RepeatAudio: Synthetic and Real-World Audio Samples for Repetition Counting

Hamozwa created RepeatAudio for research in class-agnostic audio repetition counting. The dataset contains synthetic samples with varying noise and real-world samples from mechanical, ecological, and medical domains. It was last updated on May 29, 2026.

Use Cases

• Training models for counting repetition events in audio based on the described synthetic samples.
• Benchmarking audio repetition algorithms on real-world samples from domains like clocks and heartbeats.
• Developing noise-robust audio processing models based on the described synthetic datasets with varying noise levels.

Strengths

• Contains two distinct sections: synthetic audio with controlled noise and real-world audio from multiple domains.
• Synthetic samples have a uniform 10-second length and contain 0 to 8 repetition events.
• Real-world samples are derived from mechanical, ecological, and medical sources, providing domain variety.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

Hamozwa on Hugging Face.

Collection Method

Likely a mix of synthetic generation and collection/derivation from real-world sources.

Freshness

Last updated 2026-05-29 19:38:06; freshness should be verified.