Tag
This article discusses research on measuring benchmark optimization in speech recognition, where some ASR models may optimize for test benchmarks rather than real-world performance, and introduces tests to quantify this phenomenon.