Abstract: A method of attention-based end-to-end (E2E) automatic speech recognition (ASR) training, includes performing cross-entropy training of a model, based on one or more input features of a speech signal, performing beam searching of the model of which the cross-entropy training is performed, to generate an n-best hypotheses list of output hypotheses, and determining a one-best hypothesis among the generated n-best hypotheses list. The method further includes determining a character-based gradient and a word-based gradient, based on the model of which the cross-entropy training is performed and a loss function in which a distance between a reference sequence and the determined one-best hypothesis is maximized, and performing backpropagation of the determined character-based gradient and the determined word-based gradient to the model, to update the model.

Abstract: A system enables tool to process remote log files as if the log files were located locally. The system relies on technology, such a file system in user space (FUSE), to add log file search results to the local file system. In some embodiments, the log file search results are associated with virtual files in the local file system. A remote data store system communicates with the local system to provide the search results in response to a search query.

Abstract: A method for locating a faulty photovoltaic (PV) panel includes controlling an unmanned aerial vehicle (UAV) to fly and perform image capturing, obtaining image information of the PV panel captured by a camera carried by the UAV, obtaining global positioning (GPS) information of the UAV and attitude information of the camera at a shooting time when the camera captures the image information, and, in response to determining that the image information includes fault information of the PV panel, determining a position of the PV panel according to the GPS information of the UAV and the attitude information of the camera at the shooting time.

Abstract: Methods and apparatuses are provided for performing end-to-end speech recognition training performed by at least one processor. The method includes receiving, by the at least one processor, one or more input speech frames, generating, by the at least one processor, a sequence of encoder hidden states by transforming the input speech frames, computing, by the at least one processor, attention weights based on each of the sequence of encoder hidden states and a current decoder hidden state, performing, by the at least one processor, a decoding operation based on a previous embedded label prediction information and a previous attentional hidden state information generated based on the attention weights; and generating a current embedded label prediction information based on a result of the decoding operation and the attention weights.

Abstract: A method and apparatus are provided that analyzing sequence-to-sequence data, such as sequence-to-sequence speech data or sequence-to-sequence machine translation data for example, by minimum Bayes risk (MBR) training a sequence-to-sequence model and within introduction of applications of softmax smoothing to an N-best generation of the MBR training of the sequence-to-sequence model.

Abstract: A computer-implemented method for tracking includes obtaining an infrared image and a visible image from an imaging device supported by a carrier of an unmanned aerial vehicle (UAV), obtaining a combined image based on the infrared image and the visible image, identifying a target in the combined image, and generating control signals for tracking the identified target using the imaging device.

Abstract: An image processing method includes obtaining first contour information of a first image captured by a first lens and second contour information of a second image captured by a second lens, the first and second images being captured at the same time, aligning the first contour information of the first image with the second contour information of the second image to obtain aligning contour information of the first and second contour information, and adjusting a relative position between the first and second images according to the aligning contour information to fuse the first and second images to obtain a fused image, the fused image including first edge information of the first image and second edge information of the second image.

Abstract: A method for identifying a photovoltaic panel includes: acquiring a grayscale image of an infrared image captured by a camera mounted on a UAV, the grayscale image including an image of a photovoltaic panel; performing edge extraction processing on an image in the grayscale image to obtain a monochrome image including a plurality of horizontal lines and a plurality of vertical lines, the horizontal lines being lines in a first direction, an average length of the lines in the first direction being greater than a preset length, the vertical lines being lines in a second direction, and an average length of the lines in the second direction being less than the preset length; and identifying the photovoltaic panel in the monochrome image based on a relative positional relationship between the horizontal lines and the vertical lines in the monochrome image.

Abstract: Methods and apparatuses are provided for performing sequence to sequence (Seq2Seq) speech recognition training performed by at least one processor. The method includes acquiring a training set comprising a plurality of pairs of input data and target data corresponding to the input data, encoding the input data into a sequence of hidden states, performing a connectionist temporal classification (CTC) model training based on the sequence of hidden states, performing an attention model training based on the sequence of hidden states, and decoding the sequence of hidden states to generate target labels by independently performing the CTC model training and the attention model training.

Abstract: Methods and apparatuses are provided for performing end-to-end speech recognition training performed by at least one processor. The method includes receiving, by the at least one processor, one or more input speech frames, generating, by the at least one processor, a sequence of encoder hidden states by transforming the input speech frames, computing, by the at least one processor, attention weights based on each of the sequence of encoder hidden states and a current decoder hidden state, performing, by the at least one processor, a decoding operation based on a previous embedded label prediction information and a previous attentional hidden state information generated based on the attention weights; and generating a current embedded label prediction information based on a result of the decoding operation and the attention weights.

Abstract: Methods and apparatuses are provided for performing acoustic to word (A2W) speech recognition training performed by at least one processor. The method includes initializing, by the at least one processor, one or more first layers of a neural network with phone based Connectionist Temporal Classification (CTC), initializing, by the at least one processor, one or more second layers of the neural network with grapheme based CTC, acquiring, by the at least one processor, training data and performing, by the at least one processor, A2W speech recognition training based the initialized one or more first layers and one or more second layers of the neural network using the training data.

Abstract: A control device includes a display configured to show an image captured by an imaging device supported by a movable object and one or more processors configured to obtain information about a user input indicative of a target shown within the image, and generate data, based on the information about the user input indicative of the target, to effectuate an automatic control of a zoom level of the imaging device and an attitude of the imaging device relative to the target.

Abstract: The claimed subject matter includes a system and method for recognizing mixed speech from a source. The method includes training a first neural network to recognize the speech signal spoken by the speaker with a higher level of a speech characteristic from a mixed speech sample. The method also includes training a second neural network to recognize the speech signal spoken by the speaker with a lower level of the speech characteristic from the mixed speech sample. Additionally, the method includes decoding the mixed speech sample with the first neural network and the second neural network by optimizing the joint likelihood of observing the two speech signals considering the probability that a specific frame is a switching point of the speech characteristic.

Abstract: The claimed subject matter includes a system and method for recognizing mixed speech from a source. The method includes training a first neural network to recognize the speech signal spoken by the speaker with a higher level of a speech characteristic from a mixed speech sample. The method also includes training a second neural network to recognize the speech signal spoken by the speaker with a lower level of the speech characteristic from the mixed speech sample. Additionally, the method includes decoding the mixed speech sample with the first neural network and the second neural network by optimizing the joint likelihood of observing the two speech signals considering the probability that a specific frame is a switching point of the speech characteristic.

Abstract: The claimed subject matter includes a system and method for recognizing mixed speech from a source. The method includes training a first neural network to recognize the speech signal spoken by the speaker with a higher level of a speech characteristic from a mixed speech sample. The method also includes training a second neural network to recognize the speech signal spoken by the speaker with a lower level of the speech characteristic from the mixed speech sample. Additionally, the method includes decoding the mixed speech sample with the first neural network and the second neural network by optimizing the joint likelihood of observing the two speech signals considering the probability that a specific frame is a switching point of the speech characteristic.

Abstract: The claimed subject matter includes a system and method for recognizing mixed speech from a source. The method includes training a first neural network to recognize the speech signal spoken by the speaker with a higher level of a speech characteristic from a mixed speech sample. The method also includes training a second neural network to recognize the speech signal spoken by the speaker with a lower level of the speech characteristic from the mixed speech sample. Additionally, the method includes decoding the mixed speech sample with the first neural network and the second neural network by optimizing the joint likelihood of observing the two speech signals considering the probability that a specific frame is a switching point of the speech characteristic.

Abstract: The claimed subject matter includes a system and method for recognizing mixed speech from a source. The method includes training a first neural network to recognize the speech signal spoken by the speaker with a higher level of a speech characteristic from a mixed speech sample. The method also includes training a second neural network to recognize the speech signal spoken by the speaker with a lower level of the speech characteristic from the mixed speech sample. Additionally, the method includes decoding the mixed speech sample with the first neural network and the second neural network by optimizing the joint likelihood of observing the two speech signals considering the probability that a specific frame is a switching point of the speech characteristic.

Abstract: The claimed subject matter includes a system and method for recognizing mixed speech from a source. The method includes training a first neural network to recognize the speech signal spoken by the speaker with a higher level of a speech characteristic from a mixed speech sample. The method also includes training a second neural network to recognize the speech signal spoken by the speaker with a lower level of the speech characteristic from the mixed speech sample. Additionally, the method includes decoding the mixed speech sample with the first neural network and the second neural network by optimizing the joint likelihood of observing the two speech signals considering the probability that a specific frame is a switching point of the speech characteristic.

Abstract: A wafer electroplating apparatus with a function of bubble removal includes an electroplating bath main body and a fixing device. The electroplating bath main body has an inlet device, a first de-bubble tank and at least an air hole. The fixing device has a second de-bubble tank and an outer shell. The air hole guides gathering bubbles to an outside of the electroplating bath main body so as to remove bubbles. The fixing device can be put into the first de-bubble tank within the electroplating bath main body to form a de-bubble area and is separated easily therefrom to clean the wafer electroplating apparatus. The electroplating bath main body further includes a baffle for rectifying electroplating solution flow before entering the inlet device.