A data augmentation approach for an automatic speech recognition system using laser Doppler vibrometer technology

In challenging conditions such as low signal-to-noise ratios and distant speech, microphone-based automatic speech recognition (ASR) struggles with clarity. To remedy this, laser Doppler vibrometer (LDV) technology is integrated into the ASR system and a data augmentation approach is employed to generate training data containing LDV attributes. The performance of the ASR, assessed using word error rates, showed superior results with the data augmentation approach compared to the baseline ASR system trained solely on real LDV data. Thus, with the aid of data augmentation, LDV can potentially be a sound-capturing device for ASR, offering valuable insights for future applications.