Speech Utterances: A Collection of Human Non-Speech Vocal Sounds

A collection of human non-speech vocal sound datasets in WebDataset format, useful for audio classification tasks. The collection includes the 'NonSpeech7k' subset with 7,014 samples across 7 classes like breathing and laughing, sourced from Zenodo. The dataset was authored by 'gijs' and last updated on Hugging Face in January 2026.

Use Cases

• Train audio classifiers for non-verbal vocal expressions based on the described classes like breathing and laughing.
• Develop emotion recognition models from non-speech vocal sounds mentioned in the description.
• Benchmark machine learning models for sound event detection using the provided 7-class categorization.

Strengths

• The 'NonSpeech7k' subset contains 7,014 labeled audio samples, providing a foundation for model training.
• Data is organized into 7 distinct classes of non-speech vocal sounds, such as coughing and crying.
• The dataset is provided in WebDataset format, which is a standardized format for large-scale audio data.

Limitations

• Description metadata is limited; actual data quality, sample characteristics, and full collection details require manual inspection after download.
• Column-level documentation is absent; field semantics and audio file attributes must be inferred after download.
• The total size and row count for the full 'all' collection are unknown, which may limit suitability assessment.

Provenance

Source

Hugging Face (user: gijs), with the 'NonSpeech7k' subset sourced from Zenodo.

Collection Method

Collection of existing datasets; method of original data gathering is not specified.

Freshness

Last updated 2026-01-19 16:27:52; freshness should be verified.