Name
Title | ||
|---|---|---|
Phone-Aware Multi-task Learning and Length Expanding for Short-Duration Language Recognition |
Abstract | ||
|---|---|---|
In the language recognition, the phonetic information has shown great potential for neural network to learn the high-level representations. In this paper, we explore two significant aspects to improve the system performance on oriental language recognition (OLR) challenge under the short-duration condition. Firstly, we propose to learn the language information and phonetic information jointly with multi-task learning. The classified networks can learn the extra phonetic representation from a frame-level phone-task and extract the language embedding at the segment level. Furthermore, we propose to introduce length expanding strategy to provide supplemental information of short-duration utterances by dithering the short duration evaluation utterances at different speeds. The evaluation results of the 3rd OLR Challenge showed that our proposed methods obtained the best results on the short-duration condition. |
Year | DOI | Venue |
|---|---|---|
2019 | 10.1109/APSIPAASC47483.2019.9023014 | Asia-Pacific Signal and Information Processing Association Annual Summit and Conference |
Keywords | DocType | ISSN |
phonetic information,multi-task learning,length expanding,speed perturbation pooling,short-duration,language recognition | Conference | 2309-9402 |
Citations | PageRank | References |
0 | 0.34 | 0 |
Authors | ||
8 | ||
Name | Order | Citations | PageRank |
|---|---|---|---|
| Miao Zhao | 1 | 0 | 2.70 |
| Rongjin Li | 2 | 1 | 1.35 |
| Shijiang Yan | 3 | 0 | 0.34 |
| Zheng Li | 4 | 0 | 1.35 |
| Hao Lu | 5 | 3 | 3.78 |
| Shipeng Xia | 6 | 0 | 1.01 |
| Q. Y. Hong | 7 | 50 | 15.79 |
| Lin Li | 8 | 323 | 79.92 |