Vividh-ASR Benchmark Hindi Test Split: Real-World Speech Evaluation

Vividh-ASR is a complexity-stratified benchmark for Hindi speech recognition models. It is authored by adalat-ai and was last updated on May 20, 2026. The benchmark organizes evaluation by acoustic complexity to expose performance gaps on non-studio, real-world audio.

Use Cases

• Benchmark ASR model performance on real-world Hindi audio based on the described acoustic complexity stratification.
• Identify model weaknesses on non-studio speech based on the benchmark's focus on exposing studio-bias.
• Compare model robustness across different acoustic conditions based on the complexity-stratified evaluation framework.

Strengths

• Benchmark is specifically designed to evaluate real-world, non-studio Hindi speech.
• Evaluation is structured by acoustic complexity, providing detailed performance diagnostics.
• Authored by adalat-ai and hosted on Hugging Face, suggesting a machine learning research context.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Freshness should be verified as the last update timestamp is in the future (2026-05-20).

Provenance

Source

adalat-ai

Freshness

Last updated 2026-05-20 10:10:38