WildASR: Multilingual Diagnostic Benchmark for ASR Robustness

WildASR is a multilingual diagnostic benchmark built from real human speech to stress-test automatic speech recognition (ASR) robustness under real-world out-of-distribution conditions. The dataset decomposes robustness into axes including environmental degradation and demographic shift. It was created by bosonai and last updated on 2026-03 -25.

Use Cases

• Benchmarking ASR model robustness based on environmental degradation factors like reverberation and noise
• Evaluating demographic fairness in speech recognition based on speaker characteristics like age
• Testing ASR performance on out-of-distribution conditions based on real human speech data
• Diagnosing specific failure modes in ASR systems based on the decomposed robustness axes

Strengths

• Designed specifically for stress-testing ASR robustness under real-world out-of-distribution conditions
• Decomposes robustness into specific, interpretable axes like environmental degradation and demographic shift
• Built from real human speech, likely providing authentic test scenarios

Limitations

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

Provenance

Source

bosonai

Collection Method

Built from real human speech, as stated in the description.

Time Range

null

Freshness

Last updated 2026-03-25 04:06:43; freshness should be verified

Geography

null