ARK-ASR-3B: Open ASR Leaderboard Results for English Short-Form Speech

Raw JSONL manifests contain the results for the AutoArk-AI/ARK-ASR-3B model on the public English short-form splits from the hf-audio/open-asr-leaderboard. The results were generated and scored on June 22, 2026, using the shared Open ASR Leaderboard scorer on a local machine with 8x RTX 4090 GPUs. The dataset was authored by AutoArk-AI and hosted on Hugging Face.

Use Cases

• Benchmark ASR model performance based on the official Open ASR Leaderboard splits mentioned in the description.
• Compare error rates and scores for the ARK-ASR-3B model against other models on the same test set.
• Analyze the performance of a large-scale ASR model on short-form English audio tasks.

Strengths

• Results were generated on June 22, 2026, providing a specific timestamp for the evaluation.
• Scoring was performed using the official, shared Open ASR Leaderboard scorer, suggesting standardized metrics.
• The underlying test data is from a public, established benchmark (hf-audio/open-asr-leaderboard).

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The dataset contains only results for a single model (ARK-ASR-3B), not the benchmark data itself.

Provenance

Source

Hugging Face (AutoArk-AI/ark-asr-3b-open-asr-leaderboard-results)

Collection Method

Results generated by scoring the ARK-ASR-3B model on the Open ASR Leaderboard splits.

Freshness

Last updated 2026-06-22 10:01:19