基于VQ和ASR的多媒体课件检索-计算机系统结构专业论文.docxVIP

速度。 2、 多级检索算法将利用路径搜索算法进行关键词检出（keyword spotting）作为初级检索，给出一组粗略的候选结果。利用自动 语音识别技术（ASR）作为次级检索对候选结果进行筛选，验 证等处理，最终找出与关键词相匹配的项并根据似然度评分原 则排序输出。为学生提供多媒体课件检索服务。 3、 利用上述方法实现原型系统，并在实际课件点播环境中测试其 性能。同时研究相应算法在该环境中的可行性。 关键词：矢量量化，语音识别，关键字检出，多媒体检索 VQ and ASR Based Courseware Retrieval ABSTRACT In the E-learning environment, multimedia courseware is most widely used in daily teaching events. Retrieval services on multimedia courseware can definitely help to improve the quality of the learning by providing the students a convenient way to focus on the part where they have interest. And speech signal based retrieval method is a common technology in retrieving information from multimedia files. With well-developed speech recognition techniques, context based speech retrieval can be a good practice in this field and deserves research efforts. This article introduces a multimedia courseware retrieval method in E- learning environment. Since the variety of professional knowledge courseware may contain in its content, and especially those terminologies mentioned in lecture, it’s hardly possible to build a well language for the recognizer. So the method introduced here shall directly process the acoustic level signals to avoid the “out-of-vocabulary (OOV)” problem. Besides, the courseware is recorded in the classroom which makes the audio quality comparatively poor, our method should also be an error tolerant one. The main tasks in this article include: 1、 Analyze and design the vector quantization (VQ) technique in clustering the hidden Markov model (HMM) states. Generate a codebook for pre-processing the speech signal. This step do the time-consuming feature extraction and comparison work off-line so that the retrieval request can be processed with a quick response. Also, the speech signals are converted into a sequence of state numbers to speed up the pattern matching. 2、 Multilevel retrieval algorithm afterwards introduces a path searching method for keyword spotting as first-level