BROWSE
Detail
Conference Paper
A Robust Endpoint Detection Algorithm for the Speech Recognition in Noisy Environments
Cited - time in 
Share 
Share
- Authors
- Kiyoung Park
- Issue Date
- 2013-09
- Citation
- International Congress and Exposition on Noise Control Engineering (Inter-Noise) 2013, pp.1-6
- Language
- English
- Type
- Conference Paper
- Abstract
- A method to detect voice segments in audio signals recorded in noisy environments is proposed in this paper. The endpoint detection is one of the most important part of speech recognition. Despite of many studies, it is still challenging especially for the mobile applications since they are used in very diverse conditions. The hierarchical endpoint detection algorithm which was proposed is further enhanced and tested for the real-world data which are collected using the mobile speech recognition application opened to the public. Many cases where the conventional algorithm fails to work are analyzed and categorized in several classes. A detector is designed for each class and augmented to the hierarchical detection algorithm. Experiments are performed to measure the detection accuracy and the speech recognition accuracy for the real-world database collected. The users uttered short sentences for writing e-mails, twitters and short messages in diverse environments. The voice segments are detected and then recognized by the large vocabulary continuous speech recognizer in real time. The results show that the proposed method works well in harmony with other detectors and the overall performance is improved greatly in the aspect of the detection accuracy as well as the speech recognition accuracy.
- KSP Keywords
- Audio signal, Continuous Speech, Detection accuracy, Hierarchical detection, Large vocabulary, Mobile Application(APP), Mobile speech recognition, Overall performance, Part of Speech(POS), Real-time, Real-world data