Test Instance

This is the test instance of the TU Graz Repository.

Published December 30, 2024 | Version 1.2.1.
Dataset Open

Selection of Lyric Prompts (Precise & Imprecise) for Testing of Lyric Search Tools

  • 1. ROR icon Graz University of Technology
  • 2. ROR icon University of Sarajevo

Description

Selection of 51 hand-picked songs to use as prompts for any AI or ML based song lyric search engine. The lyrics are separated into two categories:

  • Direct song lyrics (lines taken from a lyric sheet found online)
  • Vague song lyrics (misheard/vibe based/fragmentary song lyrics, with balance between the three categories)

The lyrics are selected from 51 years of the Billboard Pop 100 list, taking one song from every year (every song was a conscious decision to ensure diversity of genres, rhythms and lyric complexity). Prompts ranging from Falco to 50 Cent, from Elton John to Drake. The vague song lyrics have been created using common homophones, usual word pairs, and some songs have been selected for their general atmosphere or idea because of the lyrics being only the second or third priority of the song ("Mozart song" for Rock Me Amadeus or "Napoleon song" for Waterloo.

Due to a lack of "misheard/misinterpreted lyrics" datasets existing on the internet, this curated list will give you a rough estimate of success with human-authored queries. Columns for ranking and reranking results are available and have been left blank. 

Files

Files (19.6 kB)