Transfer of Robust ASR Improvements to End-to-End Spoken Language Systems

Investigate whether improvements in speech recognition under realistic conversational and telecommunication conditions translate into more reliable end-to-end spoken language systems.

Background

The paper introduces TeleASR-Amharic, a benchmark for evaluating Amharic automatic speech recognition under realistic telecommunication degradations, including noise, channel distortion, bandwidth limitation, and packet loss. Although data scaling and targeted augmentation improve ASR robustness in the reported experiments, the paper does not determine whether these recognition-level gains carry over to complete spoken language systems that must perform speech-to-speech processing in conversational telecommunications settings. The authors identify the existing benchmark as a foundation for evaluating this unresolved transfer question.