Presentation is loading. Please wait.

Presentation is loading. Please wait.

Compensating speaker-to-microphone playback system for robust speech recognition So-Young Jeong and Soo-Young Lee Brain Science Research Center and Department.

Similar presentations


Presentation on theme: "Compensating speaker-to-microphone playback system for robust speech recognition So-Young Jeong and Soo-Young Lee Brain Science Research Center and Department."— Presentation transcript:

1 Compensating speaker-to-microphone playback system for robust speech recognition So-Young Jeong and Soo-Young Lee Brain Science Research Center and Department of Electrical Engineering and Computer Science Korea Advanced Institute of Science and Technology

2  ASR in mismatched environments Environmental information –Background noise, acoustic/transmission channel Assume environment degradation model Motivation Clean speech Channel Additive noise Distorted speech

3 –P.S –F.B. –L.S. –C.S. Channel Impacts on feature Channel Assumption 2 Channel Assumption 1

4  Speaker-to-Microphone playback  Speaker distortion Nonlinearity caused by voice coil  Microphone distortion Frequency response caused by different fabrication Nonlinearity caused by dynamic range Ambient noise by directionality Speaker-to-Microphone compensation

5  Mapper train Where and which type of mapper should be deployed?  Mapper apply Speaker-to-Microphone mapping F.E. + clean F.E.Mapper distorted Error F.E.Trained Mapper distorted To recognizer

6  Diamond, plus, cross denotes PS,FB.LS level Mapping error at L.S.

7 Frequency correlation plots

8  Task Phoneme recognition for 40 TIMIT phone sets Phone accuracy = (N-D-S-I) * 100 /N  Database HTIMIT : re-recording TIMIT sentence thru. 10 various telephone handsets Training : 246 speaker * 8 sent. = 1968sent. Test : 48 speaker * 8 = 384 sent.  Baseline 3-state monophone HMM with 16 gaussian mixture Recognition Experiments

9 Experiment I – CI result typematchedmismatchCMSDIAGLINPERMLP senh54.7 cb153.645.850.352.652.252.451.9 cb254.948.352.455.154.854.653.7 cb348.532.338.737.340.638.241.9 cb449.835.840.837.942.942.243.3 el155.445.652.254.053.553.254.1 el253.736.749.151.852.552.652.4 el351.044.644.547.146.947.147.2 el453.743.147.649.449.649.750.1 pt152.641.143.045.246.045.445.9

10  Speech signal distorted by low-quality speaker-to- microphone playback system can be compensated with feature mapping network  Feature mapping scheme would be useful in cases that environmental condition is tough for collecting database Conclusion


Download ppt "Compensating speaker-to-microphone playback system for robust speech recognition So-Young Jeong and Soo-Young Lee Brain Science Research Center and Department."

Similar presentations


Ads by Google