unique Identifiers

In soundevent, various objects feature a field called uuid. This field stores a Universal Unique Identifier (UUID), a 128-bit label generated automatically. When created following standard methods, UUIDs are practically unique. Information labeled with UUIDs by different parties can be combined into a unified database without fear of duplication.