Research And Implementation Of Special English Audio Segmentation Based On Dual-Threshold

Mobile Learning is a new learning technique and method, which is quickly rising with the rapid development and integration of network technology, wireless communication technology, mobile computing technology and multimedia technology in recent years. Combining with mobile learning technique, building mobile English learning platform is a very valuable thing in research and practice.To meet the demand for resource in building mobile English learning platform, in this paper, we realize the MP3 decoding module, propose the audio segmentation algorithm based on dual-threshold, and explore the text's role in audio segmentation. Based on this work, an audio segmentation system is designed and implemented at last.The special English such as VOA, BBC and so on, which is widely popular on Internet, is fit for English learner. This paper mainly deals with this audio for segmentation. While this audio is mainly stored as MP3 format, instead of the waveform file format without compression on Internet. Therefore, this paper studies the theory of MP3, and introduces the process and some details of MP3 encoding and decoding algorithms firstly.Then, an audio segmentation algorithm based on dual-threshold named quiet energy threshold and quiet delay threshold is proposed in this paper, after analyzing the characteristics of the audio waveform and the differences and similarities between different language unit boundaries. At first, this algorithm estimates the dual-threshold by appropriate means, then extracts the energy sequences, and detects the audio sentence boundaries with the dual-threshold at last. The segmentation result of 28 pieces of VOA Special English audio shows that both the precision and the recall rate of this algorithm have exceeded 90%.Each piece of audio has related text file. Generally, the text effects in two ways in audio segmentation. On the one hand, it can provide sentence-level text corresponded to sentence-level audio. On the other hand, it can be used to proofread the audio segmentation result. For the former, this paper uses a rule-based method to detect text sentence boundaries; the experimental data indicates the correct rate is nearly 100%. For the latter, this paper proposes a proofread algorithm based on double thresholds, using the text segmentation information.Finally, an audio segmentation system is built, which includes MP3 decoding module, text segmentation module, audio segmentation module, and so on. Besides, a simple interface is designed in this paper.