Libri-AudioEvent: 11,700 Synthesized Noisy Speech Clips

Libri-AudioEvent is a synthesized noisy-speech dataset containing matched noisy speech, clean speech, and noise signals. The dataset contains 11,700 total clips across training, validation, and test splits, with each clip being 10 seconds long and sampled at 16 kHz. It was created by Ediethia and last updated on Hugging Face in July 2026.

Use Cases

• Train speech denoising models based on the matched noisy and clean speech pairs.
• Evaluate audio source separation algorithms based on the isolated noise and clean speech signals.
• Benchmark speech enhancement systems using the dedicated test-mix and test-iso splits.
• Synthesize custom noisy audio samples for data augmentation based on the provided clean speech and noise components.

Strengths

• Dataset contains 11,700 total audio clips, providing a substantial scale for model training and evaluation.
• All clips are standardized at 10 seconds duration and 16 kHz sampling rate, ensuring consistency.
• Data is structured into four distinct splits (training, validation, test_mix, test_iso) for rigorous machine learning workflows.
• Provides matched triplets of noisy speech, clean speech, and noise signals, which is valuable for supervised learning tasks.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• The dataset is synthesized, which may not fully capture the acoustic characteristics of real-world noisy environments.
• Row count per split is provided, but total file size and specific audio formats are unknown.

Provenance

Source

Hugging Face dataset repository by Ediethia.

Collection Method

Synthesized; method of noise synthesis and source of clean speech are not detailed in the provided description.

Freshness

Last updated 2026-07-11 09:16:01; freshness should be verified.