schema/README.md

Annotation Schema

This directory contains the versioned annotation schema for the Voice Note Dataset.

Current Schema Version

Version: 1.0.0 File: annotation_schema_v1.json Last Updated: 2025-10-26

Schema Purpose

The annotation schema defines the structure and categories for annotating voice recordings with:

1. Classification - What type of content the note represents (31 categories)
2. Audio Defects - Quality issues and environmental noise (10 categories)
3. Content Issues - Content-level problems in the recording (5 categories)
4. Languages - Languages present in multi-language recordings (7 languages)
5. Transcription Quality - Assessment of transcription accuracy (5 levels)
6. Notes - Free-form text annotations

Using the Schema

In the Web UI

The Voice Note Dataset Manager web interface automatically loads and uses this schema to:

• Dynamically generate annotation checkboxes and options
• Validate annotation data
• Track schema version with each entry

Programmatically

import json

# Load the schema
with open('schema/annotation_schema_v1.json', 'r') as f:
    schema = json.load(f)

# Access schema information
print(f"Schema Version: {schema['version']}")
print(f"Description: {schema['description']}")

# Get classification categories
classifications = schema['schema']['classification']['categories']
for category in classifications:
    print(f"- {category['label']}: {category['description']}")

Via API

The Flask backend serves the schema via the /api/schema endpoint:

// Fetch the current schema
const response = await fetch('/api/schema');
const schema = await response.json();
console.log('Schema version:', schema.version);

Schema Evolution

When updating the schema:

1. Create a new version file: annotation_schema_v{N}.json
2. Update the API in app.py to point to the new version
3. Document changes in this README
4. Maintain backward compatibility when possible
5. Update the HF dataset copy in /home/daniel/repos/hugging-face/datasets/public/Voice-Note-Audio/

Version History

Version 1.0.0 (2025-10-26)

Initial schema release:

• 31 classification categories covering communication, development, content creation, and personal notes
• 10 audio defect categories for real-world STT evaluation
• 5 content issue types
• 7 language tags for multi-language support
• 5 transcription quality levels
• Free-form notes field
• Schema versioning metadata

Schema Distribution

This schema is distributed in three locations:

1. Development Repository: /home/daniel/repos/github/Voice-Note-Dataset/schema/
2. Hugging Face Dataset: /home/daniel/repos/hugging-face/datasets/public/Voice-Note-Audio/
3. Runtime: Served via Flask API at /api/schema

When updating, ensure all three locations are synchronized.

Contact

For questions or suggestions about the schema:

• Author: Daniel Rosehill
• Email: public@danielrosehill.com