Elise: 3 Hours of Audio with 600+ Emotional Vocalization Tags

This 3-hour audio collection by MrDragonFox, updated in March 2025, provides speech recordings paired with classified emotional tags. It functions as a demo test set for voice fine-tuning, containing over 600 specific instances of non-verbal sounds such as laughs, sighs, and giggles.

Use Cases

• Fine-tuning text-to-speech models to trigger the 'laughs' or 'giggles' labels
• Training audio event detectors to recognize the 'sighs' or 'chuckles' features
• Evaluating emotional prosody in synthetic voices using the 'whispers' or 'gasps' tags

Strengths

• Contains 336 labeled laughs and 156 labeled sighs
• Released under the MIT license for permissive reuse
• Includes rare non-verbal categories such as 'clicks tongue' and 'gasps'

Limitations

• Small total volume of 3 hours limits its use for general-purpose speech models
• Extreme class imbalance, with labels like 'whispers' and 'long pause' having only 2 samples each
• Explicitly labeled as a demo test set rather than a production-scale resource

Provenance

Source

Jinsaryko/Elise

Collection Method

Annotated

Freshness

Last updated March 2025.