Anchor voiceprint recognition in live streaming via RawNet-SA and gated recurrent unit

With the sharp booming of online live streaming platforms, some anchors seek profits and accumulate popularity by mixing inappropriate content into live programs. After being blacklisted, these anchors even forged their identities to change the platform to continue live, causing great harm to the network environment. Therefore, we propose an anchor voiceprint recognition in live streaming via RawNet-SA and gated recurrent unit (GRU) for anchor identification of live platform. First, the speech of the anchor is extracted from the live streaming by using voice activation detection (VAD) and speech separation. Then, the feature sequence of anchor voiceprint is generated from the speech waveform with the self-attention network RawNet-SA. Finally, the feature sequence of anchor voiceprint is aggregated by GRU to transform into a deep voiceprint feature vector for anchor recognition. Experiments are conducted on the VoxCeleb, CN-Celeb, and MUSAN dataset, and the competitive results demonstrate that our method can effectively recognize the anchor voiceprint in video streaming.

Named Entity Recognition Model of Chinese Clinical Electronic Medical Record Based on XLNet-BiLSTM

The recognition of named entities in Chinese clinical electronic medical records is one of the basic tasks to realize smart medical care. Aiming at the insufficient text semantic representation of the traditional word vector model and the inability of the recurrent neural network (RNN) model to solve the problems of long-term dependence, a Chinese clinical electronic medical record named entity recognition model XLNet-BiLSTM-MHA-CRF based on XLNet is proposed. Use the XLNet pre-training language model as the embedding layer to vectorize the medical record text to solve the problem of ambiguity; use the bidirectional long and short-term memory network (BiLSTM) gate control unit to obtain the forward and backward semantic feature information of the sentence; Then input the feature sequence to the multi-head attention layer (multi-head attention, MHA), use MHA to obtain information represented by different subspaces of the feature sequence, enhance the relevance of context semantics and eliminate noise; finally, input the conditional random field CRF to identify the global maximum 优 sequence. The experimental results show that the XLNet-BiLSTM-Attention-CRF model has achieved good results on the CCKS-2017 named entity recognition data set.

Video Summarization Based on Multimodal Features

In this manuscript, the authors present a keyshots-based supervised video summarization method, where feature fusion and LSTM networks are used for summarization. The framework can be divided into three folds: 1) The authors formulate video summarization as a sequence to sequence problem, which should predict the importance score of video content based on video feature sequence. 2) By simultaneously considering visual features and textual features, the authors present the deep fusion multimodal features and summarize videos based on recurrent encoder-decoder architecture with bi-directional LSTM. 3) Most importantly, in order to train the supervised video summarization framework, the authors adopt the number of users who decided to select current video clip in their final video summary as the importance scores and ground truth. Comparisons are performed with the state-of-the-art methods and different variants of FLSum and T-FLSum. The results of F-score and rank correlation coefficients on TVSum and SumMe shows the outstanding performance of the method proposed in this manuscript.

TBDB: a database of structurally annotated T-box riboswitch:tRNA pairs

T-box riboswitches constitute a large family of tRNA-binding leader sequences that play a central role in gene regulation in many gram-positive bacteria. Accurate inference of the tRNA binding to T-box riboswitches is critical to predict their cis-regulatory activity. However, there is no central repository of information on the tRNA binding specificities of T-box riboswitches, and de novo prediction of binding specificities requires advanced knowledge of computational tools to annotate riboswitch secondary structure features. Here, we present the T-box Riboswitch Annotation Database (TBDB, https://tbdb.io), an open-access database with a collection of 23,535 T-box riboswitch sequences, spanning the major phyla of 3,632 bacterial species. Among structural predictions, the TBDB also identifies specifier sequences, cognate tRNA binding partners, and downstream regulatory targets. To our knowledge, the TBDB presents the largest collection of feature, sequence, and structural annotations carried out on this important family of regulatory RNA.

TBDB – A database of structurally annotated T-box riboswitch:tRNA pairs

T-box riboswitches constitute a large family of tRNA-binding leader sequences that play a central role in gene regulation in many gram-positive bacteria. Accurate inference of the tRNA binding to T-boxes is critical to predict their cis-regulatory activity. However, there is no central repository of information on the tRNA binding specificities of T-box riboswitches and de novo prediction of binding specificities requires advance knowledge of computational tools to annotate riboswitch secondary structure features. Here we present T-box annotation Database (TBDB,https://tbdb.io), an open-access database with a collection of 23,497 T-box sequences, spanning the major phyla of 3,621 bacterial species. Among structural predictions, the TBDB also identifies specifier sequences, cognate tRNA binding partners, and downstream regulatory target. To our knowledge, the TBDB presents the largest collection of feature, sequence, and structural annotations carried out on this important family of regulatory RNA.

IsoDetect: Detection of splice isoforms from third generation long reads based on short feature sequences

Aims: Accurately detect isoforms from third generation sequencing data. Background: Transcriptome annotation is the basis for the analysis of gene expression and regulation. The transcriptome annotation of many organisms such as humans is far from incomplete, due partly to the challenge in the identification of isoforms that are produced from the same gene through alternative splicing. Third generation sequencing (TGS) reads provide unprecedented opportunity for detecting isoforms due to their long length that exceeds the length of most isoforms. One limitation of current TGS reads-based isoform detection methods is that they are exclusively based on sequence reads, without incorporating the sequence information of known isoforms. Objective: Develop an efficient method for isoform detection. Method: Based on annotated isoforms, we propose a splice isoform detection method called IsoDetect. First, the sequence at exon-exon junction is extracted from annotated isoforms as the “short feature sequence”, which is used to distinguish different splice isoforms. Second, we aligned these feature sequences to long reads and divided long reads into groups that contain the same set of feature sequences, thereby avoiding the pair-wise comparison among the large number of long reads. Third, clustering and consensus generation are carried out based on sequence similarity. For the long reads that do not contain any short feature sequence, clustering analysis based on sequence similarity is performed to identify isoforms. Result: Tested on two datasets from Calypte Anna and Zebra Finch, IsoDetect showed higher speed and compelling accuracy compared with four existing methods. Conclusion: IsoDetect is a promising method for isoform detection. Other: This paper was accepted by the CBC2019 conference.

Detection method for dam deformation of tailing pond based on fault diagnosis algorithm

The existing methods of dam deformation detection of tailings reservoir have the problems of poor accuracy and slow speed. Therefore, a fault diagnosis algorithm based on tailing dam deformation detection method is proposed. The grey theory is used to accumulate the original feature sequence, and the first cumulative sequence is obtained. Based on this, the grey detection model is constructed, and then the concrete deformation of tailings dam body is accurately detected by precision test. Experimental results show that the method has high accuracy, high speed and practicability