基本信息:
- 专利标题: 성문 특징 모델 갱신 방법 및 단말
- 专利标题(英):Method for updating voiceprint feature model and terminal
- 专利标题(中):用于更新VOICEPRINT特征模型和终端的方法
- 申请号:KR1020147029482 申请日:2013-07-08
- 公开(公告)号:KR1020140144233A 公开(公告)日:2014-12-18
- 发明人: 루팅
- 申请人: 후아웨이 디바이스 컴퍼니 리미티드
- 申请人地址: Building B*, Huawei Industrial Base, Bantian, Longgang District, Shenzhen, P.R. China
- 专利权人: 후아웨이 디바이스 컴퍼니 리미티드
- 当前专利权人: 후아웨이 디바이스 컴퍼니 리미티드
- 当前专利权人地址: Building B*, Huawei Industrial Base, Bantian, Longgang District, Shenzhen, P.R. China
- 代理人: 유미특허법인
- 优先权: CN2012102355930 2012-07-09
- 国际申请: PCT/CN2013/079005 2013-07-08
- 国际公布: WO2014008843 2014-01-16
- 主分类号: G10L15/02
- IPC分类号: G10L15/02 ; G10L15/07
The present invention is applicable to a speech recognition technology, and provides a gate characteristic model updating method and a terminal. Features gates model updating method comprising the steps of: obtaining the original audio stream comprising at least one speaker; According to the preset speaker segmentation and clustering algorithm comprising: obtaining each of the audio streams of each of the speakers of at least one speaker in the original audio stream; A step of matching each of the audio stream and the original model of the gate, characterized in that of the at least one speaker of each speaker individually to obtain a successful mating in the audio stream; And a step of using the audio stream as the audio stream matched to the successful learning samples added for generating the original model of the gate feature, and updating the original gate characteristic model. In the present invention, is used valid audio stream is extracted adaptively as a further audio stream learning sample of the call, the dynamic correction to the original gate, characterized in module, and carries a relatively high accuracy of the written feature model on the premise of practical by and achieve the purpose to increase the recognition accuracy.
信息查询:
EspacenetIPC结构图谱:
G | 物理 |
--G10 | 乐器;声学 |
----G10L | 语言分析或合成;语言识别 |
------G10L15/00 | 语音识别 |
--------G10L15/02 | .语音识别的特征提取;识别单位的选择 |