In Automatic speech recognition paradigm has been shifted from statistical model (GMM-HMM) to deep neural network. Among various types of Deep neural networks architecture, CNNs have been most broadly used and considered. There are many advanced features in CNNs like weight sharing, local filters and pooling etc. An ideal pooling method is to be expected to extract only useful information and discards irrelevant details. Pooling is an important component of CNN to be highlighted. Hence, to understand and select the best pooling technique, this paper aims to provide a broad survey on different types of pooling techniques applied in CNNs architecture. This paper will also summarize the improvements of CNNs, including key properties of CNNs, architecture of CNNs.
Convolutional Neural Network for ASR
2018-03-01
5747022 byte
Conference paper
Electronic Resource
English
European Patent Office | 2024
|Pedestrian Detection with Deep Convolutional Neural Network
British Library Conference Proceedings | 2015
|Phase unwrapping with the convolutional neural network
British Library Conference Proceedings | 2020
|