使用线性预测编码（Lpc），线性预测倒谱系数（Lpcc）和感知线性预测（Plp）对孤立单词的语音识别进行比较，以及模型阶数变化对语音识别率的影响-研究论文_线性预测倒谱系数

需积分: 46 1.5k 浏览量更新于2023-05-27 评论收藏 245KB PDF 举报

身份认证购VIP最低享 7 折!

领优惠券(最高得80元）

资源详情

资源评论

资源推荐

International Journal of Electronics and Communication Engineering Research and Development (IJECERD),

ISSN 2228 – 9282 (Print), ISSN 2248 –9290 (Online), Volume 1, Number 1, January - April (2011)

COMPARISON OF SPEECH RECOGNITION OF ISOLATED

WORDS USING LINEAR PREDICTIVE CODING (LPC), LINEAR

PREDICTIVE CEPSTRAL COEFFICIENT (LPCC) & PERCEPTUAL

LINEAR PREDICTION (PLP) AND THE EFFECT OF VARIATION

OF MODEL ORDER ON SPEECH RECOGNITION RATE

Yogesh S Angal

, R.H.Chile

, R.S.Holambe

Department of Instrumentation Engineering,

Padmshree Dr.D.Y.Patil Institute of Engineering and Technology, Pimpri, Pune, M.S, India.

2, 3

Department of Instrumentation Engineering,

S.G.G.S.Institution of Engineering and Technology, Vishnupuri, Nanded, M.S.,India.

ABSTRACT

Automatic Speech Recognition (ASR) by machine has been a goal of research from more

than six decades. In spite of all advances, machines cannot match the performance of

their human counterparts in terms of accuracy and speed, especially in case of speaker

independent speech recognition. So, today significant portion of speech recognition

research is focused on speaker independent speech recognition problem. Before

recognition, speech processing has to be carried out to get feature vectors of the signal.

So, front end analysis plays an important role. The reasons are its wide range of

applications and limitations of available techniques of speech recognition. The aim of

this paper is to study, implement and compare those widely used parameterization

method Linear Predictive Coding (LPC) techniques and Linear Predictive Cepstral

Coefficient (LPCC) and Perceptual Linear Prediction (PLP) in speech technology for

speech recognition .We have also observed the effect of variation of model parameter

on recognition rate. Vector Quantization (VQ) is used to prepare word model as a

template for each utterance. Moreover Euclidean distance is used as a classifier.

Front ends were compared in clean speech and with speech degraded by noise and

spectral variability, using the TI-46 word database. We have studied some aspects of

speech recognition and effect noise at various SNR levels 40dB, 35dB, 30dB, 20dB,

15dB, 10dB, 5dB, 0dB and-5dB. Experimentation is carried out with white Gaussian

noise.

It has been observed that LPCC gives better results as compared LPC in clean as well as

in noisy environment.PLP with DELTA and DELTA-DELTA works better than LPC and

LPCC in clean and noisy speech. Comparison of all above techniques for speech

IJECERD

International Journal of Electronics and Communication

Engineering Research and Development (IJECERD), ISSN

2228 – 9282 (Print), ISSN 2248 –9290 (Online), Volume 1,

Number 1, January - September (2011), pp. 07-19

http://www.prjpublication.com/IJECERD.asp

Electronic copy available at: https://ssrn.com/abstract=3527930

International Journal of Electronics and Communication Engineering Research and Development (IJECERD),

ISSN 2228 – 9282 (Print), ISSN 2248 –9290 (Online), Volume 1, Number 1, January - April (2011)

recognition is carried out in this paper. Suitability of each technique for different

environment is also discussed.

Keywords: ASR, LPC, LPCC, PLP, VQ.

1. INTRODUCTION

Speech recognition systems perform two fundamental operations: Signal modeling and

pattern matching. Signal modeling represents process of converting speech signal into a

set of parameters. Pattern matching is the task of finding parameter sets from memory

which closely matches the parameter set obtained from the input speech signal.

An Isolated Word Recognition (IWR) system consists of four main steps: preprocessing,

framing and windowing, feature extraction and pattern classification as shown in Figure

1. The function of preprocessing step is to make the spectrum flat (especially at high

frequency) and possibly removal of noise. The framing and windowing operation divides

the speech signal into overlapping frames and multiplies each frame by window. The

feature extractor does the job of mapping each frame of the speech signal into the set of

features called the feature vector which best approximates the signals property so

that efficient computations and compact representation of speech signal is possible. In the

pattern classifier module, word models are built for each word (available in the

vocabulary) from the feature vectors during the training phase. In order to recognize a

word, testing speech is passed through the same feature extractor and the test features are

compared with each of the stored models of utterances. The test sample of speech is

assigned to the word whose model gives the minimum distance or maximum probability

(based on the classifier used) with the test sample. Hence for an IWR system to work,

following major tasks are to be accomplished.

Training Phase

-Preprocessing of training speech samples.

-Framing and Windowing.

-Extracting appropriate speech features (feature vectors) from each windowed frame.

-Building a model for each word from feature vectors.

Testing Phase

-Preprocessing of test speech samples.

-Framing and Windowing.

-Extracting appropriate speech features from the given speech data.

-Obtain Euclidean distances (or Probabilities) of feature vectors with each word

model.

-Decision based on minimum distance (or maximum probability).

The main steps of extracting the important information from the speech signal

(During both training and testing phase) are Preprocessing, Framing and Windowing

and Feature extraction

Electronic copy available at: https://ssrn.com/abstract=3527930

International Journal of Electronics and Communication Engineering Research and Development (IJECERD),

ISSN 2228 – 9282 (Print), ISSN 2248 –9290 (Online), Volume 1, Number 1, January - April (2011)

Figure 1 Main steps in Isolated Words Speech Recognition (a) Training Phase (b)

Testing Phase

2. FEATURE EXTRACTION TECHNIQUES

Feature extraction aims at giving a useful representation of the speech signal by capturing

the important information from it. A common division of the feature extraction

approaches is into production-based and perception-based methods. Linear predictive

coding (LPC), Linear Predictive Cepstral Coefficient (LPCC) is an example from the first

group while Mel-frequency cepstral coefficients (MFCC) and Perceptual Linear

Prediction (PLP) belong to the perception-based approaches family. In speech

recognition, a premium is placed on extracting features that are somewhat invariant to

changes in the speaker [1, 2]. So feature extraction involves analysis of speech signal.

Broadly the feature extraction techniques are classified as temporal analysis and spectral

analysis technique. In temporal analysis the speech waveform itself is used for analysis.

In spectral analysis spectral representation of speech signal is used for analysis [3]. In

theory, it should be possible to recognize speech directly from the digitized waveform; it

should be represented directly in terms of the signal’s Fourier coefficients or as the set

of power values at the outputs from a bank of filters [4]. The envelope of the spectrum

can be represented indirectly in terms of the parameters of an all-pole model, using linear

predictive coding (LPC), or in terms of the first dozen or so coefficients of the

cepstrum—the inverse Fourier transform of the logarithm of the spectrum[5].

One reason for computing the short-term spectrum is that the cochlea of the human ear

performs a quasi-frequency analysis. The analysis in the cochlea takes place on a

nonlinear frequency scale (known as the Bark-scale or the mel-scale). This scale is

approximately linear up to about 1000 Hz and is approximately logarithmic thereafter.

Electronic copy available at: https://ssrn.com/abstract=3527930

剩余12页未读，继续阅读

weixin_38665944

粉丝: 6
资源: 915

会员权益专享

使用线性预测编码（Lpc），线性预测倒谱系数（Lpcc）和感知线性预测（Plp）对孤立单词的语音识别进行比较，以及模型阶数变化对...

评论0

会员权益专享

最新资源

使用线性预测编码（Lpc），线性预测倒谱系数（Lpcc）和感知线性预测（Plp）对孤立单词的语音识别进行比较，以及模型阶数变化对...

评论0

lpc 特征提取csdn

语音信号处理-LPC特征提取实验报告

号倒谱分析和 lpc 谱分析的编程流程图(图 2.1、图 2.2)编制浊音

用matlab识别电话音

论文中语音特征提取的一部分

基于matlab的语音信号处理

用python提取语音文件的lpcc、lpc、lsp特征

matlab语音识别系统

matlab在语音信号分析与合成中的应用

matble中线性预测参数频谱lpcar2pf函数如何实现

自己编写matble函数实现线性预测谱的绘制

写一个语音识别数据预处理算法

lpc55s69代码加密

lpc1768 app 加密

LPC2132\硬件资料\使用LPC2000 Flash Utility下载程序方法.doc

怎么写基于LPC1114的温度记录仪的程序

使用C语言编写LPC2138 通过SD更新程序

LPC2138 通过Sd卡更新Flash程序

LPC4357中文介绍

lpc1768的swd烧录方法

会员权益专享

最新资源