论文部分内容阅读
语音识别作为计算机的输入手段,国内外的研究工作进展甚快。美国、日本和我国哈工大研究的连续词汇的识别系统,日趋实用化。但是基于过零率特征的语音识别的研究工作不多。我们使用一台BCM-80单板计算机,提取语音过零率,判定过零率和能量特征的端点,并用线性正参数和多模板技术,识别汉语十个数字。正识率为86%;识别规范的命令,精度在95%以上。可望在声控操作传递规范的指令等方面有一定的应用价值。
Speech recognition as a means of computer input, research work at home and abroad has progressed rapidly. The continuous vocabulary recognition system of the HIT’s research in the United States, Japan and China is becoming more and more practical. However, the research on speech recognition based on the characteristics of zero-crossing rate is limited. We use a BCM-80 single-board computer to extract the zero-crossing rate of speech, determine the endpoint of the zero-crossing rate and energy characteristics, and identify ten Chinese digits using linear positive parameters and multi-template techniques. Mastery rate was 86%; identification of the order, accuracy of 95% or more. It is expected that there will be some application value in the aspects of voice control operation such as transferring the normative instructions.