Loading...
Please wait, while we are loading the content...
A Novel Front-end Based on Variable Frame Rate Analysis and Mel-filterbank Output Compensation for Robust ASR
| Content Provider | Semantic Scholar |
|---|---|
| Author | Choi, Eric H. C. Epps, Julien |
| Copyright Year | 2006 |
| Abstract | For automatic speech recognition (ASR) systems, robustness in the presence of various types and levels of environmental noise remains an important issue, despite the various advances of recent years. This paper describes a new noise-robust ASR front-end employing a combination of variable frame rate processing based on the sample-by-sample delta energy parameter, Melfilterbank output compensation and cumulative distribution mapping. Recognition experiments on the Aurora II connected digits database reveal that the proposed front-end achieves an average digit recognition accuracy of 84.3% for a model set trained from clean speech data. Compared with the ETSI standard Mel-cepstral front-end, the proposed front-end is found to obtain a relative error rate reduction of around 60%. Moreover, the proposed front-end can provide almost comparable recognition accuracy with the ETSI advanced front-end, at roughly half the computational complexity. |
| File Format | PDF HTM / HTML |
| Alternate Webpage(s) | http://www.assta.org/sst/2006/sst2006-19.pdf |
| Language | English |
| Access Restriction | Open |
| Content Type | Text |
| Resource Type | Article |
