专利快速检索-快速检索全球专利，免费商用专利数据库-IPRDB

1. 发明申请

US20160322042A1 FAST DEEP NEURAL NETWORK FEATURE TRANSFORMATION VIA OPTIMIZED MEMORY BANDWIDTH UTILIZATION 有权
标题翻译：通过优化的记忆带宽利用快速深入神经网络特征转换
公开(公告)号：US20160322042A1
公开(公告)日：2016-11-03
申请号：US14699778
申请日：2015-04-29
申请人： Nuance Communications, Inc.
发明人： Jan Vlietinck , Stephan Kanthak , Rudi Vuerinckx , Christophe Ris
IPC分类号： G10L15/06 , G06N3/08
CPC分类号： G06N3/08 , G06N3/0454 , G10L15/02 , G10L15/16 , G10L2015/0635
摘要： Deep Neural Networks (DNNs) with many hidden layers and many units per layer are very flexible models with a very large number of parameters. As such, DNNs are challenging to optimize. To achieve real-time computation, embodiments disclosed herein enable fast DNN feature transformation via optimized memory bandwidth utilization. To optimize memory bandwidth utilization, a rate of accessing memory may be reduced based on a batch setting. A memory, corresponding to a selected given output neuron of a current layer of the DNN, may be updated with an incremental output value computed for the selected given output neuron as a function of input values of a selected few non-zero input neurons of a previous layer of the DNN in combination with weights between the selected few non-zero input neurons and the selected given output neuron, wherein a number of the selected few corresponds to the batch setting.
摘要翻译：具有许多隐藏层和每层许多单元的深层神经网络（DNN）是具有非常大数量参数的非常灵活的模型。因此，DNN具有挑战性来进行优化。为了实现实时计算，本文公开的实施例通过优化的存储器带宽利用率实现快速DNN特征变换。为了优化存储器带宽利用率，可以基于批量设置来减少访问存储器的速率。对应于DNN的当前层的所选给定输出神经元的存储器可以用针对所选择的给定输出神经元计算的增量输出值作为所选择的几个非零输入神经元的输入值的函数来更新结合所选择的几个非零输入神经元和所选择的给定输出神经元之间的权重组合的DNN的先前层，其中所选择的几个的数量对应于批次设置。

2. 发明授权

US10013652B2 Fast deep neural network feature transformation via optimized memory bandwidth utilization 有权
公开(公告)号：US10013652B2
公开(公告)日：2018-07-03
申请号：US14699778
申请日：2015-04-29
申请人： Nuance Communications, Inc.
发明人： Jan Vlietinck , Stephan Kanthak , Rudi Vuerinckx , Christophe Ris
IPC分类号： G10L15/00 , G06N3/08 , G06N3/04 , G10L15/16 , G10L15/06 , G10L15/02
CPC分类号： G06N3/08 , G06N3/0454 , G10L15/02 , G10L15/16 , G10L2015/0635
摘要： Deep Neural Networks (DNNs) with many hidden layers and many units per layer are very flexible models with a very large number of parameters. As such, DNNs are challenging to optimize. To achieve real-time computation, embodiments disclosed herein enable fast DNN feature transformation via optimized memory bandwidth utilization. To optimize memory bandwidth utilization, a rate of accessing memory may be reduced based on a batch setting. A memory, corresponding to a selected given output neuron of a current layer of the DNN, may be updated with an incremental output value computed for the selected given output neuron as a function of input values of a selected few non-zero input neurons of a previous layer of the DNN in combination with weights between the selected few non-zero input neurons and the selected given output neuron, wherein a number of the selected few corresponds to the batch setting.

3. 发明授权

US11367437B2 Multi-microphone speech dialog system for multiple spatial zones 有权
公开(公告)号：US11367437B2
公开(公告)日：2022-06-21
申请号：US16426356
申请日：2019-05-30
申请人： NUANCE COMMUNICATIONS, INC.
发明人： Timo Matheja , Markus Buck , Andreas Kirbach , Martin Roessler , Tim Haulick , Julien Premont , Josef Anastasiadis , Rudi Vuerinckx , Christophe Ris , Stijn Verschaeren , Hakan Ari , Dieter Ranz
IPC分类号： G10L15/20 , G10L17/00 , G10L15/22 , H04R1/40 , G10L21/02
摘要： There is provided a speech dialog system that includes a first microphone, a second microphone, a processor and a memory. The first microphone captures first audio from a first spatial zone, and produces a first audio signal. The second microphone captures second audio from a second spatial zone, and produces a second audio signal. The processor receives the first audio signal and the second audio signal, and the memory contains instructions that control the processor to perform operations of a speech enhancement module, an automatic speech recognition module, and a speech dialog module that performs a zone-dedicated speech dialog.

4. 发明授权

US10229686B2 Methods and apparatus for speech segmentation using multiple metadata 有权
公开(公告)号：US10229686B2
公开(公告)日：2019-03-12
申请号：US15329354
申请日：2014-08-18
申请人： NUANCE COMMUNICATIONS, INC.
发明人： Markus Buck , Tobias Herbig , Simon Graf , Christophe Ris
IPC分类号： G10L15/28 , G10L15/22 , G10L25/78 , G10L15/04 , G10L15/20 , G10L21/02
摘要： Methods and apparatus to process microphone signals by a speech enhancement module to generate an audio stream signal including first and second metadata for use by a speech recognition module. In an embodiment, speech recognition is performed using endpointing information including transitioning from a silence state to a maybe speech state, in which data is buffered, based on the first metadata and transitioning to a speech state, in which speech recognition is performed, based upon the second metadata.

你已经成功收藏专利！

检索式保存成功!

IPRDB

热门服务

关于我们

友情链接

联系方式