Synthetic Lyrics Dataset

Dataset Overview

Dataset Name

• Synthetic Lyrics Dataset with Mistral 7B

Data Collection Method

• Artist ID acquisition
• Song URL acquisition
• Lyrics web crawling

Data Annotation

• Using Mistral API combined with LangChain for annotation of theme, emotion, style, tone, and narrative
• Annotation example: obtain text style via a specific template, requesting a concise answer with only three words and no explanation

Dataset Scale and Cost

• Approximately 350 input tokens per lyric annotation
• Total about 14,700,000 tokens
• Total cost roughly $4

Dataset Applications

• Fine‑tune language models to support
  • Song classification
  • Lyrics generation
  • Recommendation systems