New dataset aids automatic classification of singing vocal modes

Researchers have developed a new dataset for automatically classifying vocal modes, a technique important for technology-assisted singing instruction. The dataset, comprising over 3,752 sustained vowel samples from four singers, aims to address a previous lack of data that hindered classification efforts. Baseline results using a ResNet18 model achieved a balanced accuracy of 81.3% in cross-validation, indicating the dataset's potential utility. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

IMPACT Provides a new dataset that could improve AI-powered singing education tools.

RANK_REASON This is a research paper introducing a new dataset for a specific classification task.

Read on arXiv cs.LG →

• arXiv
• Cathrin Sadolin
• Complete Vocal Technique
• ResNet18

• paper
• other