| 语音MFCC特征提取的FPGA实现 |
| |
| 引用本文： | 谢秋云,肖铁军.语音MFCC特征提取的FPGA实现[J].计算机工程与设计,2008,29(21). |
| |
| 作者姓名： | 谢秋云 肖铁军 |
| |
| 作者单位： | 江苏大学计算机科学与通信工程学院,江苏镇江,212013 |
| |
| 摘 要： | 提出了在FPGA上实现语音MFCC特征提取的方法,巧妙设计了FFT、三角滤波、取对数和DCT的硬件结构,采hl多时钟、状态机、模块复用、IP核和多级流水技术,大大提高了运算速度,实验表明与软件相比,结果误差在l%以内,50M时钟频率下每秒语音数据计算时间仅为8.5 ms,与VQ、HMM结合后,对系统识别率影响可以忽略,达到了嵌入式实时系统的应用要求.
|
| 关 键 词： | 语音识别 特征提取 嵌入式计算 |
Implementation of MFCC extraction in speech recognition based on FPGA |
| |
| XIE Qiu-yun,XIAO Tie-jun.Implementation of MFCC extraction in speech recognition based on FPGA[J].Computer Engineering and Design,2008,29(21). |
| |
| Authors: | XIE Qiu-yun XIAO Tie-jun |
| |
| Affiliation: | XIE Qiu-yun,XIAO Tie-jun(College of Computer Science , Telecommunications Engineering,Jiangsu University,Zhenjiang 212013,China) |
| |
| Abstract: | A method of MFCC extraction based on FPGA is presented.The hardware structure are designed skillfully of FFT,triangular filter,logarithm and DCT.Many techniques are used to obtain high speed,such as multiply clock,FSM,module reuse,IP core,and multiply pipeline.Experimental results show that the time of MFCC extraction per-second under 50M clock is only 8.5ms,and the percent error of final result is less than 1% compared with software,after combination with VQ and HMM,the effect on recognition rate is neglec... |
| |
| Keywords: | MFCC FPGA |
| 本文献已被 CNKI 维普 万方数据 等数据库收录！ |
|
