DNN-based speech enhancement with self-attention on feature dimension

Cheng, Jiaming; Liang, Ruiyu; Zhao, Li

doi:10.1007/s11042-020-09345-z

Cited by 11 publications

(6 citation statements)

References 35 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Multi-head attention is a set of multiple heads that jointly learn different representations at each position in the sequence. A DNN-based model [38] for speech enhancement with self-attention on feature dimension is proposed, which can make full use of the key information in frame-level feature.…”

Section: Attention Mechanismmentioning

confidence: 99%

An Attention-BiLSTM Model for Satellite Operation Prediction with Correlation Telemetry

Peng

Jia

Xie

et al. 2023

Preprint

View full text Add to dashboard Cite

At present, satellite anomaly is mostly solved after the event, and rarely predicted in advance in satellite health management. Thus, satellite trend prediction is quietly important for avoiding the fault which perhaps affects data accuracy and service quality of satellite, and even impacts greatly on satellite safety. However, it is difficult to predict satellite operation through a simple model because satellite system is complex, and telemetry data is numerous, coupled and spatiotemporal. Therefore, this paper proposes a model combing attention mechanism and Bidirectional Long Short-term Memory (Attention-BiLSTM) with correlation telemetry to predict the situation of satellite operation. Firstly, high-dimensional K-NearestNeighbor Mutual Information (HKNN-MI) method is performed to select the related telemetry variables from multiple variables of satellite telemetry data. Secondly, we put forward to a new BiLSTM model with attention mechanism for telemetry prediction. The dataset for the research is generated and transmitted from the power system of FY3E meteorological satellite. In order to verify the superiority of the proposed model, it is compared with other method based on the same dataset in the experiment. The result shows that the method outperforms other methods due to its better accuracy and prediction precision.

show abstract

Section: Attention Mechanismmentioning

confidence: 99%

An Attention-BiLSTM Model for Satellite Operation Prediction with Correlation Telemetry

Peng

Jia

Xie

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…A similar technique can be found in speech recognition where time series is involved. Cheng et al [53] proposed a DNN-based model for speech enhancement using attention mechanisms on the feature dimension. They suggested that the attention model can make full use of the key information in features and improve accuracy.…”

Section: B Rail Break Predictionmentioning

confidence: 99%

Rail Break Prediction and Cause Analysis Using Imbalanced In-Service Train Data

Zeng

Huang

Wang

et al. 2022

IEEE Trans. Instrum. Meas.

View full text Add to dashboard Cite

Timely detection and identification of rail breaks are crucial for safety and reliability of railway networks. This article proposes a new deep learning-based approach using the daily monitoring data from in-service trains. A time-series generative adversarial network (TimeGAN) is employed to mitigate the problem of data imbalance and preserve the temporal dynamics for generating synthetic rail breaks. A feature-level attentionbased bidirectional recurrent neural network (AM-BRNN) is proposed to enhance feature extraction and capture two-direction dependencies in sequential data for accurate prediction. The proposed approach is implemented on a three-year dataset collected from a section of railroads (up to 350 km) in Australia. A real-life validation is carried out to evaluate the prediction performance of the proposed model, where historical data are used to train the model and future "unseen" rail breaks along the whole track section are used for testing. The results show that the model can successfully predict nine out of 11 rail breaks three months ahead of time with a false prediction of nonbreak of 8.2%. Predicting rail breaks three months ahead of time will provide railroads enough time for maintenance planning. Given the prediction results, a Shapley additive explanations (SHAP) method is employed to perform a cause analysis for individual rail break. The results of cause analysis can assist railroads to plan appropriate maintenance to prevent rail breaks.

show abstract

“…This kind of approach has a low computational and hardware need and does not rely on priori speech information. As a result, its realtime performance is usually good [11]. The spectralsubtraction approach [2,12], Wiener filtering [2,13], minimum mean square error (MMSE) method [14] and subspace method [2] are examples of traditional speech enhancement algorithms.…”

Section: Introductionmentioning

confidence: 99%

Performance analysis of speech enhancement using spectral gating with U-Net

Agrawal,

Gupta,

Garg

2023

Journal of Electrical Engineering

View full text Add to dashboard Cite

Many speech processing systems’ crucial frontends include speech enhancement. Single-channel speech enhancement experiences a number of technological challenges. Due to the advent of cloud-based technology and the use of deep learning systems in big data, deep neural networks in particular have recently been seen as a potent means for complex classification and regression. In this work, spectral gating noise filter is combined with deep neural network U-Net to enhance the performance of speech enhancement network. Further, for performance analysis three distinct objective functions namely, Mean Square Error, Huber Loss and Mean Absolute Error are considered as loss functions. In addition, comparison of three different optimizers Adam, Adagrad and Stochastic Gradient Descent is presented. Proposed system is tested and evaluated on LibriSpeech and NOIZEUS datasets and compared to other state-of-the-art systems. It demonstrates that, in comparison to other state-of-the-art models, the proposed network outperformed them with PESQ scores of 2.737420 for training and 2.67857 for testing, along with better generalization ability.

show abstract

DNN-based speech enhancement with self-attention on feature dimension

Cited by 11 publications

References 35 publications

An Attention-BiLSTM Model for Satellite Operation Prediction with Correlation Telemetry

An Attention-BiLSTM Model for Satellite Operation Prediction with Correlation Telemetry

Rail Break Prediction and Cause Analysis Using Imbalanced In-Service Train Data

Performance analysis of speech enhancement using spectral gating with U-Net

Contact Info

Product

Resources

About