Innovation in Language Learning
Speech Recognition Technique in Oral English Learning for Mobile Devices
Xia Liu, Harbin University of Science and Technology (China)
Abstract
With the scientific and technological progress as well as the rapid economic growth, the terminal capabilities of Smartphone have globally improved in a large degree. Mobile phone manufacturers intensify software development regarding Smartphone in succession and variety of voice phone comes into being thereupon. However, the speech recognition of ordinary phones aims at realizing operation through human-mobile interaction, in lack of the applied research and development specializing in the demand of English learners. English oral communication has always played a leading role in international exchanges. Thus, with portable terminal of Smartphone, an intellectual learning system free from the limitation of time, location and teacher resources will provide users with better and faster electronic learning
method.
Presently, the intellectual English learning software ground on PC has already been able to provide PC-based learning skills, whose intelligentized function enables learners to obtain grades on pronunciation quality in time. The transplantation of such software to cellphone platform would be limited by factors as operation speed of cellphone, memory space and bus bandwidth. With regard to the software and hardware limitation of embedded system, this paper has studied an English learning system on cellphone platform on the basis of continuous speech recognition technology. The system implements effective appraisals of learners' pronunciation through speech recognition and feedbacks the information to users. Taking SPHINX of Carnegie Mellon University as the system's core recognition decoder, the system development has predominance in the spontaneous and conversational speech. In the meantime, the transformation of general HIDDEN MARKOV MODELS (HMM) to semi-continuous HIDDEN MARKOV MODELS (CHMM) lessens the amount of calculation to a large extent in the process of recognition. Moreover, according to the special application of English pronunciation learning while taking sentence-learning as a priori knowledge, the system clips only sentence read during the recognition, decreasing search space greatly as well as shortening the response time of system.
At the end of this paper, I compare the PC system and my system,
and I found that, my system lose some accuracy, but it run faster, and it is more suitable for mobile device.
Keywords: Speech Recognition Technique,Hidden Markov Model,Language learning
