Curriculum learning based approach for noise robust language identification using DNN with attention

Vuddagiri, Ravi Kumar; Vydana, Hari Krishna; Vuppala, Anil Kumar

doi:10.1016/j.eswa.2018.06.004

Cited by 28 publications

(13 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…As seen in Table II, in terms of the C avg , the baseline neural models like LSTM [36] and DNN with attention [27] perform comparatively well on short durations (3 sec. and 10 sec.).…”

Section: A Lre Evaluation Resultsmentioning

confidence: 99%

“…The end-to-end approaches to language recognition have been explored with long short term memory (LSTM) networks and with DNNs [26]. A recent approach using curriculum learning had also been applied for noise robust language recognition [27]. However, the state-ofthe-art language recognition systems using large scale NIST language recognition evaluation (LRE) challenges, continue to use the i-vector/x-vector based approaches with support vector machine classifier [28].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Towards Relevance and Sequence Modeling in Language Recognition

Padi¹,

Mohan²,

Ganapathy

2020

IEEE/ACM Trans. Audio Speech Lang. Process.

View full text Add to dashboard Cite

The task of automatic language identification (LID) involving multiple dialects of the same language family in the presence of noise is a challenging problem. In these scenarios, the identity of the language/dialect may be reliably present only in parts of the temporal sequence of the speech signal. The conventional approaches to LID (and for speaker recognition) ignore the sequence information by extracting long-term statistical summary of the recording assuming an independence of the feature frames. In this paper, we propose a neural network framework utilizing short-sequence information in language recognition. In particular, a new model is proposed for incorporating relevance in language recognition, where parts of speech data are weighted more based on their relevance for the language recognition task. This relevance weighting is achieved using the bidirectional long short-term memory (BLSTM) network with attention modeling. We explore two approaches, the first approach uses segment level i-vector/x-vector representations that are aggregated in the neural model and the second approach where the acoustic features are directly modeled in an end-to-end neural model. Experiments are performed using the language recognition task in NIST LRE 2017 Challenge using clean, noisy and multi-speaker speech data as well as in the RATS language recognition corpus. In these experiments on noisy LRE tasks as well as the RATS dataset, the proposed approach yields significant improvements over the conventional i-vector/x-vector based language recognition approaches as well as with other previous models incorporating sequence information.

show abstract

“…As seen in Table II, in terms of the C avg , the baseline neural models like LSTM [36] and DNN with attention [27] perform comparatively well on short durations (3 sec. and 10 sec.).…”

Section: A Lre Evaluation Resultsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Towards Relevance and Sequence Modeling in Language Recognition

Padi¹,

Mohan²,

Ganapathy

2020

IEEE/ACM Trans. Audio Speech Lang. Process.

View full text Add to dashboard Cite

show abstract

“…Due to the deep mining characteristics of data features of CNN network, it has made great progress in the identification and analysis of targets in the field of statistical analysis and audio signal processing [14], [15], and it also provides the possibility of deep mining and utilization of seismic information. Vuddagiri et al [16] proposed curriculum learning based approach to noise robust language identification using DNN with attention, Lee et al [17] proposed a regression approach to single-channel speech separation via high-resolution deep neural networks, they performed accurate speech recognition with background noise in the environment. An automatic microseismic or acoustic emission arrival identification scheme with deep recurrent neural networks proposed by Zheng et al [18].…”

Section: Introductionmentioning

confidence: 99%

Research on Microseismic Source Locations Based on Deep Reinforcement Learning

Wei

Pan

et al. 2019

IEEE Access

View full text Add to dashboard Cite

The technology used to locate the sources of shallow underground seismic activity is mainly used to assess underground damage, the positioning of bombs in ordnance test fields, and the positioning of charges in engineering blasting. Due to sensor acquisition clock errors and the lack of generalizability of vibration data, the current sourced location methods cannot easily meet the requirements of precise positioning and system reaction time in assessing near-field microseismic activity. Based on the seismic wave analysis technique and the deep reinforcement learning method, this paper proposes a dynamic model for locating shallow underground seismic sources. The model allows the generalization of low-dimensional vibration waves information to high dimensions through spatial scanning. The process of source detection is treated as a Markov process. The correspondence between the center of a source and a high-dimensional energy distribution is established, and training is then gradually performed using the deep reinforcement learning method. The optimization of the center of the source to accurately determine the position of the focal center is provisionally called the reinforcement learning source scanning algorithm (RL-SSA). In addition, a small site-based static explosion test shows that the positioning method can greatly improve the positioning accuracy (<1 m) in the location of near-field microseismic sources.INDEX TERMS Convolutional neural network (CNN), deep reinforcement learning (DRL), deep Q-learning (DQN), source-scanning algorithm (SSA).

show abstract

“…neural network (CNN), recurrent neural network (RNN), deep autoencoder (DAE), long short-term memory network, and deep belief network (DBN) are considered deep learning algorithms. Initially, deep learning was applied to the fields of image identification [13][14][15][16] and natural language processing [17][18][19]. Today, it is also gradually applied on process monitoring and modeling prediction [20][21][22][23].…”

Section: Introductionmentioning

confidence: 99%

Soft Sensor Modeling Method by Maximizing Output-Related Variable Characteristics Based on a Stacked Autoencoder and Maximal Information Coefficients

Wang¹,

Yan²

2019

IJCIS

View full text Add to dashboard Cite

KeywordsSoft sensor Deep learning Stacked autoencoder (SAE) Maximal information coefficient (MIC) Modeling method A B S T R A C TThe key factors required to establish a precise soft sensor model for industrial processes include selection of variables affecting vital indicators from a large number of online measurement variables and elimination of the effects of unrelated disturbance variables. How to compress redundant information and retain the unique characteristic information contained by the selected variables is worthy of in-depth research. A novel soft sensor modeling method based on weighted maximal information coefficients (MICs) and a stacked autoencoder (SAE), hereinafter referred to as MICW-SAE, is proposed in this work. In our model, the MICs between each input and output variable are calculated and compared with the threshold before training each network in SAE. Then, input variables with low MICs are selected, and the average MIC index is calculated using other input variables. If the index is higher than the second threshold, the MIC of this specific variable is set to 0. Finally, the weights of all input variables are determined in accordance with the scale and placed into the loss function for training. The Boston house-price and naphtha dry point temperature datasets are used to prove the prediction ability of our model. Results demonstrate that MICW-SAE can enhance the output-related features of the input variables. Moreover, redundant information that can also be represented by other input variables are identified and excluded.

show abstract

Curriculum learning based approach for noise robust language identification using DNN with attention

Cited by 28 publications

References 13 publications

Towards Relevance and Sequence Modeling in Language Recognition

Towards Relevance and Sequence Modeling in Language Recognition

Research on Microseismic Source Locations Based on Deep Reinforcement Learning

Soft Sensor Modeling Method by Maximizing Output-Related Variable Characteristics Based on a Stacked Autoencoder and Maximal Information Coefficients

Contact Info

Product

Resources

About