2 papers
cs.CL2022
Improving Generalization of Deep Neural Network Acoustic Models with Length Perturbation and N-best Based Label Smoothing
Xiaodong Cui, George Saon, Tohru Nagano +4
We introduce two techniques, length perturbation and n-best based label smoothing, to improve generalization of deep neural network (DNN) acoustic models for automatic speech recog…
cs.SD2021
Knowledge Distillation Leveraging Alternative Soft Targets from Non-Parallel Qualified Speech Data
Tohru Nagano, Takashi Fukuda, Gakuto Kurata
This paper describes a novel knowledge distillation framework that leverages acoustically qualified speech data included in an existing training data pool as privileged information…