Speech/music indexing for audio life-logs from portable device record

Audio plays an important role among information sources in our life. As a result of current technology, it is available to record people´s huge personal life activities as life-logs from long-term and multi-dimensional point of view on portable device. In order to make record be effective, this paper focuses on implementing the classification among speech, music and other kinds of sound around, which are collected from peoples´ daily acoustic life logs by smart phone. The separation experiments were carried out on daily life logs recording, while the performance of discrimination has been achieved by using Mel-Frequency Cepstrum Coefficients (MFCC) and Short-Term Energy based on multilayer feed forward Artificial Neural Network (ANN). Tests have shown encouraging result with accuracy around 86% by means of the proposed method, compared to 80% accuracy by only MFCC feature, which is completely enough for vague search among long-term audio life-logs.