Transfer Learning for Low Resource Spoken Language Understanding without Speech-to-Text

Spoken Language Understanding (SLU) without speech-to-text conversion is more promising in low resource scenarios. These could be applications where there is not enough labeled data to train reliable speech recognition and language understanding systems, or where running SLU on edge is preferred over cloud based services. In this paper, we present an approach for building SLU without speech-to-text conversion in low resource scenarios using a transfer learning approach. We show that the intermediate layer representations from a pre-trained model outperforms the typically used Mel filter bank features. Moreover, the representations extracted from a model pre-trained on one language perform well even for SLU tasks on a different language.