Problem
Research questions and friction points this paper is trying to address.
British accent robustness
Automatic Speech Recognition
financial voice assistants
ASR-LLM pipeline
accent gap
Mind the Accent Gap: British Accent Robustness in Speech-Driven Financial Voice Assistants
🤖 AI Summary
This study addresses the limitation that existing automatic speech recognition (ASR) models are predominantly biased toward American English, which compromises the robustness of financial voice assistants to regional British accents and consequently degrades downstream tool-calling accuracy. To investigate this issue, we introduce CavaBench, the first internally collected benchmark of British-accented financial voice queries, and design an end-to-end evaluation pipeline integrating ASR with large language model reasoning to systematically assess performance from speech recognition through task execution. Our findings reveal that while word error rate (WER) serves as a strong predictor of tool-calling accuracy, it fails to adequately capture task-level performance. This work provides critical empirical evidence for designing inclusive and reliable voice interaction systems in the financial domain.
📝 Abstract
AI voice assistants often use Automatic Speech Recognition (ASR) with LLM-based reasoning, yet existing systems struggle with regional British accents, including Scottish, Irish, and Welsh accents, since most ASR models are trained predominantly on American English voice data. Consequently, errors can carry through to the LLM stage, corrupting tool-call arguments and producing wrong or missing responses, which is especially costly in finance. Deployable ASR must also meet tight latency and memory budgets, making an accent-robust model choice even harder. We introduce CavaBench, the first internally collected benchmark of spoken financial queries, and use it to evaluate a range of ASR models and their end-to-end ASR-LLM pipeline behaviour across self-reported British accents. We find that WER strongly predicts downstream tool-calling accuracy ($r = -0.93$) but can fail to reflect task-level performance, with accent-related failures varying substantially across models and acoustic conditions. These findings guide the design of more inclusive, reliable voice-based financial assistants.
Innovation
Methods, ideas, or system contributions that make the work stand out.
CavaBench
Accent Robustness
ASR-LLM Pipeline
Financial Voice Assistants
Tool-Calling Accuracy
🔎 Similar Papers
No similar papers found.
