LabelCraft

LabelCraft is an audio and text annotation platform. Annotators label, reviewers check, and managers accept the work.

🌐 labelcraft.org · X · LinkedIn · Telegram · Instagram

What you can label

• Speech and audio: transcription with speakers and timestamps, speaker diarization, voice activity, sound events, and a category for each segment.
• Text and NLP: named entities, text classification, translation and summaries.
• Conversations and LLM data: dialogue labels, written and rated responses, preference pairs and rankings. These are in development.

How work moves through a project

1. Upload audio files, or text as .txt, .json or .jsonl.
2. Annotate in the studio. AI can prepare a first draft; the annotator corrects it.
3. Review: a reviewer approves the work or corrects it, and each review gets a quality score.
4. Accept: a manager accepts the item. Accepted items make up your dataset.

Every save is a new version, so any two versions can be compared segment by segment.

Your data

Each organization's files are kept in separate storage. You can export the full dataset yourself at any time, free of charge, or read accepted data through the API and the MCP server with access keys you control.

Most of the data on the platform so far is Uzbek and multilingual speech.

▶ Watch the intro