{"title":"基于支持向量机的语音活动检测","authors":"M. Baig, S. Masud, Mian M. Awais","doi":"10.1109/ISPACS.2006.364896","DOIUrl":null,"url":null,"abstract":"Voice activity detection (VAD) is important for efficient speech coding and accurate automatic speech recognition (ASR). Most of the algorithms proposed in the past, for solving the VAD problem, have been based on some deterministic feature of the speech signal such as zero crossing rate. The speech/non-speech decisions are then taken using suitably chosen thresholds. This paper presents the application of support vector machines (SVM) for classifying the voice activity. The speech signal has been divided into labeled overlapping frames and pattern classification has subsequently been performed by using a supervised learning algorithm. It has been observed that the SVM based solution is computationally efficient and provides around 90% accuracy for speech signals directly recorded using a microphone and an accuracy of over 85% for noisy speech","PeriodicalId":178644,"journal":{"name":"2006 International Symposium on Intelligent Signal Processing and Communications","volume":"67 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2006-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"8","resultStr":"{\"title\":\"Support Vector Machine based Voice Activity Detection\",\"authors\":\"M. Baig, S. Masud, Mian M. Awais\",\"doi\":\"10.1109/ISPACS.2006.364896\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"Voice activity detection (VAD) is important for efficient speech coding and accurate automatic speech recognition (ASR). Most of the algorithms proposed in the past, for solving the VAD problem, have been based on some deterministic feature of the speech signal such as zero crossing rate. The speech/non-speech decisions are then taken using suitably chosen thresholds. This paper presents the application of support vector machines (SVM) for classifying the voice activity. The speech signal has been divided into labeled overlapping frames and pattern classification has subsequently been performed by using a supervised learning algorithm. It has been observed that the SVM based solution is computationally efficient and provides around 90% accuracy for speech signals directly recorded using a microphone and an accuracy of over 85% for noisy speech\",\"PeriodicalId\":178644,\"journal\":{\"name\":\"2006 International Symposium on Intelligent Signal Processing and Communications\",\"volume\":\"67 1\",\"pages\":\"0\"},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2006-12-01\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"8\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"2006 International Symposium on Intelligent Signal Processing and Communications\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/10.1109/ISPACS.2006.364896\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"2006 International Symposium on Intelligent Signal Processing and Communications","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/ISPACS.2006.364896","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
Support Vector Machine based Voice Activity Detection
Voice activity detection (VAD) is important for efficient speech coding and accurate automatic speech recognition (ASR). Most of the algorithms proposed in the past, for solving the VAD problem, have been based on some deterministic feature of the speech signal such as zero crossing rate. The speech/non-speech decisions are then taken using suitably chosen thresholds. This paper presents the application of support vector machines (SVM) for classifying the voice activity. The speech signal has been divided into labeled overlapping frames and pattern classification has subsequently been performed by using a supervised learning algorithm. It has been observed that the SVM based solution is computationally efficient and provides around 90% accuracy for speech signals directly recorded using a microphone and an accuracy of over 85% for noisy speech