公开(公告)号：WO2014187069A1

公开(公告)日：2014-11-27

申请号：PCT/CN2013/085347

申请日：2013-10-16

Applicant: TENCENT TECHNOLOGY (SHENZHEN) COMPANY LIMITED

Inventor： LIU, Haibo ,  WANG, Eryu ,  ZHANG, Xiang ,  YUE, Shuai ,  LI, Lu ,  LU, Li ,  LIU, Jian ,  CHEN, Bo

IPC: G10L15/26 ,  G10L15/02 ,  G10L15/14 ,  G10L15/18 ,  G06F17/30

CPC classification number: G10L15/1815 ,  G06F17/27 ,  G06F17/2725 ,  G10L15/04 ,  G10L15/183 ,  G10L15/187 ,  G10L15/26 ,  G10L15/265

Abstract: Systems and methods are provided for adding punctuations. For example, one or more first feature units are identified in a voice file taken as a whole; the voice file is divided into multiple segments; one or more second feature units are identified in the voice file; a first aggregate weight of first punctuation states of the voice file and a second aggregate weight of second punctuation states of the voice file are determined, using a language model established based on word separation and third semantic features; a weighted calculation is performed to generate a third aggregate weight based on at least information associated with the first aggregate weight and the second aggregate weight; and one or more final punctuations are added to the voice file based on at least information associated with the third aggregate weight.

Abstract translation: 提供了系统和方法来添加标点符号。 例如，一个或多个第一特征单元在作为整体而言的语音文件中被识别; 语音文件分为多个段; 在语音文件中识别一个或多个第二特征单元; 使用基于词分离和第三语义特征建立的语言模型来确定语音文件的第一标点状态的第一聚合权重和语音文件的第二标点状态的第二聚合权重; 基于至少与第一聚集权重和第二聚集权重相关联的信息来执行加权计算以产生第三聚集权重; 并且基于至少与第三聚合权重相关联的信息将一个或多个最终标点符号添加到语音文件。

公开(公告)号：WO2014117577A1

公开(公告)日：2014-08-07

申请号：PCT/CN2013/087816

申请日：2013-11-26

Applicant: TENCENT TECHNOLOGY (SHENZHEN) COMPANY LIMITED

Inventor： YUE, Shuai ,  LU, Li ,  ZHANG, Xiang ,  XIE, Dadong ,  CHEN, Bo ,  RAO, Feng

IPC: G10L15/193 ,  G10L15/28 ,  G10L15/00

CPC classification number: G10L15/193 ,  G10L15/083

Abstract: A method and system for automatic speech recognition is provided. The method includes generating a decoding network that includes a primary sub-network and a classification sub-network. The primary sub-network includes a classification node corresponding to the classification sub-network. The classification sub-network corresponds to a group of uncommon words. Speech input is received and decoded by instantiating a token in the primary sub-network and passing the token through the primary network. When the token reaches the classification node, the method includes transferring the token to the classification sub-network and passing the token through the classification sub-network. When the token reaches an accept node of the classification sub-network, the method includes returning a result of the token passing through the classification sub-network to the primary sub-network. The result includes one or more words in the group of uncommon words. A string corresponding to the speech input is output that includes the one or more words.

Abstract translation: 提供了一种自动语音识别的方法和系统。 该方法包括生成包括主子网和分类子网的解码网络。 主子网包括与分类子网对应的分类节点。 分类子网对应于一组不常见的单词。 通过在主子网中实例化令牌并传递令牌通过主网络来接收和解码语音输入。 当令牌到达分类节点时，该方法包括将令牌传送到分类子网，并通过分类子网传递令牌。 当令牌到达分类子网络的接受节点时，该方法包括将通过分类子网络的令牌的结果返回到主子网络。 结果包括不常见词组中的一个或多个单词。 输出与语音输入对应的字符串，其中包含一个或多个单词。

公开(公告)号：WO2014117549A1

公开(公告)日：2014-08-07

申请号：PCT/CN2013/086152

申请日：2013-10-29

Applicant: TENCENT TECHNOLOGY (SHENZHEN) COMPANY LIMITED

Inventor： LI, Lou ,  CHENG, Qiang ,  RAO, Feng ,  LU, Li ,  ZHANG, Xiang ,  YUE, Shuai ,  CHEN, Bo ,  LU, Duling

IPC: G06F17/27

CPC classification number: G06F17/273 ,  G06F17/2775

Abstract: A computer-implemented method is performed at a device having one or more processors and memory storing programs executed by the one or more processors. The method comprises: selecting a target word in a target sentence; from the target sentence, acquiring a first sequence of words that precede the target word and a second sequence of words that succeed the target word; from a sentence database, searching and acquiring a group of words, each of which separates the first sequence of words from the second sequence of words in a sentence; creating a candidate sentence for each of the candidate words by replacing the target word in the target sentence with each of the candidate words; determining the fittest sentence among the candidate sentences according to a linguistic model; and suggesting the candidate word within the fittest sentence as a correction.

Abstract translation: 在具有一个或多个处理器的设备和由一个或多个处理器执行的存储器存储程序的设备上执行计算机实现的方法。 该方法包括：在目标句子中选择目标词; 从所述目标句子中获取所述目标词之前的第一个单词序列以及继续所述目标单词的第二个单词序列; 从句子数据库中搜索并获取一组单词，每一个单词都将第一个单词序列与一个句子中的第二个单词序列进行分隔; 通过用目标句子中的每个候选词替换目标词来为每个候选词创建候选句子; 根据语言模型确定候选句子中最适合的句子; 并建议适者生词中的候选词作为校正。

公开(公告)号：WO2015024431A1

公开(公告)日：2015-02-26

申请号：PCT/CN2014/083531

申请日：2014-08-01

Applicant: TENCENT TECHNOLOGY (SHENZHEN) COMPANY LIMITED

Inventor： MA, Jianxiong ,  LI, Lu ,  LU, Li ,  ZHANG, Xiang ,  YUE, Shuai ,  RAO, Feng ,  WANG, Eryu ,  KONG, Linghui

IPC: G10L15/08

CPC classification number: G10L15/18 ,  G10L15/08 ,  G10L15/28 ,  G10L15/32 ,  G10L2015/088

Abstract: A method and an apparatus are provided for retrieving keyword. The apparatus configures at least two types of language models in a model file, where each type of language model includes a recognition model and a corresponding decoding model; the apparatus extracts a speech feature from the to-be-processed speech data; performs language matching on the extracted speech feature by using recognition models in the model file one by one, and determines a recognition model based on a language matching rate; and determines a decoding model corresponding to the recognition model; decoding the extracted speech feature by using the determined decoding model, and obtains a word recognition result after the decoding; and matches a keyword in a keyword dictionary and the word recognition result, and outputs a matched keyword.

Abstract translation: 提供了一种用于检索关键字的方法和装置。 该装置在模型文件中配置至少两种类型的语言模型，其中每种类型的语言模型包括识别模型和相应的解码模型; 该设备从待处理语音数据中提取语音特征; 通过在模型文件中逐一使用识别模型对提取出的语音特征进行语言匹配，并根据语言匹配率确定识别模型; 并确定与识别模型相对应的解码模型; 通过使用所确定的解码模型来解码所提取的语音特征，并且在解码之后获得字识别结果; 并且将关键词字典中的关键词与单词识别结果进行匹配，并输出匹配关键字。

公开(公告)号：WO2015003436A1

公开(公告)日：2015-01-15

申请号：PCT/CN2013/085568

申请日：2013-10-21

Applicant: TENCENT TECHNOLOGY (SHENZHEN) COMPANY LIMITED

Inventor： WANG, Eryu ,  LU, Li ,  ZHANG, Xiang ,  LIU, Haibo ,  RAO, Feng ,  LI, Lou ,  YUE, Shuai ,  CHEN, Bo

IPC: G06N3/08

CPC classification number: G06N3/08

Abstract: A method and a device for training a DNN model includes: at a device includes one or more processors and memory: establishing an initial DNN model; dividing a training data corpus into a plurality of disjoint data subsets; for each of the plurality of disjoint data subsets, providing the data subset to a respective training processing unit of a plurality of training processing units operating in parallel, wherein the respective training processing unit applies a Stochastic Gradient Descent (SGD) process to update the initial DNN model to generate a respective DNN sub-model based on the data subset; and merging the respective DNN sub- models generated by the plurality of training processing units to obtain an intermediate DNN model, wherein the intermediate DNN model is established as either the initial DNN model for a next training iteration or a final DNN model in accordance with a preset convergence condition.

Abstract translation: 用于训练DNN模型的方法和设备包括：在设备上包括一个或多个处理器和存储器：建立初始DNN模型; 将训练数据语料库划分为多个不相交的数据子集; 对于多个不相交数据子集中的每一个，将数据子集提供给并行操作的多个训练处理单元的相应训练处理单元，其中各训练处理单元应用随机梯度下降（SGD）过程来更新初始 DNN模型基于数据子集生成相应的DNN子模型; 并且合并由多个训练处理单元生成的各个DNN子模型，以获得中间DNN模型，其中，中间DNN模型被建立为用于下一个训练迭代的初始DNN模型或根据下一个训练迭代的最终DNN模型 预设收敛条件。

公开(公告)号：WO2014183373A1

公开(公告)日：2014-11-20

申请号：PCT/CN2013/085155

申请日：2013-10-14

Applicant: TENCENT TECHNOLOGY (SHENZHEN) COMPANY LIMITED

Inventor： LI, Lou ,  LU, Li ,  ZHANG, Xiang ,  RAO, Feng ,  YUE, Shuai ,  CHEN, Bo ,  MA, Jianxiong ,  LIU, Haibo

IPC: G10L15/18

CPC classification number: G10L15/083 ,  G10L15/1815 ,  G10L15/183

Abstract: Systems and methods are provided for voice identification. For example, audio characteristics are extracted from acquired voice signals; a syllable confusion network is identified based on at least information associated with the audio characteristics; a word lattice is generated based on at least information associated with the syllable confusion network and a predetermined phonetic dictionary; and an optimal character sequence is calculated in the word lattice as an identification result.

Abstract translation: 为语音识别提供了系统和方法。 例如，从获取的语音信号中提取音频特性; 至少基于与音频特征相关联的信息来识别音节混淆网络; 基于至少与音节混淆网络和预定语音字典相关联的信息生成单词格点; 并且在字格中计算最佳字符序列作为识别结果。

公开(公告)号：WO2014117583A1

公开(公告)日：2014-08-07

申请号：PCT/CN2013/087994

申请日：2013-11-28

Applicant: TENCENT TECHNOLOGY (SHENZHEN) COMPANY LIMITED

Inventor： ZHANG, Xiang ,  LU, Li ,  WANG, Eryu ,  YUE, Shuai ,  RAO, Feng ,  LIU, Haibo ,  LI, Lou ,  LU, Duling ,  CHEN, Bo

IPC: H04L9/32

CPC classification number: G06F21/32 ,  G06F2221/2117

Abstract: A computer-implemented method is performed at a server having one or more processors and memory storing programs executed by the one or more processors for authenticating a user from video and audio data. The method includes: receiving a login request from a mobile device, the login request including video data and audio data; extracting a group of facial features from the video data; extracting a group of audio features from the audio data and recognizing a sequence of words in the audio data; identifying a first user account whose respective facial features match the group of facial features and a second user account whose respective audio features match the group of audio features. If the first user account is the same as the second user account, retrieve the sequence of words associated with the user account and compare the sequences of words for authentication purpose.

Abstract translation: 在具有一个或多个处理器的服务器和由一个或多个处理器执行的用于从视频和音频数据认证用户的存储器存储程序的服务器执行计算机实现的方法。 该方法包括：从移动设备接收登录请求，登录请求包括视频数据和音频数据; 从视频数据中提取一组面部特征; 从音频数据提取一组音频特征并识别音频数据中的单词序列; 识别其各自的面部特征与该组面部特征相匹配的第一用户帐户和其各个音频特征与该组音频特征相匹配的第二用户帐户。 如果第一个用户帐户与第二个用户帐户相同，则检索与用户帐户相关联的单词序列，并比较用于验证目的的单词序列。

公开(公告)号：WO2014117547A1

公开(公告)日：2014-08-07

申请号：PCT/CN2013/085905

申请日：2013-10-24

Applicant: TENCENT TECHNOLOGY (SHENZHEN) COMPANY LIMITED

Inventor： LU, Li ,  ZHANG, Xiang ,  YUE, Shuai ,  RAO, Feng ,  WANG, Eryu ,  LI, Lu

IPC: G10L15/08

CPC classification number: G10L15/063 ,  G10L15/08 ,  G10L2015/088

Abstract: An electronic device with one or more processors and memory trains an acoustic model with an international phonetic alphabet (IPA) phoneme mapping collection and audio samples in different languages, where the acoustic model includes: a foreground model; and a background model. The device generates a phone decoder based on the trained acoustic model. The device collects keyword audio samples, decodes the keyword audio samples with the phone decoder to generate phoneme sequence candidates, and selects a keyword phoneme sequence from the phoneme sequence candidates. After obtaining the keyword phoneme sequence, the device detects one or more keywords in an input audio signal with the trained acoustic model, including: matching phonemic keyword portions of the input audio signal with phonemes in the keyword phoneme sequence with the foreground model; and filtering out phonemic non-keyword portions of the input audio signal with the background model.

Abstract translation: 具有一个或多个处理器和存储器的电子设备具有使用不同语言的国际语音字母（IPA）音素映射收集和音频样本的声学模型，其中声学模型包括：前景模型; 和背景模型。 该设备基于经过训练的声学模型生成电话解码器。 设备收集关键字音频样本，用手机解码器对关键词音频样本进行解码，以产生音素序列候选，并从音素序列候选中选择关键词音素序列。 在获得关键字音素序列之后，设备利用经训练的声学模型检测输入音频信号中的一个或多个关键词，包括：使用前景模型将关键词音素序列中的输入音频信号的音素关键字部分与音素相匹配; 并用背景模型滤出输入音频信号的音素非关键字部分。

公开(公告)号：WO2014117544A1

公开(公告)日：2014-08-07

申请号：PCT/CN2013/085738

申请日：2013-11-21

Applicant: TENCENT TECHNOLOGY (SHENZHEN) COMPANY LIMITED

Inventor： YUE, Shuai ,  LU, Li ,  ZHANG, Xiang ,  XIE, Dadong ,  LIU, Haibo ,  CHEN, Bo ,  LIU, Jian

IPC: G10L15/20

CPC classification number: G10L15/14 ,  G10L15/063 ,  G10L15/083 ,  G10L15/32 ,  G10L2015/088 ,  G10L2015/223

Abstract: A method of recognizing speech commands includes generating a background acoustic model for a sound using a first sound sample, the background acoustic model characterized by a first precision metric. A foreground acoustic model is generated for the sound using a second sound sample, the foreground acoustic model characterized by a second precision metric. A third sound sample is received and decoded by assigning a weight to the third sound sample corresponding to a probability that the sound sample originated in a foreground using the foreground acoustic model and the background acoustic model. The method further includes determining if the weight meets predefined criteria for assigning the third sound sample to the foreground and, when the weight meets the predefined criteria, interpreting the third sound sample as a portion of a speech command. Otherwise, recognition of the third sound sample as a portion of a speech command is forgone.

Abstract translation: 识别语音命令的方法包括使用第一声音样本产生用于声音的背景声学模型，所述背景声学模型由第一精度度量表征。 使用第二声音样本为声音生成前景声学模型，前景声学模型以第二精度度量为特征。 通过使用前景声学模型和背景声学模型通过对与声音样本始发于前景的概率相对应的第三声音样本分配权重来接收和解码第三声音样本。 该方法还包括确定权重是否满足用于将第三声音样本分配给前景的预定准则，并且当权重满足预定标准时，将第三声音样本解释为语音命令的一部分。 否则，放弃了作为语音命令的一部分的第三声音样本的识别。

公开(公告)号：WO2015021844A1

公开(公告)日：2015-02-19

申请号：PCT/CN2014/082332

申请日：2014-07-16

Applicant: TENCENT TECHNOLOGY (SHENZHEN) COMPANY LIMITED

Inventor： LI, Lu ,  LU, Li ,  MA, Jianxiong ,  KONG, Linghui ,  RAO, Feng ,  YUE, Shuai ,  ZHANG, Xiang ,  LIU, Haibo ,  WANG, Eryu ,  CHEN, Bo

IPC: G10L15/08

CPC classification number: G10L15/08 ,  G10L15/083 ,  G10L2015/088

Abstract: Disclosed is a method implemented of recognizing a keyword in a speech that includes a sequence of audio frames further including a current frame and a subsequent frame. A candidate keyword is determined for the current frame using a decoding network that includes keywords and filler words of multiple languages, and used to determine a confidence score for the audio frame sequence. A word option is also determined for the subsequent frame based on the decoding network, and when the candidate keyword and the word option are associated with two distinct types of languages, the confidence score of the audio frame sequence is updated at least based on a penalty factor associated with the two distinct types of languages. The audio frame sequence is then determined to include both the candidate keyword and the word option by evaluating the updated confidence score according to a keyword determination criterion.

Abstract translation: 公开了一种在语音中识别关键字的方法，该方法包括进一步包括当前帧和后续帧的音频帧序列。 使用包括多种语言的关键词和填充词的解码网络为当前帧确定候选关键字，并且用于确定音频帧序列的置信度分数。 还基于解码网络为后续帧确定字选项，并且当候选关键词和词选项与两种不同类型的语言相关联时，至少基于惩罚来更新音频帧序列的置信度得分 与两种不同类型语言相关联的因素。 然后通过根据关键字确定标准评估更新的可信度得分，确定音频帧序列以包括候选关键词和词选项。