Fundamental Frequency Feature Normalization and Data Augmentation for\n Child Speech Recognition

Automatic speech recognition (ASR) systems for young children are needed due\nto the importance of age-appropriate educational technology. Because of the\nlack of publicly available young child speech data, feature extraction\nstrategies such as feature normalization and data augmentation must be\nconsidered to successfully train child ASR systems. This study proposes a novel\ntechnique for child ASR using both feature normalization and data augmentation\nmethods based on the relationship between formants and fundamental frequency\n($f_o$). Both the $f_o$ feature normalization and data augmentation techniques\nare implemented as a frequency shift in the Mel domain. These techniques are\nevaluated on a child read speech ASR task. Child ASR systems are trained by\nadapting a BLSTM-based acoustic model trained on adult speech. Using both $f_o$\nnormalization and data augmentation results in a relative word error rate (WER)\nimprovement of 19.3% over the baseline when tested on the OGI Kids' Speech\nCorpus, and the resulting child ASR system achieves the best WER currently\nreported on this corpus.\n