Impact of Age in ASR for the Elderly: Preliminary Experiments in European Portuguese

摘要: Standard automatic speech recognition (ASR) systems use acoustic models typically trained with of young adult speakers. Ageing is known to alter production in ways that require ASR be adapted, particular at the level modeling. This paper reports experiments illustrate impact speaker age on performance. A large read corpus European Portuguese allowed us measure statistically significant performance differences among groups ranging from 60- 90-year-old An increase 41% relative (11.9% absolute) word error rate was observed between 60-65-year-old and 81-86-year-old also retraining (AMs), further illustrating ageing Differentiated gains were depending range adaptation data retrain models.