Abstract: Disclosed are a speech recognition method, apparatus, computer device and storage medium. The method includes: performing a framing and an acoustic feature extraction of a speech-information-to-be-tested according to a default rule to obtain a frame-level speech feature sequence; dividing the frame-level speech feature sequence into n blocks sequentially; inputting all blocks into a preset bidirectional LSTM-RNN model parallelly to obtain an output result of the corresponding neuron in an output layer of the preset bidirectional LSTM-RNN model corresponding to the forward recognition result and backward recognition result of each block to obtain a speech recognition result of the speech-information-to-be-tested. The present application can improve the speech recognition effect significantly and reduce the time delay of the speech decoding effectively.

Abstract: A system for directory level atomic commits includes an interface and a processor. The interface is configured to receive an indication to provide a set of files. The processor is configured to determine whether a file in a directory has been either 1) atomically committed or 2) written by a non-atomic process and not designated as deleted and provide the file as one file of the set of files in the event that the file in the directory has been either 1) atomically committed or 2) written by a non-atomic process and not designated as deleted.

Abstract: A speech recognition method, comprising: acquiring speech data to be recognized; extracting a Filter Bank feature and a Mel-Frequency Cepstral Coefficient (MFCC) feature in the speech data; using the MFCC feature as input data of a trained Gaussian Mixture Model-Hidden Markov Model (GMM-HMM) model, acquiring a first likelihood probability matrix outputted by the trained GMM-HMM model; using the Filter Bank feature as an input feature of a trained long short-term memory (LSTM) model which has a connection unit, acquiring a posterior probability matrix outputted by the LSTM model; using the posterior probability matrix and the first likelihood probability matrix as input data of a trained HMM model, acquiring an second likelihood probability matrix; and acquiring a target word sequence corresponding to the speech data to be recognized from a phoneme decoding network according to the second likelihood probability matrix.

Abstract: The present invention relates to a dual-arm clamping type holder for transmission electron microscopy grids and preparation method thereof. The preparation method comprises: firstly manufacturing a frame with two adjacent arms located on a same plane and have a clamping structure by a hard material; then putting the frame in a molten adhering liquid so that the frame is dipped with the adhering liquid; and finally, taking out the frame dipped with the adhering liquid, and waiting for the adhering liquid to solidify into adhering layers along the arms. The dual-arm clamping type holder manufactured by the method of the present invention comprises a frame and adhering layers; and the adhering layers adhere to the inner sides of the clamping structure between the two adjacent arms of the frame.

Abstract: The present invention relates to a magnetic holder for immunoelectron microscopy grids. The holder comprises a frame, a magnet and a hydrophobic layer. The device can use a magnetic force to simultaneously attach the outer rings of nickel grids to the frame, so that a batch operation (such as rinsing, immunolabeling and dyeing) of the nickel grids can be realized. In addition, due to the hydrophobic effect of the hydrophobic layer, the holder can reduce the amount of the liquid carried by the nickel grids in the process of continuously transferring the nickel grids between different types of liquids to almost zero. Compared with the prior art, the magnetic holder effectively reduces the probability of cross-contamination between reagents.

Abstract: A laser cleaning method and device for improving uniformity of a laser cleaning surface are provided. The laser cleaning method includes: applying a peaked-top sine wave signal to a motor; controlling a galvanometer to swing in a reciprocated manner by the motor; shaping a laser beam emitted by a laser to a linear beam by the reciprocated swing of the galvanometer; and performing laser cleaning using the shaped linear beam.

Abstract: An acousto-optic Q switch, a resonant cavity, and a pulse laser device for improving laser device power. The acousto-optic Q switch includes: a transparent optical element configured to form a phase grating that diffracts laser; a piezoelectric transducer arranged at one end of the transparent optical element and configured to convert electrical energy into ultrasonic energy to form the phase grating in the transparent optical element; and an absorber arranged at the other end of the transparent optical element to absorb the ultrasonic energy.

Abstract: The present invention relates to a dual-arm clamping type holder for transmission electron microscopy grids and preparation method thereof. The preparation method comprises: firstly manufacturing a frame with two adjacent arms located on a same plane and have a clamping structure by a hard material; then putting the frame in a molten adhering liquid so that the frame is dipped with the adhering liquid; and finally, taking out the frame dipped with the adhering liquid, and waiting for the adhering liquid to solidify into adhering layers along the arms. The dual-arm clamping type holder manufactured by the method of the present invention comprises a frame and adhering layers; and the adhering layers adhere to the inner sides of the clamping structure between the two adjacent arms of the frame.

Abstract: The present invention relates to a magnetic holder for immunoelectron microscopy grids. The holder comprises a frame, a magnet and a hydrophobic layer. The device can use a magnetic force to simultaneously attach the outer rings of nickel grids to the frame, so that a batch operation (such as rinsing, immunolabeling and dyeing) of the nickel grids can be realized. In addition, due to the hydrophobic effect of the hydrophobic layer, the holder can reduce the amount of the liquid carried by the nickel grids in the process of continuously transferring the nickel grids between different types of liquids to almost zero. Compared with the prior art, the magnetic holder effectively reduces the probability of cross-contamination between reagents.

Abstract: An acoustic model training method, a speech recognition method, an apparatus, a device and a medium. The acoustic model training method comprises: performing feature extraction on a training speech signal to obtain an audio feature sequence; training the audio feature sequence by a phoneme mixed Gaussian Model-Hidden Markov Model to obtain a phoneme feature sequence; and training the phoneme feature sequence by a Deep Neural Net-Hidden Markov Model-sequence training model to obtain a target acoustic model. The acoustic model training method can effectively save time required for an acoustic model training, improve the training efficiency, and ensure the recognition efficiency.

Abstract: A laser cleaning system including a laser source, an energy-transferring optical fiber, a laser cleaning head, a coreless motor, a connection lens barrel, and a mirror.

Abstract: A semiconductor device includes a substrate having a NMOS region and a PMOS region; a gate structure extending along a first direction from the NMOS region to the PMOS region on the substrate; and a first contact plug landing directly on the gate structure closer to the PMOS region from a boundary separating the NMOS region and the PMOS region. Preferably, the semiconductor device further includes a first source/drain region extending along a second direction adjacent to two sides of the gate structure on the NMOS region and a second source/drain region extending along the second direction adjacent to two sides of the gate structure on the PMOS region.

Abstract: An acoustic model training method, a speech recognition method, an apparatus, a device and a medium. The acoustic model training method comprises: performing feature extraction on a training speech signal to obtain an audio feature sequence; training the audio feature sequence by a phoneme mixed Gaussian Model-Hidden Markov Model to obtain a phoneme feature sequence; and training the phoneme feature sequence by a Deep Neural Net-Hidden Markov Model-sequence training model to obtain a target acoustic model. The acoustic model training method can effectively save time required for an acoustic model training, improve the training efficiency, and ensure the recognition efficiency.

Abstract: A contact switch, contact switch system, and interactive building block system carrying the contact switch based on UHF RFID technology for detecting mutual contact events are provided. The contact switch system includes a plurality of contact switches each including a chip and an antenna not connected to the chip. When the contact switch is in contact with another contact switch, the chip is connected to another antenna of the another contact switch and the antenna is connected to another chip of the another contact switch. The chip receives and converts a radio wave from the another antenna into power, and the power converts identifying information of the chip into a radio wave, which is sent to a reader. When the reader reads two encoded identifying information at the same time, whether the two contact switches are in contact with each other is determined by the contact switch system.

Abstract: Included is a semiconductor multilayer film in which a non-doped InAlN layer and a GaN layer formed on said InAlN layer and containing a dopant are stacked a plurality of times.

Abstract: A light-emitting element and a light-emitting device having low light loss, high luminance, and high light extraction efficiency are provided. The light-emitting element includes: a semiconductor structure layer having a light-emitting layer; a light-transmitting substrate provided on the semiconductor structure layer; a wavelength conversion layer disposed on the light-transmitting substrate; a light-transmitting covering member configured to cover at least a part of a side surface of the light-transmitting substrate and have transparency to light from the light-emitting layer; and a light-shielding member configured to entirely cover surfaces including a surface of the light-transmitting covering member, and including a side surface of the semiconductor structure layer, a side surface of the light-transmitting substrate, and a side surface of the wavelength conversion layer.

Abstract: A method and a system for determining a discharging process of a battery are provided. The method includes the following steps. Measuring charging/discharging information of the battery. Calculating a charging/discharging characteristic of the battery according to the charging/discharging information. Aligning the charging/discharging characteristic of the battery according to a comparison characteristic point of a comparison characteristic to obtain an aligned charging/discharging characteristic. Determining whether the battery is normal according to the aligned charging/discharging characteristic or a coulombic efficiency of the battery. Calculating a safety probability of the battery according to the aligned charging/discharging characteristic and resistance of an internal short circuit of the battery when the battery is determined as abnormal. Determining a discharging process of the battery according to the safety probability of the battery.

Abstract: A fire control device comprises a box, a power wire, a pressure relieving check valve, a fire extinguishing check valve and a fire extinguisher. The box is configured to accommodate a battery system, and the power wire is configured to couple to the battery system. The pressure relieving check valve and the fire extinguishing check valve extend through the box, and a state of the pressure relieving check valve is switched between open and closed states according to a pressure difference between an inside and an outside of the box. A state of the fire extinguishing check valve is switched between open and close states according to a pressure difference between the inside and the outside of the box. The fire extinguisher is connected to the pressure relieving check valve. The fire extinguisher is switched between starting and stopping modes according to the state of the pressure relieving check valve.

Abstract: Disclosed are a speech recognition method, apparatus, computer device and storage medium. The method includes: performing a framing and an acoustic feature extraction of a speech-information-to-be-tested according to a default rule to obtain a frame-level speech feature sequence; dividing the frame-level speech feature sequence into n blocks sequentially; inputting all blocks into a preset bidirectional LSTM-RNN model parallelly to obtain an output result of the corresponding neuron in an output layer of the preset bidirectional LSTM-RNN model corresponding to the forward recognition result and backward recognition result of each block to obtain a speech recognition result of the speech-information-to-be-tested. The present application can improve the speech recognition effect significantly and reduce the time delay of the speech decoding effectively.

Abstract: The present invention discloses a method for interpreting and evaluating production profile of multi-layer gas reservoir based on downhole distributed temperature monitoring, including: obtaining downhole distributed temperature monitoring data of target well; preprocessing the downhole distributed temperature monitoring data; segmenting the temperature monitoring data according to test curve characteristics of the target well and logging interpretation results; using a multi-layer gas reservoir seepage pressure field—temperature field coupled model to calculate temperatures of each layer in the borehole production profile of the target well by numerical simulation method; comparing the temperatures of each layer of the borehole production profile with the temperature monitoring data after segmentation, obtaining the optimal flow rate of each production layer with optimization theories, and obtaining the production profile of the target well based on the optimal flow rate of each production layer.