Abstract
The purpose of this paper is to propose an improved FM radio music extraction system we have developed. Usually, the following three processes are implemented to extract music. In the first process, power change is utilized to extract verge parts between music and talk. The second process is fixing the verge parts according to the frequency feature. The final process extracts the music part which has longer interval of fixed verge parts than threshold. However, the above algorithm might divide a single music into multi parts, which decreases the usability. Therefore, we propose an improved algorithm, which infuses the difference between left and right sound data channels as a new feature and decides the end time referring to the continuous talk time. The usability evaluation experimental results show that our proposed algorithm has better performance than the usual one.
