基于双向循环神经网络的汉语语音识别*

当前基于深度神经网络模型中,虽然其隐含层可设置多层,对复杂问题适应能力强,但每层之间的节点连接是相互独立的,这种结构特性导致了在语音序列中无法利用上下文相关信息来提高识别效果,而传统的循环神经网络虽然做出了改进,但是只能对上文信息进行利用。针对以上问题,该文采用可以同时利用语音序列中上下文相关信息的双向循环神经网络模型与深度神经网络模型相结合,并应用于语音识别。构建具有5层隐含层的模型,其中第3层为双向循环神经网络结构,其他层采用深度神经网络结构。实验结果表明:加入了双向循环神经网络结构的模型与其他模型相比,较好地提高了识别正确率;噪声对双向循环神经网络汉语识别有重要影响,尤其是训练集和测试集附加噪声类型不同时,单一的含噪声语音的训练模型无法适应不同噪声类型的语音识别;调整神经网络模型中隐含层神经元数量后,识别正确率并不是一直随着隐含层中神经元数量的增加而增加,神经元数量数目增加到一定程度后正确率出现了降低的趋势。     

A solvent-governed surface state strategy for rational synthesis of N and S co-doped carbon dots with multicolour fluorescence

ABSTRACT

Multicolour emissive carbon dots (CDs) are widely investigated by virtue of their merits on fluorescent properties. Method on heteroatom doping assisted with various solvents has been proved efficient in achieving multiple-colour-emissive CDs, especially long-wavelength emission. Herein, a synthesis of multicolour-emissive CDs by controlled surface function is reported. By tuning the thermal-pyrolysis temperature and molar ratio of reactants, optimal emission of the resulted CDs gradually shifts from blue to yellow light with the assistance of different solvents. According to the emissive relationship dependent on excitation, fluorescence lifetimes, and FT-IR of these CDs, the different surface states participated with S and N elements on the surface of carbogenic core govern fluorescent colours of the CDs. In terms of the applications, blue CDs (B-CDs) exhibits high sensitivity for ion detections of Ag⁺ and Fe³⁺, which is further illustrated to have different quenching mechanisms each other because that these ions have the affinity interaction with different surface groups of the CDs. Moreover, blue and yellow CDs solutions are mixed with PVP water solution to fabricate white-light CDs/PVP film, which exhibits stable fluorescence with a CIE coordinate of (0.32, 0.33) and endows these CDs as potentially fluorescent nanomaterial in the solid state lighting field.     

Volatile organic compounds in head and neck squamous cell carcinoma—An in vitro pilot study

Owing to the lack of specific symptoms, diagnosis of head and neck squamous cell carcinoma (HNSCC) may be delayed. We evaluated volatile organic compounds in tumor samples from patients suffering from HNSCC and tested the hypothesis that there is a characteristic altered composition in the headspace of HNSCC compared with control samples from the same patient with normal squamous epithelium. These results provide the basis for future noninvasive breath analysis in HNSCC. Headspace air of suspected tumor and contralateral control samples in 20 patients were analyzed using ion-mobility spectrometry. Squamous cell carcinoma was diagnosed in 16 patients. In total, we observed 93 different signals in headspace measurements. Squamous cell carcinomas revealed significantly higher levels of volatile cyclohexanol (0.54 ppb_v, 25th to 75th percentiles 0.35–0.86) compared with healthy squamous epithelium (0.24 ppb_v, 25th to 75th percentiles 0.12–0.3; p < 0.001). In conclusion, head and neck squamous cell carcinoma emitted significantly higher levels of volatile cyclohexanol in headspace compared with normal squamous epithelium. These findings form the basis for future breath analysis for diagnosis, therapy control and the follow-up of HNSSC to improve therapy and aftercare.     

A DNA Switch for Detecting Single Nucleotide Polymorphism within a Long DNA Sequence Under Denaturing Conditions

DNA detection is usually conducted under nondenaturing conditions to favor the formation of Watson–Crick base-paring interactions. However, although such a setting is excellent for distinguishing a single-nucleotide polymorphism (SNP) within short DNA sequences (15–25 nucleotides), it does not offer a good solution to SNP detection within much longer sequences. Here we report on a new detection method capable of detecting SNP in a DNA sequence containing 35–90 nucleotides. This is achieved through incorporating into the recognition DNA sequence a previously discovered DNA molecule that forms a stable G-quadruplex in the presence of 7 molar urea, a known condition for denaturing DNA structures. The systems are configured to produce both colorimetric and fluorescent signals upon target binding.     

贝叶斯优化卷积神经网络公共场所异常声识别*

针对公共场所异常声的感知和识别问题,提出一种基于贝叶斯优化卷积神经网络的识别方法。提取声信号的Gammatone倒谱系数、倍频程功率谱、短时能量和谱质心,组合成声信号的特征图。构建卷积神经网络作为分类器,利用递增的卷积核设置和池化操作处理不同尺度的特征。基于贝叶斯优化算法优化卷积神经网络的模型参数,对包括火苗噼啪声、婴儿啼哭声、烟花燃放声、玻璃破碎声和警报声的5种公共场所异常声进行识别。该方法的识别结果与基于不同的特征提取和分类器方案得到的识别结果进行比较,结果表明该方法的识别效果优于其他特征提取和分类器方案的识别效果。最后分析了该方法在不同信噪比噪声干扰下的识别结果,验证了该方法的有效性。     

Intertwined Detection and Recognition Roles of Tetrazine in Synergistic Anion-π and H-bond Based Anion Receptor

The intrinsic properties of tetrazine as a π-anion receptor and as an on/off recognition probe merged with H-bond ability of an urea motif into a single architecture constitutes a new generation of well-defined anion receptors. Complexation properties directly benefit from the dual and synergistic contribution of tetrazine and urea. In this study, we report on the synthesis and assessment of binding properties to anions of diverse geometries. Association constants have been predicted by theoretical calculations and evaluated by multiple and complementary experimental techniques including electrospray-mass tandem spectroscopy, NMR, UV-visible, steady state fluorescence spectroscopies and time resolved fluorescence. These results provide the basis for a better understanding of both the complexation and the anion-dependent quenching mechanism.     

OCS经由F 31Π里德堡态生成CO(X1Σ+)+S(1D2)产物的波长相关性光解离

本文利用时间切片离子速度成像技术在134∽140 nm波段研究了OCS分子经由F 3¹Π里德堡态的真空紫外光解离动力学. 在选取的5个分别对应OCS(F 3¹Π, v₁=0∽4)的伸缩振动激发的光解波长，实验测得了来自CO(X¹Σ⁺)+S(¹D₂)产物通道的SS(¹D₂))实验影像，并获得了总平动能谱和CO(X¹Σ⁺, v)共生产物的振动布居及角分布. 结果分析表明OCS分子解离生成CO(X¹Σ⁺)+S(¹D₂)产物的过程经历了上态F 3¹Π 与C_?v和C_s构型的下电子态间非绝热耦合过程. 实验结果显示了很强的波长相关性：OCS (F 3¹Π, v₁)的较低转动激发态(v₁=0∽2)和较高转动激发态(v₁=3, 4)的CO(X¹Σ⁺)产物的振动布居和角分布具有显著差异，表明该解离过程中具有不同的解离机理. 本结果提供了振动耦合可能对真空紫外光解离动力学产生关键作用的相关证据.     

Study of water uptake in poly(ether ether ketone) irradiated by MeV O+ and Au+ ions

Poly(ether ether ketone) (PEEK) was irradiated with 4?MeV O⁺ and 5 and 10?MeV Au⁺ ions to the fluences from 10¹² to 10^14?cm^?2 and then treated in 5 M/l water solution of LiCl for one month at room temperature. After drying and removal of LiCl surface contamination, the depth distribution of LiCl embeded in PEEK was measured by the neutron depth profilig method (NDP) sensitive to ⁶Li isotope. Embeded LiCl is believed to map distribution of water diffusing into PEEK interior. The results show that the PEEK irradiated to the fluences above 1.10¹³cm^?2 is prone to water penetration to the depths of few microns. On the pristine PEEK and that irradiated to lower ion fluences only a surface Li contamination is observed.     

Self‐Assembled Discrete Molecules for Sensing Nitroaromatics

Efficient sensing of trace amount nitroaromatic (NAC) explosives has become a major research focus in recent time due to concerns over national security as well as their role as environment pollutants. NO₂‐containing electron‐deficient aromatic compounds, such as picric acid (PA), trinitrotoluene (TNT), and dinitrotoluene (DNT), are the common constituents of many commercially available chemical explosives. In this article, we have summarized our recent developments on the rational design of electron‐rich self‐assembled discrete molecular sensors and their efficacy in sensing nitroaromatics both in solution as well as in vapor phase. Several π‐electron‐rich fluorescent metallacycles (squares, rectangles, and tweezers/pincers) and metallacages (trigonal and tetragonal prisms) have been synthesized by means of metal–ligand coordination‐bonding interactions, with enough internal space to accommodate electron‐deficient nitroaromatics at the molecular level by multiple supramolecular interactions. Such interactions subsequently result in the detectable fluorescence quenching of sensors even in the presence of trace quantities of nitroaromatics. The fascinating sensing characteristics of molecular architectures discussed in this article may enable future development of improved sensors for nitroaromatic explosives.     

Synthesis of Mesoporous Wall‐Structured TiO2 on Reduced Graphene Oxide Nanosheets with High Rate Performance for Lithium‐Ion Batteries

Mesoporous wall‐structured TiO₂ on reduced graphene oxide (RGO) nanosheets were successfully fabricated through a simple hydrothermal process without any surfactants and annealed at 400 °C for 2 h under argon. The obtained mesoporous structured TiO₂–RGO composites had a high surface area (99 0307 m² g^?1) and exhibited excellent electrochemical cycling (a reversible capacity of 260 mAh g^?1 at 1.2 C and 180 mAh g^?1 at 5 C after 400 cycles), demonstrating it to be a promising method for the development of high‐performance Li‐ion batteries.