Performance Prediction Of Automatic Speech Recognition Systems Using Convolutional Neural Networks

This paper focuses on the ASR performance prediction task. Two prediction approaches are compared: a state-of-the-art performance prediction based on engineered features and a new strategy based on learnt features using convolutional neural networks. We also try to better understand which information is captured by the deep model and its relation with different conditioning factors. To take advantage of this analysis, we then try to leverage these 3 types of information at training time through multi-task learning, which is slightly more efficient on ASR performance prediction task.