Introducing shared-hidden-layer autoencoders for transfer learning and their application in acoustic emotion recognition

Deng, Jun; Xia, Rui; Liu, Yang; Schuller, Björn

doi:10.1109/icassp.2014.6854517

Cited by 70 publications

(47 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The encoder function can be written in the form (3) where is a nonlinear activation function, typically a logistic sigmoid [6], [5], is the weight matrix, and is the bias vector.…”

Section: A Local Invariant Feature Learningmentioning

confidence: 99%

“…is the weight matrix shared with the encoder and is the bias vector. Auto-encoder trains the network by adjusting the parameters on the collected training set to minimize the total reconstruction error (5) where is the reconstruction error that is computed using squared error . In order to encourage units to maintain a low average activation, an additional penalty term is added into (5) (6) where is the sparse penalty term, is the average activation of hidden unit (averaged over the training set), denotes the number of active units, and denotes the number of training samples.…”

Section: A Local Invariant Feature Learningmentioning

confidence: 99%

See 1 more Smart Citation

Learning Salient Features for Speech Emotion <newline/>Recognition Using Convolutional <newline/>Neural Networks

Mao

Dong

Huang

et al. 2014

IEEE Trans. Multimedia

536

198

View full text Add to dashboard Cite

As an essential way of human emotional behavior understanding, speech emotion recognition (SER) has attracted a great deal of attention in human-centered signal processing. Accuracy in SER heavily depends on finding good affect-related, discriminative features. In this paper, we propose to learn affect-salient features for SER using convolutional neural networks (CNN). The training of CNN involves two stages. In the first stage, unlabeled samples are used to learn local invariant features (LIF) using a variant of sparse auto-encoder (SAE) with reconstruction penalization. In the second step, LIF is used as the input to a feature extractor, salient discriminative feature analysis (SDFA), to learn affect-salient, discriminative features using a novel objective function that encourages feature saliency, orthogonality, and discrimination for SER. Our experimental results on benchmark datasets show that our approach leads to stable and robust recognition performance in complex scenes (e.g., with speaker and language variation, and environment distortion) and outperforms several well-established SER features.Index Terms-Affective-salient discriminative feature analysis, convolutional neural networks, feature learning, speech emotion recognition.

show abstract

“…The encoder function can be written in the form (3) where is a nonlinear activation function, typically a logistic sigmoid [6], [5], is the weight matrix, and is the bias vector.…”

Section: A Local Invariant Feature Learningmentioning

confidence: 99%

Section: A Local Invariant Feature Learningmentioning

confidence: 99%

Learning Salient Features for Speech Emotion <newline/>Recognition Using Convolutional <newline/>Neural Networks

Mao

Dong

Huang

et al. 2014

IEEE Trans. Multimedia

536

198

View full text Add to dashboard Cite

show abstract

“….g., by DAE neural networks [ [12,11,13]]. In fact, emotional data resources often come in very different labellings such as (often different) categories or dimensions -transferring by suited approaches of machine learning can also be of help in this respect.…”

Section: Transfer Learningmentioning

confidence: 99%

Deep Learning Our Everyday Emotions

Schuller

2015

Advances in Neural Networks: Computational and Theoretical Issues

View full text Add to dashboard Cite

Abstract. Emotion is omnipresent in our daily lives and has a significant influence on our functional activities. Thus, computer-based recognising and monitoring of affective cues can be of interest such as when interacting with intelligent systems, or for health-care and security reasons. In this light, this short overview focuses on audio/visual and textual cues as input feature modality for automatic emotion recognition. In particular, it shows how these can best be modelled in a Neural Network context. This includes deep learning, and sparse auto-encoders for transfer learning of a compact task and population representation. It further shows avenues towards massively autonomous rich multitasklearning and required confidence estimation as is needed to prepare such technology for real-life application.

show abstract

“…Prior works on speech emotion recognition utilize various methods to deal with low-resource training samples, including unsupervised representation learning [6,7] and transfer learning [8,9]. Unsupervised representation learning takes full advantage of the information from unlabeled data.…”

Section: Introductionmentioning

confidence: 99%

Unsupervised Representation Learning with Future Observation Prediction for Speech Emotion Recognition

Lian¹,

Tao²,

Liu³

et al. 2019

Interspeech 2019

View full text Add to dashboard Cite

Prior works on speech emotion recognition utilize various unsupervised learning approaches to deal with low-resource samples. However, these methods pay less attention to modeling the long-term dynamic dependency, which is important for speech emotion recognition. To deal with this problem, this paper combines the unsupervised representation learning strategy -Future Observation Prediction (FOP), with transfer learning approaches (such as Fine-tuning and Hypercolumns). To verify the effectiveness of the proposed method, we conduct experiments on the IEMOCAP database. Experimental results demonstrate that our method is superior to currently advanced unsupervised learning strategies.

show abstract

Introducing shared-hidden-layer autoencoders for transfer learning and their application in acoustic emotion recognition

Cited by 70 publications

References 15 publications

Learning Salient Features for Speech Emotion <newline/>Recognition Using Convolutional <newline/>Neural Networks

Learning Salient Features for Speech Emotion <newline/>Recognition Using Convolutional <newline/>Neural Networks

Deep Learning Our Everyday Emotions

Unsupervised Representation Learning with Future Observation Prediction for Speech Emotion Recognition

Contact Info

Product

Resources

About