I
摘
要
语音识别主要是让机器听懂人说的话,即在各种情况下,准确地识别出语音的内容,
从而根据其信息执行人的各种意图。语音识别技术既是国际竞争的一项重要技术,也是每
一个国家经济发展不可缺少的重要技术支撑。
本文基于语音信号产生的数学模型,
从时域、
频域出发对语音信号进行分析,论述了语音识别的基本理论。在此基础上讨论了语音识别
的五种算法:
动态时间伸缩算法
(Dynamic Time Warping
,
DTW)
、
基于规则的人工智能方法、
人工神经网络
(Artificial Neural Network
,
ANN)
方法、隐马尔可夫
(Hidden Markov Model
,
HMM)
方法、
HMM
和
ANN
的混合模型。
重点是从理论上研究隐马尔可夫
(HMM)
模型算法,
对经典的
HMM
模型算法进行改进。
语音识别算法有多种实现方案,本文采取的方法是利用
Matlab
强大的数学运算能力,
实现孤立语音信号的识别。
Matlab
是一款功能强大的数学软件,它附带大量的信号处理工
具箱为信号分析研究,特别是文中主要探讨的声波分析研究带来极大便利。本文应用隐马
尔科夫模型
(HMM)
为识别算法,采用
MFCC(MEL
频率倒谱系数
)
为主要语音特征参数,
建立了一个汉语数字语音识别系统,其中包括语音信号的预处理、特征参数的提取、识别
模板的训练、识别匹配算法;同时,提出利用
Matlab
图形用户界面开发环境设计语音识别
系统界面,设计简单,使用方便,系统界面友好。经过统计,识别效果明显达到了预期目
标。
关键词
:语音识别算法;
HMM
模型;
Matlab
;
GUI
ABSTRACT
Speech
Recognition
is
designed
to
allow
machines
to
understand
what
people
say,and
accurately
identify
the
contents
of
voice
to
execute
the
intent
of
people.Speech
recognition
technology
is
not
only
an
important
internationally
competed
technology,but
also
an
indispensable
foundational
technology
for
the
national
economic
development.Based
on
the
mathematical
model
from
the
speech
signal,this
paper
analyze
audio
signal
from
the
time
domain,frequency
domain
proceeding,and
discussed
the
basic
theory
of
speech
recognition
technology.Five
algorithm
are
discussed:Dynamic
Time
Warping(DTW)
、
Rule-based
Artificial
Intelligence,Artificial
Neural
Network(ANN),Hidden
Markov
Model(HMM),HMM
combined
with
ANN.The
focus
is
put
in
the
theoretical
studies
of
Hidden
Markov(HMM)
model
algorithm,and the classical HMM algorithm is improved.
Speech recognition algorithm is realized in various programs,this article taking the method
is
to
use
Matlab
powerful
mathematical
operation
ability
to
realize
the
recognition
of
speech
signal isolation. Matlab is a powerful mathematic software with a mass of toolboxes dealing with
signal
processing.
It
gives
a
terrific
shortcut
to
the
research
of
signal
processing,especially
the
wave analysis. We can characterize the sound with key parameters such as intensity, frequency
etc.
In
this
paper,
hidden
Markov
model
(HMM)
recognition
algorithm
using
MFCC
(MEL