摘要
选用200个拉祜语常用词,以词为识别单元,利用语音识别工具箱HTK开展非特定人拉祜语孤立词的语音识别研究.研究发现:特定发音人的识别率比较稳定,针对非特定发音孤立词,通过增加模型的混合高斯数目,其识别率得到显著提高,使得识别正确率保持在99%以内,为拉祜语孤立词识别提供了有效方法.
This research selects 200 common Luhuwords with each as the recognition unit, and uses voice recognition toolkit HTK for the study of the speech recognition of the isolated words of speaker dependent Luhus. The study reveals that speaker dependent Luhus have a stable recognition rate;as to the speaker independence, it increases the number in the Gaussian mixture model, and its recognition rate is increased significantly with the recognition accuracy within 99% ,which provides an effective method for the speech recognition of the isolated words in the Lahu language.
出处
《云南民族大学学报(自然科学版)》
CAS
2015年第4期337-340,共4页
Journal of Yunnan Minzu University:Natural Sciences Edition
基金
云南省教育厅科学研究基金(2014Z091)
云南省自然科学基金(2011FZ176)
关键词
语音识别
拉祜语
孤立词
非特定人
高斯混合模型
speech recognition
Lahu language
isolated word
speaker independent
Gaussian mixture model