Soundgen: An open-source tool for synthesizing nonverbal vocalizations.

Area of Science:

• Bioacoustics
• Speech Synthesis
• Psychology

Background:

• Investigating the communicative role of acoustic features in vocalizations is crucial.
• Existing text-to-speech systems are not always suitable for non-speech sound research.
• Bioacousticians and researchers of human nonverbal vocalizations need specialized tools.

Purpose of the Study:

• To introduce Soundgen, an open-source R package for synthesizing and manipulating natural-sounding nonverbal vocalizations.
• To validate Soundgen's effectiveness by comparing synthetic sounds with original recordings.
• To enable precise control over acoustic features for research in vocal communication and perception.

Main Methods:

• Developed Soundgen, an R package with command-line and interactive app interfaces.
• Synthesized nonverbal vocalizations (screams, moans, laughs) using high-level acoustic parameters.
• Validated synthetic sounds by comparing perceived emotion, valence, arousal, and authenticity against 60 original recordings.

Main Results:

• Synthetic vocalizations generated by Soundgen showed similar valence and arousal ratings compared to original recordings.
• Authenticity ratings for synthetic sounds were comparable to originals, especially for less complex vocalizations.
• Manual specification of a few high-level parameters enabled accurate sound reproduction.

Conclusions:

• Soundgen provides a validated tool for synthesizing and manipulating nonverbal vocalizations.
• The package facilitates research into the acoustic predictors of emotion and authenticity in human voices.
• Soundgen is broadly applicable to studies requiring precise control over nonspeech sound acoustics, including animal vocalizations and auditory perception research.