arXiv:2605.17443v2 Announce Type: replace Abstract: We analyze how automatic speech recognition (ASR) errors propagate through ASR-LLM cascades in Korean spoken question answering (SQA), focusing on downstream semantic failures that conventional ASR metrics cannot fully capture. Our analysis shows that the relative downstream degradation caused by ASR errors is consistent across LLMs with different absolute performance, suggesting that cascade degradation largely tracks ASR-stage information loss. We further identify single-character Korean ASR errors as a Korean-specific loss channel, where e
The proliferation of LLMs and the increasing demand for spoken interfaces drive research into understanding and mitigating error propagation in ASR-LLM cascades.
This study highlights a critical vulnerability in spoken AI systems, particularly in non-English languages, affecting the reliability and trustworthiness of AI applications.
The identified 'single-character Korean ASR errors' introduce a specific and significant challenge for developing robust multilingual AI, particularly for languages with complex orthographies.
- · ASR research institutions
- · Multilingual AI developers
- · Speech technology companies
- · LLM developers without robust ASR integration
- · Korean SQA service providers using current cascades
- · Early adopters of cascaded SQA systems
Improved Korean ASR models specifically designed to mitigate propagation errors will emerge.
Similar, language-specific error propagation challenges will be identified and addressed in other non-English languages.
The development of truly robust, multilingual, end-to-end spoken AI systems will accelerate, leading to more inclusive global AI applications.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.CL