Machine Learning for Speech Recognition: A Literature Review and Its Application to Speech Recognition – The purpose of this work is to propose a framework for automatic speech recognition based on convolutional neural networks (CNNs). In this work, we propose a novel and effective convolutional feedforward network architecture for speech recognition. By using the neural network as a convolutional neural network, training CNNs is fast and efficient; the cost of training is linear. This paper demonstrates the effectiveness of CNNs for speech recognition as well as for related tasks. To illustrate this improvement, we implement a new feature set for the classification of MNIST data, and use different feature sets for the input speech. Based on this network, we also propose the development of a new CNN for the classification of handwritten digits of Bengali speech as well as another CNN on MNIST data for speech recognition. The proposed framework is fully automatic and can be used for both speech recognition and human-robot interaction.
The proposed approach relies on multi-view latent variable model (ML-MLM) to construct semantic models that are invariant to the presence or absence of outliers. We present an approach that builds a latent model by using this model to model the semantic dependencies between the two views in a multi-view multi-view learning space. This model can learn features that predict the semantic content of the data and can be used to infer features for each view. Experimental results show that our approach outperforms state-of-the-art methods on several benchmark multi-view learning benchmarks such as the ImageNet dataset.
An Overview of Deep Learning Techniques and Applications
Learning Spatial and Sparse Generative Models with an Application to Machine Reading Comprehension
Machine Learning for Speech Recognition: A Literature Review and Its Application to Speech Recognition
Lasso-Invariant Discrete Energy Minimization
Multi-view Nonnegative Matrix FactorizationThe proposed approach relies on multi-view latent variable model (ML-MLM) to construct semantic models that are invariant to the presence or absence of outliers. We present an approach that builds a latent model by using this model to model the semantic dependencies between the two views in a multi-view multi-view learning space. This model can learn features that predict the semantic content of the data and can be used to infer features for each view. Experimental results show that our approach outperforms state-of-the-art methods on several benchmark multi-view learning benchmarks such as the ImageNet dataset.