*This dataset supports online use.Click here to jump.

NonverbalTTS is a non-verbal audio generation dataset released by VK Lab and Yandex in 2025. The related paper results are "NonverbalTTS: A Public English Corpus of Text-Aligned Nonverbal Vocalizations with Emotion Annotations for Text-to-Speech", which aims to promote expressive text-to-audio (TTS) research and support models to generate natural speech that contains emotions and non-verbal sounds.

The dataset contains 17 hours of high-quality speech data from 2,296 participants (60% males, 40% females), covering 10 non-verbal speech types (breathing, laughing, sighing, sneezing, coughing, throat clearing, groaning, grunting, snoring, and inhaling) and 8 emotion categories (anger, disgust, fear, happiness, neutral, sadness, surprise, and other).

Dataset features:

• Multi-source data: derived from VoxCeleb and Expresso corpora

• Rich metadata: emotion tags, non-verbal speech annotations, speaker IDs, audio quality metrics

• Sampling rate: 16kHz for audio from VoxCeleb, 48kHz for audio from Expresso