Note: Your custom schedule will not be saved unless you create a new account or login to an existing account.
| Paper: | SP-P16.9 |
| Session: | Speech Modeling for Robust Speech Recognition |
| Time: | Friday, May 21, 15:30 - 17:30 |
| Presentation: |
Poster |
| Topic: |
Speech Processing: Robust Speech Recognition |
| Title: |
CAN BACK-ENDS BE MORE ROBUST THAN FRONT-ENDS? INVESTIGATION OVER THE AURORA-2 DATABASE |
| Authors: |
Alexis Bernard; Texas Instruments, Inc. | | |
| | Yifan Gong; Texas Instruments, Inc. | | |
| | Xiaodong Cui; University of California, Los Angeles | | |
| Abstract: |
We present a back-end solution developed at Texas Instruments for noise robust speechrecognition. The solution consists of three techniques: 1) a joint additive andconvolutive noise compensation (JAC) which adapts speech acoustic models, 2) an enhancedchannel estimation procedure which extends JAC performance towards lower SNR ranges, and3) an N-pass decoding algorithm. The performance the proposed back-end is evaluated onthe Aurora-2 database. With 20% less model parameters and without the need for secondorder derivative of the recognition features, the performance of the proposed solution is91.86%, which outperforms that of the ETSI Advanced Front-End standard (88.19%) by morethan 30% relative word error rate reduction. |
| |
| Back | |