Adding Robust Code-Switching Capabilities to High Performance Multilingual ASR

The paper introduces a method for enhancing multilingual automatic speech recognition (ASR) systems with robust code-switching (CSW) capabilities through Bayesian factorized adaptation. This approach improves transcription accuracy for code-switched words by 32.87% and overall word error rate (WER) by 5.31%, while preserving monolingual performance, indicating that effective CSW adaptation relies on knowledge integration rather than merely increasing data complexity. This advancement is significant for practitioners aiming to deploy ASR systems in multilingual environments where code-switching is prevalent.