CoVox: A dataset of contrasting vocalizations

Area of Science:

• Speech and Hearing Sciences
• Acoustic Phonetics
• Music Psychology

Background:

• The human voice exhibits significant variability across singing and speaking tasks.
• Existing vocalization datasets often lack ecological validity and stylistic diversity.
• Developing controlled, yet naturalistic, vocal stimuli is crucial for voice research.

Purpose of the Study:

• To introduce a novel, ecologically valid dataset of vocalizations encompassing diverse singing and speaking styles.
• To provide a resource for studying the acoustic and perceptual differences between vocal styles.
• To facilitate research on the human singing and speaking voice.

Main Methods:

• 22 female singers recorded melody excerpts in three singing styles (lullaby, pop, opera) and two speaking styles (adult, infant).
• Stimuli were produced using original lyrics in Brazilian Portuguese and a/lu/ sound.
• A forced-choice experiment validated the dataset, with listeners achieving >69% accuracy in style recognition.

Main Results:

• A validated dataset of 1320 vocalizations was created, demonstrating high accuracy in style recognition.
• Acoustic analyses revealed distinct and contrasting acoustic profiles for each vocalization style.
• The dataset provides a valuable resource for comparative voice research.

Conclusions:

• The developed vocalization dataset is ecologically valid and perceptually distinct across styles.
• This resource supports further investigation into the acoustic characteristics and perception of singing and speaking voices.
• Freely available recordings promote wider accessibility and advancement in voice science.