用于拟人机器人的嵌入式语音交互系统研究

SPEECH RECOGNITON & RESPONSION SYSTEM BASED EMBEDDED TECHNOLOGY FOR HUMANOID ROBOT

  • 摘要: 本文介绍了一种用于拟人机器人的嵌入式语音交互系统.系统采用高质量的语音采集模块及语音输出模块,以高性能数字信号处理器(DSP)TMS320VC5402为硬件核心.HMM语音识别引擎以LPC倒谱及其差分分量作为语音特征表达,改进的BaumWelch重估算法完成了多观察值序列下的语音模板训练.同时进行了语音特征不同表达形式对识别结果影响的对比实验.系统外围控制程序完成识别结果提示以及与上位机的通讯.系统在词汇量为200的非特定人、孤立词识别上取得了很好的效果.

     

    Abstract: This paper presents a speech recognition & responsion system based on embedded technology for humanoid robot.In this system,the high quality speech sampling module and response module are adopted.The physical hardware system is based on the 16 bit fixed point TMS320VC5402 DSP form Texas Instruments.It also includes a description of LPC analysis and Baum Welch re estimation,and shows how the Hidden Markov Models (HMM) recognition Engine and the system algorithm interact.An experiment is carried out to describe how different feature extractors influence the recognition result.The paper concludes with a description of a 200 vocabulary speaker independent and isolated word recognizer using the TMS320VC5402 DSP.

     

/

返回文章
返回