摘要
设计并实现了一种嵌入式语音识别系统。硬件核心处理器是ADSP2181,语音接口芯片为AD1847。软件模块包括语音端点检测、MFCC求取、动态时间弯折算法、识别结果判定、模板训练等。系统使用定点DSP实现了浮点DSP运算,提高了运算的精度,扩大了信号处理的动态范围。实验结果表明,该系统对孤立词特定人的识别率为98%。它可用于声控玩具、智能家电等,具有较大的市场应用价值。
An embedded speech recognition system is designed and realized in this paper. Its hardware＇s core processor is ADSP2181.ADI847 is used as the system＇s speech interface chip. Speech recognition system＇s software consists of several modules such as endpoint detection, computation of MFCC, dynamic time warping algorithm, training of speech recognition referenced vectors, etc. The system realizes float-point operation on fixed-point digital signal processor, it has higher precision in calculation and wider signal processing dynamic range comparing to fixed-point realization scheme. The experiment confirms that its speech recognition accuracy reaches 98 percent for special person and small vocabulary. The system can be used in speech controlling toys, intelligent home appliances etc. It has much market potential.
出处
《计算机工程》
EI
CAS
CSCD
北大核心
2005年第16期160-162,共3页
Computer Engineering
关键词
语音识别
数字信号处理
MEL倒谱系数
Speech recognition
Digital signal processor
Mel frequency cepstrum coefficients （MFCC）
作者简介
李鹏怀（1978-），男，博士生，研究方向为语音识别、多媒体通信；E-mail：penghli@mail.ustc.edu.cn
徐佩霞，教授、博导
