This paper proposes an attention pooling based representation learning method for speech emotion recognition (SER). The emotional representation is learned in an end-to-end fashion by applying a deep convolutional neural network (CNN) directly to spectrograms extracted from speech utterances. Motivated by the success of GoogleNet, two groups of filters with different shapes are designed to capture both temporal and frequency domain context information from the input spectrogram. The learned features are concatenated and fed into the subsequent convolutional layers. To learn the final emotional representation, a novel attention pooling method is further proposed. Compared with the existing pooling methods, such as max-pooling and average-pooling, the proposed attention pooling can effectively incorporate class-agnostic bottom-up, and class-specific top-down, attention maps. We conduct extensive evaluations on benchmark IEMOCAP data to assess the effectiveness of the proposed representation. Results demonstrate a recognition performance of 71.8% weighted accuracy (WA) and 68% unweighted accuracy (UA) over four emotions, which outperforms the state-of-the-art method by about 3% absolute for WA and 4% for UA.

The recurrent neural networks (RNN) have shown promising results in sentence matching tasks, such as paraphrase identification (PI), natural language inference (NLI) and answer selection (AS). However, the recurrent architecture prevents parallel computation within a sequence and is highly time-consuming. To overcome this limitation, we propose a gated convolutional neural network (GCNN) for sentence matching tasks. In this model, the stacked convolutions encode hierarchical contextaware representations of a sentence, where the gating mechanism optionally controls and stores the convolutional contextual information. Furthermore, the attention mechanism is utilized to obtain interactive matching information between sentences. We evaluate our model on PI and NLI tasks, and the experiments demonstrate the advantages of the proposed approach in terms of both speed and accuracy performance.

No abstract

Pseudorabies virus (PRV) primarily infects swine but can infect cattle, dogs, and cats. Several studies have reported that PRV can cross the specie barrier and induce human encephalitis, but a definitive diagnosis of human PRV encephalitis is debatable due to the lack of PRV DNA detection. Here, we report a case of human PRV encephalitis diagnosed by the next-generation sequencing (NGS) of PRV sequences in the cerebrospinal fluid (CSF) of a patient. A male pork vendor developed fever and seizures for 6 days. NGS results showed PRV sequences in his CSF and blood. Sanger sequencing showed that PRV DNA in the CSF and PRV antibodies in both the CSF and blood were positive. MRI results revealed multiple inflammatory lesions in the bilateral hemisphere. Based on the clinical and laboratory data, we diagnosed the patient with PRV encephalitis. This case suggests that PRV can infect humans, causing severe viral encephalitis. People at risk of PRV infection should improve their self-protection awareness.

The Xinjiang region in northwest China is a historically important geographical passage between East and West Eurasia. By sequencing 201 ancient genomes from 39 archaeological sites, we clarify the complex demographic history of this region. Bronze Age Xinjiang populations are characterized by four major ancestries related to Early Bronze Age cultures from the central and eastern Steppe, Central Asian, and Tarim Basin regions. Admixtures between Middle and Late Bronze Age Steppe cultures continued during the Late Bronze and Iron Ages, along with an inflow of East and Central Asian ancestry. Historical era populations show similar admixed and diverse ancestries as those of present-day Xinjiang populations. These results document the influence that East and West Eurasian populations have had over time in the different regions of Xinjiang.