Can Voice Agents Handle Bilingual Customers? Benchmarking Frontier ASR on Code-Switched Speech

The study benchmarks the performance of Frontier ASR on code-switched speech, specifically assessing its ability to handle bilingual customers. The evaluation reveals that Frontier ASR achieves a word error rate (WER) of X% on code-switched datasets, significantly outperforming baseline models in multilingual contexts. This research is crucial for practitioners developing voice agents, as it highlights the importance of robust ASR systems in diverse linguistic environments, enhancing user experience and accessibility.