Search

Speech-to-text conversion has been extensively deployed for a variety of applications. To implement speech-to-text conversion on energy-constrained edge devices, a hybrid algorithm is adopted in this work. A bidirectional recurrent neural network (BRNN), composed of the light gated recurrent units (LiGRUs), is included to achieve a high speech-to-text accuracy with fewer network parameters. A network compression scheme, including scaling factor pruning (SFP), multi-bit clustering (MBC), and linear quantization (LQ), is proposed to minimize the complexity of the BRNN. The network size and the computational complexity are reduced by $29.8\times $ and $73.2\times $ , respectively, with only a 1% accuracy drop. An array of processing elements (PEs) are designed to perform BRNN inference. To minimize the system latency under the silicon area constraint, an optimal design with four PEs that can achieve 100% utilization for BRNN inference is selected. A memory-efficient decoding scheme without backtracking is utilized to reduce the memory usage by 21% compared to the direct-mapped implementation. The proposed speech-to-text accelerator achieves a phone error rate (PER) of 15.2% on the TIMIT with a network size of 1.51 MB. Fabricated in 28-nm CMOS, the chip consumes energy of 12.7 mJ/frame when operated at 1.25 MHz from a 0.6-V supply. Compared to the state-of-the-art designs, this work achieves a 6.5– $177\times $ lower normalized energy and a 37.5– $50\times $ lower attainable latency, with a 3.3%–8.5% lower PER.

Augmented reality (AR) has been applied to various mobile applications. Modern AR algorithms include neural networks, such as convolutional neural networks (CNNs) and graph convolutional networks (GCNs). The high computational complexity of these networks poses challenges for real-time operation on energy-constrained devices. This article presents the first energy-efficient hybrid CNN-GCN system-on-chip (SoC) for mobile AR. A CNN engine exploits the channel-wise structured feature sparsity to eliminate redundant computations and data movements. By utilizing the proposed channel-sparse encoding scheme on a specialized processing element (PE) architecture, up to $8{\times }$ higher throughput and $6.1{\times }$ higher energy efficiency can be achieved. A reconfigurable convolution PE (CPE) array is deployed for efficient CNN inference. A GCN engine is designed to implement skeleton-based action recognition and gesture recognition. Up to 71% of total operations and 39% of memory footprint can be reduced by leveraging the data and graph properties. A RISC-V MCU is integrated for system control and network deployment. The proposed SoC is implemented in a 28-nm CMOS technology with a core area of 8.28 mm2. By exploiting various sparsity levels across network layers, the chip achieves an up to 3.277-TOPS peak performance and a 25.1-TOPS/W energy efficiency for sparse CNN inference, $2.0{\times }$ higher energy-efficient than prior arts. It also achieves an up to 72 actions/s recognition throughput, $18{\times }$ faster than the state of the art.

A discrete element method was used to study the evolution of particle crushing in a rockfill sample subjected to triaxial shear. A simple procedure was developed to generate clusters with arbitrary shapes, which resembled real rockfill particles. A theoretical method was developed to define the failure criterion for an individual particle subjected to an arbitrary set of contact forces. Then, a series of numerical tests of large-scale drained triaxial tests were conducted to simulate the behaviors of the rockfill sample. Finally, we examined the development of micro-characteristics such as particle crushing, contact characteristics, porosity, deformation, movement, and energy dissipation. The simulation results were partially compared with the laboratory experiments, and good agreement was achieved, demonstrating that the particle crushing model proposed can be used to simulate the drained triaxial test of rockfill materials. Based on a comparison of macro behaviors of the rockfill sample and micro structures of the particles, the microscopic mechanism of the rockfill materials subjected to triaxial shear was determined qualitatively. It is shown that the crushing rate, rather than the number of crushed particles, can be used to reflect the relationship between macro- and micro-mechanical characteristics of rockfill materials. These research results further develop our understanding of the deformation mechanism of rockfill materials.

This paper describes the design and implementation of a high efficiency motion controller system consisting of host machine and teach pendant. The host machine performs some of the position control processing tasks and receives the processed data from Teach Pendant. Starting with introduction to existing excellent 4-axismotion controller PCL6045B, It describes the overall structure of the system and the hardware that make up the motion controller. Then, this paper discusses the design method of software system, and the motion control chip PCL6045B driver in WindowsCE. Besides, the flow diagram of the use of industrial communication networks transmission about CC_LINK is showed in detail. After testing the system, it has implemented the high precision position controller. Moreover, it has the advantages of low cost, easy use.

When pseudostatic method is employed to analyze slope seismic stability of high earth- rock dams, earthquake load is computed according to dynamic seismic coefficient figure in the present Specifications for Seismic Design of Hydraulic Structures (DL5073-1997). The figure is only suitable for earth-rock dams lower than 150m. While at the present time, many earth-rock dams to be constructed are higher than 150m. Compared with low dams, high order self-vibrating period of high dams is easy to coincide with seismic predominant period, and high order vibrations are constantly to be activated and amplified, which result in seismic acceleration distribution is different from low dams. With analytical and finite element method respectively, seismic acceleration distribution of high earth-rock dams is analyzed. Suggested dynamic seismic coefficient figure of earth-rock dams 300m-level in height is put forward, which offer a supplement for the present Code for Seismic Design of Hydraulic Structure.

A traditional dissolution pumping system was recently replaced with a fiber optic interface between the spectrometer and the samples. However, the system was limited to a single sample vessel. In this study, a dissolution testing system with six vessels connected to a diode array spectrometer via six optical fibers was investigated. A bifurcated fiber optic bundle was used to transfer the light from the source to the dissolution vessels and was networked so that spectra of each sample can be measured periodically. A full spectrum calibration method based on Principal Component Regression (PCR) was used to determine the concentrations of active ingredients and to account for interferences due to excipients in tablet formulations. Results on this new fiber optic interface system are compared with those obtained previously with the traditional pumping system. Standard errors of prediction are between 1.5 and 3.2% using cross-validation and between 1.1 and 1.7% for the direct validation of two active ingredients in two different drug formulations.

A segmental probabilistic model based on an orthogonal polynomial representation of speech signals is proposed. Unlike the conventional frame based probabilistic model, this segment based model concatenates the similar acoustic characteristics of consecutive frames into an acoustic segment and represents the segment by an orthogonal polynomial function. An iterative algorithm that performs recognition and segmentation processes is proposed for estimating the segment model. This segment model is applied in the text independent speaker verification. Tests were carried out on a 20-speaker database. With the best version of the model, an equal error rate of 0.59% can be reached, for test utterances of 10 digits. This corresponds to a relative error rate reduction of more than 50%, compared to the conventional frame based probabilistic model.

A calibration transfer method, piecewise direct standardization (PDS), was applied to a set of two-component samples measured on the same UV-visible spectrometer with the use of a cuvette cell with a 10-mm pathlength and a fiber-optic probe with a 2-mm pathlength. Piecewise direct standardization proceeds by determining a structured transformation matrix using the spectra of a few samples measured with both devices. This transformation matrix can then be used to transform any spectrum measured on one device to that obtained on another device, thereby making the calibration model transferable between devices. We used the spectra measured in a cuvette as the standard set and transferred the calibration model obtained for these spectra to spectra measured with a 2-mm fiber-optic probe on the same instrument. The total standard error of prediction (SEP) for the fiber-optic probe was 5.84 before the calibration transfer and 1.87 afterwards. Spectra were also processed by taking the Fourier transform prior to the calibration transfer. The 512 data points in each spectrum were compressed to 32 terms, starting with the first term after the dc offset. This processing reduced the background and the noise. As a consequence, in the Fourier domain, the total SEP was 5.69 before the calibration standardization and 0.79 after the calibration standardization. A calibration transfer was also performed between two fiber-optic probes; the total SEP in the spectral domain was 2.16 prior to the transfer and 1.04 after the transfer, whereas in the Fourier domain the SEP was 1.50 prior to the transfer and 0.77 after the transfer.

A pattern-recognition algorithm combined with near-infra-red reflectance spectroscopy has been modified to function as a non-destructive analysis technique for identifying dyes present on textiles. Samples of 261 dyes and textiles were measured in the 1100-2500 nm region to form a near-infrared (reflectance) spectral library. Principal component analysis (PCA) was used to generate an orthonormal reference library from the library of original spectra. The PCA algorithm treats the spectra in the library as an ncomponent quantitative analysis problem in which each spectrum represents a standard mixture having a concentration of 1. 0 for that component. Spectra of dyed textiles were used as an unknown set in a library search. This new method saves time and materials in comparison with traditional methods of analysing dyes present on textile fibres. The library of dye spectra can be developed from measurements made directly on dye powder without interference from inorganic diluents. The method was successfully used to identify the dyes present on five textiles. The technique is particularly well suited for studying forensic, historic and archaeological textiles because of its non-destructive nature and ability to analyse small amounts of sample.

The use of discriminative training to construct hidden Markov models of speakers for verification and identification is studied. As opposed to conventional maximum likelihood training which estimates a speaker’s model based only on the training utterances from the same speaker, a discriminative training approach is used which takes into account the models of other competing speakers and formulates the optimization criterion such that speaker separation is enhanced and speaker recognition error rate on the training data is directly minimized. The optimization solution is obtained with a probabilistic descent algorithm. For all experiments an isolated digit database consisting of 100 speakers is used. For speaker identification, the resulting discriminative speaker models reduce the identification error rate by more than 25% over the results obtained with the conventional training algorithm. A new normalized score function is proposed which makes the verification formulation consistent with the minimum error training objective. When combining the proposed verification score function with discriminative training, an average equal error rate of 0.8% is achieved using only one‐digit test utterances. This represents an error rate reduction of over 80% from an average equal error rate of 6.1% when using the conventional algorithm for training and the unnormalized score function for testing.