To solve the problems of background music extraction, seperation and robust recognition on the radio,a voice and background music seperation algorithm independent of speaker and relevant with background music types is proposed accor-ding to the types stability of the background music and speakers′ uncertainty in broadcast voice. In order to better seperate the background music and voice signal, a cascade system was designed and implemented, which combines the voice activity detec-tion system with speech and background music separation system in the music background. Aiming at the practical broadcast cor-pus, a method with entire supervision is used to extract the music background. The experimental data show this system can im-prove the recognition ratio of background music and speech to a certain extent.%为了解决广播中背景音乐的提取、分离和鲁棒识别,考虑到广播语音中背景音乐的类型稳定性和说话人的不确定性,提出一种说话人无关背景音乐类型相关的语音和背景音乐分离算法,为了更好地分离音乐背景与语音信号,设计并实现了一个级联系统,即音乐背景下的语音活动检测系统加语音和背景音乐分离系统.针对实际的广播语料,采用一种全监督的方法提取背景音乐,实验数据表明,该系统在一定程度上能够提升背景音乐和语音的识别率.
展开▼
