基于DFT的嵌入式普通话语音快速识别

发布时间：2018-04-19 16:22

本文选题：语音识别 + 频谱分析　；参考：《实验室研究与探索》2017年06期

【摘要】：基于选定的频率普通话发音识别技术,由定义的输入普通话语音拾取的普通话语音识别腔共振曲线,通过对中文分词、韵母声母频谱识别,对中文语音的声母、韵母、声调集成、输出的中文拼音词序列组合,形成句子的中文拼音序列。通过韵母分析方法进行单音节,频域幅频特性分析技术,不需要学习或培训事先要用于一个特定的人的声音,能够识别非特定的单词和句子的普通话语音,采用速率自适应梳状(Discrete Fourier Transform,DFT)谱分析算法,只需极少量的计算,就能够确定地包含4个声调的普通话,具有响应速度快,所需存储空间小,能方便的移植到嵌入式设备的应用程序中。
[Abstract]:Based on the selected frequency Putonghua pronunciation recognition technology, the resonance curve of Putonghua speech recognition cavity picked up by the defined input Putonghua speech is used to recognize the initial consonant and vowel of Chinese speech through the recognition of Chinese participle, vowel consonant spectrum, Chinese consonant and vowel.Tone integration, output Chinese phonetic word sequence combination to form sentence Chinese phonetic sequence.Single syllable and frequency-domain amplitude-frequency characteristic analysis are carried out by vowel analysis method. It does not need to be learned or trained to be used in a specific person's voice in advance, and can recognize the Mandarin pronunciation of non-specific words and sentences.Using the rate adaptive comb Fourier transform (DFT) spectral analysis algorithm, it is possible to contain four tones of Putonghua with a very small amount of calculation, which has the advantages of fast response speed and small storage space.Can be easily transplanted to embedded device applications.
【作者单位】：桂林理工大学南宁分校;
【基金】：广西科学研究与技术开发计划项目(桂科攻12118017-10) 2016年度广西职业教育教学改革研究立项项目(重点)(桂教职成[2016]26号) 广西高等学校优秀中青年骨干培养工程第一期(桂教人[2013]16号
【分类号】：TN912.34

【相似文献】