Microsoft Technology Licensing, LLC

International Business Machines Corporation

University of California

Beijing Volcano Engine Technology Co., Ltd.

Centaurus Biopharma Co., Ltd.

Sheng Zhao

Chen Wang

Zan Zhou

Jing Jing Liu

Guang Zhou Zhang

Fu Fei Xu

Lei Qiu

Lei He

Fileno A Alleva

Jingyang Xu

Max Leung

Yulin Li

Chiwei Che

Yun Feng Sun

Binggong Ding

Shifeng Pan

Chunling Ma

Yang Liu

Li Wang

Gen Li

Peng Wang

Peng Li

Jinyu Li

Wei Peng Huang

Wei Wu

Di Wang

Liang Zhang

Ji Li

Yong Zhao

Amit Srivastava

Qi Chen

Zhi Hong Li

Lin Zhu

Zhihua Liu

Yu Shen

Min Chu

Liang Cheng

Yuandong Hu

Shenghua Ye

Dengming Xiao

Kaisheng N Yao

Jian Luan

Shu Kai Chen

Xijie Liu

Bo Yan

Yongji Xie

Liming Wang

Derek Martin Johnson

Honghao Yu

Wen Chai

Pei Zhao

Ramya Tridandapani

Shulong Wang

Xinhe Xu

Lutz Thomas Finger

Haiyu Wang

Gencheng Wu

Konstantin Seleskerov

Bohan Li

Yin Wang

Yanqing Liu

Hai Feng Li

Jiachen Zhang

Muin Barkatali Momin

Huey-Ru Tsai

Linfang Wang

Difei Gao

Yijian Wu

Shang Shun Lei

Jian-Chao Wang

Lu-Jun Yuan

Shenghao Qin

Sindhu Vigasini Jambunathan

Jay A Farrell

Xinfeng Chen

Xiaoqiang Wang

Yiming Chen

Qiming Guan

Tianci Zhang

Jinfeng Yang

Zewen Jin

Rezaur Rahman

DaJiang Tao

Lipeng Zhang

Chonglong Li

Jia Tang

Hairong Xia

Daniela Braga

Novel solutions for speech recognition provide contextual spelling correction (CSC) for automatic speech recognition (ASR). Disclosed examples include receiving an audio stream; performing an ASR process on the audio stream to produce an ASR hypothesis; receiving a context list; and, based on at least the ASR hypothesis and the context list, performing spelling correction to produce an output text sequence. A contextual spelling correction (CSC) model is used on top of an ASR model, precluding the need for changing the original ASR model. This permits run-time user customization based on contextual data, even for large-size context lists. Some examples include filtering ASR hypotheses for the audio stream and, based on at least the ASR hypotheses filtering, determining whether to trigger spelling correction for the ASR hypothesis. Some examples include generating text to speech (TTS) audio using preprocessed transcriptions with context phrases to train the CSC model.

Contextual spelling correction (CSC) for automatic speech recognition (ASR)

Embodiments of the present disclosure provide a method, apparatus, device and storage medium for data processing. If a kernel module sends a data calling request to a userspace process, first modification time information and second modification time information of data corresponding to the data calling request are obtained by the userspace process. The first and second modification time information are used to indicate modification time information of the data in the kernel module and in a file service end, respectively. The first and second modification time information are compared by the userspace process. If the first and second modification time information are inconsistent, a verification invalidation result is returned to the kernel module. The data in the kernel module is invalidated by the kernel module according to the verification invalidation result. The data in the file service is synchronized by the userspace process to the kernel module.

Method, apparatus, device and storage medium for data processing

Automatic generation of intelligent content is created using a system of computers including a user device and a cloud-based component that processes the user information. The system performs a process that includes receiving an input document and parsing the input document to generate inputs for a natural language generation model using a text analysis model. The natural language generation model generates one or more candidate presentation scripts based on the inputs. A presentation script is selected from the candidate presentation scripts and displayed. A text-to-speech model may be used to generate a synthesized audio presentation of the presentation script. A final presentation may be generated that includes a visual display of the input document and the corresponding audio presentation in sync with the visual display.

Automated script generation and audio-visual presentations

Systems are configured for generating spectrogram data characterized by a voice timbre of a target speaker and a prosody style of source speaker by converting a waveform of source speaker data to phonetic posterior gram (PPG) data, extracting additional prosody features from the source speaker data, and generating a spectrogram based on the PPG data and the extracted prosody features. The systems are configured to utilize/train a machine learning model for generating spectrogram data and for training a neural text-to-speech model with the generated spectrogram data.

System and method for cross-speaker style transfer in text-to-speech and training data generation

Embodiments relate to an eigenvalue-based data query. An aspect includes receiving a query request that includes a query statement. Another aspect includes calculating eigenvalues of key component elements in the query statement. Another aspect includes matching eigenvalues of nodes in an execution plan of a historical query statement to the eigenvalues of the key component elements. Yet another aspect includes based on determining success of matching the eigenvalues of the key component elements to the eigenvalues of the nodes in an execution plan of the historical query statement, generating an execution plan of the query statement.

Eigenvalue-based data query

Various embodiments each include systems, methods, devices, or software for integer ambiguity resolution approach over a time window of GNSS/IMU data. One purpose of processing a window of data is to enhance the reliability of obtaining high-accuracy position estimation, using carrier-phase measurements, even in challenging environments.

Multiple epoch GNSS carrier phase integer resolution

Methods, systems, and computer programs are presented for recommending new connections based on profile similarity and existing interconnections within a social network. One method includes an operation for detecting a request for new connections for a member of the social network, where the profile of the member includes values for certain attributes. Additionally, the method includes operations for identifying members that have at least one equal attribute to the attributes of the member, and for calculating a connection score for each identified member based on the respective values of the identified members attributes. Members are selected from the identified members based on the connection scores, and a ranking score for each selected member is obtained utilizing a machine learning algorithm that utilizes similarity analysis of the attributes to calculate the ranking score. The selected members are presented to the member as the possible new connections based on the ranking scores.

Finding members with similar data attributes of a user for recommending new social connections

Embodiments of a system and method for controlling access to personal information are generally described herein. In an example, a method may include receiving, at an internal server, data including personal information from a third party website, the personal information including a member identification to a social networking service, adding a time to live envelope to the data, receiving a request for the data, determining whether the time to live envelope indicates that the data is accessible, in response to determining that the time to live envelope indicates that the data is accessible, releasing the data according to the request, and aggregating the data with other information.

Controlling access to data originating from a third party website

The technology relates to converting text to speech utilizing recurrent neural networks (RNNs). The recurrent neural networks may be implemented as multiple modules for determining properties of the text. In embodiments, a part-of-speech RNN module, letter-to-sound RNN module, a linguistic prosody tagger RNN module, and a context awareness and semantic mining RNN module may all be utilized. The properties from the RNN modules are processed by a hyper-structure RNN module that determine the phonetic properties of the input text based on the outputs of the other RNN modules. The hyper-structure RNN module may generate a generation sequence that is capable of being converting to audible speech by a speech synthesizer. The generation sequence may also be optimized by a global optimization module prior to being synthesized into audible speech.

Hyper-structure recurrent neural networks for text-to-speech

An index for large databases is disclosed. Data is grouped into clusters and the clusters are grouped into levels of detail. Analysis results are determined based on progressive data sampling. Sampling is conducted based on the level of detail required and/or the resources (time or computing resources) that are available. Larger, more concentrated clusters, at higher levels of detail, are sampled more sparsely. Smaller, more diffuse clusters, at lower levels of detail, are sampled more intensively. Analysis results, including outlier data, include proportional representation from the whole database up to the level of detail required. Results are quickly determined with specified degree of accuracy, based on initial sampling, and are refined with subsequent sampling.

Database index for constructing large scale data level of details

The invention relates to a novel class of 2,4-diamino-6,7-dihydro-5H-pyrrolo[2,3]pyrimidine derivatives as a FAK and/or Pyk2 inhibitor, to a process for their preparation, and to a composition thereof, as well as to use of the compounds for the inhibiting FAK and/or Pyk2 and method for the treatment of a FAK and/or Pyk2 mediated disorder or disease.

2,4-diamino-6,7-dihydro-5H-pyrrolo[2,3]pyrimidine derivatives as FAK/Pyk2 inhibitors

Dynamic features are utilized with CRFs to handle long-distance dependencies of output labels. The dynamic features present a probability distribution involved in explicit distance from/to a special output label that is pre-defined according to each application scenario. Besides the number of units in the segment (from the previous special output label to the current unit), the dynamic features may also include the sum of any basic features of units in the segment. Since the added dynamic features are involved in the distance from the previous specific label, the searching lattice associated with Viterbi searching is expanded to distinguish the nodes with various distances. The dynamic features may be used in a variety of different applications, such as Natural Language Processing, Text-To-Speech and Automatic Speech Recognition. For example, the dynamic features may be used to assist in prosodic break and pause prediction.

Dynamic long-distance dependency with conditional random fields

Growing community of inventors

Beijing, China

Sheng Zhao