Benchmark
KuralHub Speech Emotion Recognition
Compare how accurately speech models recognize emotion across 29 languages.
Explore the original evidence
Explore the original evaluation, its methods, and reported findings.
Open evalCoverage
The paper compares 11 pretrained models on 33 datasets. The surrounding resource surveys 70 languages; only 29 are represented in the benchmark.
Methodology
Accuracy counts correctly classified clips. Macro-F1 gives each emotion class equal weight; weighted F1 gives common classes more influence. The project reports that the best single run and the paper’s best model family differ for seven languages. These summaries answer different questions. The HN announcement identifies project release 1.0; it does not establish a separate numbered benchmark edition.