APPROACHES TO LANGUAGE IDENTIFICATION USING FEATURE-LEVEL AND DECISION-LEVEL FUSION

摘要：

The PRLM (phone recognizer followed by language model) system has the best performance on NIST language recognition evaluation tasks. However, its orthographical or phonetic transcription and heavy computational demands may preclude their use in low cost, realistic applications. An alternative approach to LID uses Gaussian mixture models (GMMs) to classify languages using the acoustic content of the speech signal. In this paper, we propose the feature-level and decision-level multiple features fusion to the GMM-LM (GMM recognizer followed by language model) system for a higher level of performance. The experiments show that decision-level fusion produces the best accuracy 83.6% which is 16.3% and 52% relative improvement over feature-level fusion and energy result.

关键词： Feature-level fusion Decision-level fusion GMM-LM

作者: XIUHUA ZENG JIAN YANG DAN XU

作者单位: School of Information Science and Engineering, Yunnan University, Kunming, China

会议类型: 国际会议

会议名称: The Second International Conference on Information & Systems Sciences(ICISS2008)(第二届信息与系统科学国际会议)

会议地点: 大连

会议语种:英文

页码: 795-801

在线出版日期: 2008-12-18（万方平台首次上网日期，不代表论文的发表时间）