Faculty Profile

1. Continuous voice recognition method based on deep long-and-short-term memory recurrent neural network. ZL2014108216466. PCT/CN2015/092380.

2. Construction method for deep long short-term memory recurrent neural network acoustic model based on selective attention principle. ZL2015101229826. WO2016145850-A1.

3. An acoustic model modeling method of residual long and short-term memory cyclic neural network. CN20171043247.

4. A Method and system for command word recognition using mutual a priori constraint model. CN2020105931541.

5. Robustness sound source space positioning method of distributed microphone array network. ZL201510808753.X.

6. Audio detection classification method with a custom function. WO2015124006-A1.  PCT/CN2014/091959. 

7. Gauss mix model parameter based scene audio index optimizing method, involves sensing input signal by using sensor, and arranging voice frequency section with classifier model corresponding to information classification. ZL201410209057.2. 

8. Speaker identifying method based on sparse dimension reduction, involves treating acoustic characteristics of speech signal recognition shadow sparse matrix according to judgment of classifier training. ZL201310298757.9.

9. Partial learning based speaker recognition method, involves dividing training data into multi-class training set, extracting characteristic space features of each training set, and calculating average value of training data. ZL201210548563.5. 

10. Audio indexing method based on multi-distance sound sensor. ZL201110303580.8. 

11. Method for orienting sound source space of distributed asynchronous acoustic sensor, involves calculating space position coordinate of each sound source according to time delay estimation values. ZL201010568360.3. 

12. Speaker clustering method of distributed microphone, involves calculating time delay of sound source signal segment according to index value of certain frame of numbered sub-band. ZL201010568386.8. 

13. Pitch post processing method involves performing filter process to pitch synthesized signal by obtaining and combining gain control values of local adjustment factors to acquire filer function, for performing pitch post process. WO2009140896A1. PCT/CN2009/071708. 

14. Voice enhancing method, involves obtaining enhanced frequency domain noised voice signal according to prior signal-to-noise ratio of present frame in frequency domain clean voice signal based on least-mean-square-error criterion. ZL200910132345.1.

15. Voice signal enhancement method, involves weighing corresponding voice signals by weighing coefficient of voice signals to acquire weighed multiple voice signals that are synthesized to acquire enhanced voice signal. ZL200810198772.5. 

1. NSFC: Speaker diarization based on distributed acoustic sensor networks and quantum optimization learning technology research. Project chief.
2. NSFC: Multiplayer multiparty dialogue speech separation, content analysis and understanding. Participated.
3. 863: Humane voice interactive technology research. Participated.
4. Beijing Natural Science Foundation Program: The complex scenes efficient voice enhancement technology. Participated.
5. ITU Standard: As the technology backbone to achieve China's first speech in the ITU ISO codec technology patents embedded success, won the gold medal team Huawei Medal. Participated.
6. AVS Standard: Responsible for domestic AVS codec technology standard technology embedded promote AVS audio application GB. One patent has been embedded to national standards. Group leader.

No content

No content