Turkish ASR Benchmark: 20 Models Evaluated on 1,060 Speech Clips

Turkish speech recognition benchmark comparing 20 automatic speech recognition models on 1,060 audio clips totaling about 105 minutes. The benchmark was produced by Muhammed Kumcu and Yagmur Tuncer and draws from three established Turkish speech sources. It contains performance metrics but does not distribute the source audio, reference transcripts, or generated hypotheses.

Use Cases

• Benchmarking ASR model performance based on comparative metrics
• Selecting a Turkish ASR model based on evaluation results
• Tracking progress in Turkish speech recognition technology

Strengths

• Compares 20 distinct ASR models
• Evaluates performance on 1,060 audio clips
• Draws from three established Turkish speech sources

Limitations

• Contains metrics only; source audio and transcripts are not distributed
• Column-level documentation is absent; field semantics must be inferred after download
• Row count is unknown, which may limit suitability assessment

Provenance

Source

huggingface

Collection Method

Benchmark metrics produced from three established Turkish speech sources.

Freshness

Last updated 2026-06-10 16:21:37