基于LPC美尔倒谱特征的带噪语音端点检测

复杂的噪声环境是语音识别系统在实际应用中性能下降的原因之一,识别预处理中的带噪端点检测作为关键技术,其性能的优劣某种程度上决定了识别率的高低。笔者提出了基于LPC美尔倒谱特征的带噪端点检测方法,对语音信号分高低频段分别提取IPC美尔倒谱特征分析,根据Mel倒谱距离判决,采用自适应噪声估计,实验结果表明,该方法计算效率较高,低信噪比下有较好的检测性能。     

一种新的抗噪语音特征的提取

针对LPCC易受噪声干扰和不能反映人耳听觉特性的缺点,提出了新的抗噪声特征,实验表明,在各种信噪比（SNR）环境中,该方法的精度高于线性预测倒谱系数（LPCC）和美尔倒谱系数（MFCC）.     

一种基于概率模型和倒谱差分的特征补偿算法

本文详细给出了概率模型中引入倒谱预测值的动态相关性来进行特征补偿的方法.该方法采用期望最大化(EM)算法来估计联合分布参数,基于语音和噪声的先验概率密度、在倒谱域对语音特征参数进行最小均方误差预测(MMSE),以提高语音识别精度.不同噪声环境和不同信噪比下的实验结果表明,本文方法能有效提高噪声环境下的中文连续语音识别的正确率.     

基于倒谱特征的带噪语音端点检测

在语音识别系统中产生错误识别的原因之一是端点检测有误差.在高信噪比情况下,正确地确定语音的端点并不困难.然而,大多数实际的语音识别系统需工作在低信噪比情况下,一些常规的端点检测方法,例如基于能量的端点检测方法在噪声环境下不能有效地工作.本文利用倒谱特征来检测语音端点,提出了带噪语音端点检测的两个算法,第一个算法利用倒谱距离代替短时能量作为判决的门限,第二个算法改进了基于隐马尔柯夫模型(HMM)的语音检测以适应噪声的变化,实验结果表明本方法可得到高正确率的带噪语音端点检测.     

基于倒谱特征的语音端点检测算法研究

本文在讨论基于倒谱距离语音端点检测算法不足的基础上,提出了两种改进方案.通过对三种典型噪声环境下信噪比(SNR)从-5dB到20dB的带噪语音信号进行的仿真实验结果表明,所提两种改进方案能更为准确地检测到语音的端点.     

用于分布式识别的语音参数的提取和性能分析

主要研究用于分布式语音识别(DSR)的语音参数的提取方法以及参数性能分析。以前所用到的语音参数大部分是LPC例谱参数，但其抗噪声性能较差。文中主要讨论了MEL倒谱参数。并在移动通信环境下，比较了两者的性能。     

基于多频带谱减法的抗噪声语音识别研究

为了减少在噪声环境下测试条件与训练条件不匹配导致的语音识别性能下降,提出了一种结合多频带谱减法的抗噪声语音识别系统。首先提取带噪语音的前几帧作为估计的噪声信号,将带噪语音、估计的噪声信号按频率划分M个互不相交的频带,然后根据每个频带内带噪语音与估计的噪声信号的性噪比,来确定该频带噪声的谱减参数。语音增强作为前端处理,与语音识别器级连构成抗噪声语音识别系统。通过实验仿真表明,基于多频带谱减法的抗噪声语音识别系统在不同信噪比不同类型的噪声下,识别性能明显优于基本谱减法。     

基于拉格朗日乘子SSMD和SSA的通信信号降噪方法

针对强噪声背景下通信信号的分析识别困难问题,提出一种基于拉格朗日乘子-辛奇异值模态分解(vSSMD)的奇异谱分析(SSA)降噪方法.鉴于噪声的随机变化使得采用功率谱密度方法计算嵌入维度时有较大误差,引入蒙特卡洛思想确定嵌入维数.噪声较大时,vSSMD通过构建拉格朗日乘子矩阵增强有用分量并抑制表示为噪声的残余信号,然后采用SSA方法去除vSSMD重构信号中的微弱噪声.将vSSMD-SSA算法的去噪效果与SSA、vSSMD方法进行比较,当信噪比为-14dB时,vSSMD-SSA算法相较于传统算法SSA信噪比提升了4.49dB,均方误差提升了38.25%.实验结果说明在低信噪环境比下,vSSMD-SSA算法的去噪效果最好.将vSSMD-SSA算法用于无人机通信信号去噪,降噪效果最明显.     

基于自适应滤波的噪声抵消法

语音降噪就是从带噪语音信号中提取尽可能纯净的原始语音。文中介绍了一种基于自适应滤波的噪声抵消法，采用归一化最小均方误差算法，采集实际噪声环境下各种不同信噪比的带噪语音样本进行降噪处理，实验结果表明，处理后信号的信噪比得到了较大程度的提高，大大改善了听音效果，具有很高的可懂度，且语音自然度好，没有失真；并与谱减法进行了比较，自适应噪声抵消法的降噪幅度比谱减法有一定提高，在听音效果上，用自适应噪声抵消法处理后的语音在清晰度、自然度方面优于谱减法。     

低信噪比环境下改进的倒谱距离语音端点检测算法

在强背景噪声的情况下,针对传统倒谱距离法端点检测难以判断语音段起止点的问题,提出了一种基于多窗谱估计的谱减法与改进的倒谱距离语音端点检测新方法.首先对每一帧带噪信号进行多窗谱估计得到平滑功率谱,提取前导无话段平均功率谱,再利用谱减法对带噪语音信号进行减噪处理,对语音的减噪是为了更好地进行下一步的端点检测,然后对传统的倒谱距离门限阈值进行改进,得到一种改进的自适应阈值,并结合倒谱距离法进行端点检测.通过仿真实验结果表明,与传统倒谱距离端点检测算法相比,本文方法提高了低信噪比语音端点检测的精度,具有良好的鲁棒性能.     

基于特征融合的海洋哺乳动物声音识别

为了提高海洋哺乳动物声音识别算法的识别率和鲁棒性,提出了一种将梅尔倒谱系数MFCC、线性倒谱系数LFCC和时域特征融合作为特征参数进行声音识别的方法。该方法通过融合不同倒谱系数以增强对不同频段的表征能力,通过融合时域特征来更全面地描述声音信息。声音样本通过基于海洋环境下的预处理、特征提取与融合后,用支持向量机进行分类识别。相对于传统算法只针对一种或几种哺乳动物进行识别,该方法在包含61种海洋哺乳动物声音的样本库中进行测试。测试结果显示该算法较传统的梅尔倒谱系数在识别率上提升了5.5%,且在海洋低信噪比环境下有更好的识别表现。     

OMP-BASED MULTI-BAND SIGNAL RECONSTRUCTION FOR ECOLOGICAL SOUNDS RECOGNITION简

The paper proposes a new method of multi-band signal reconstruction based on Orthogonal Matching Pursuit （OMP）, which aims to develop a robust Ecological Sounds Recognition （ESR） system Firstly, the OMP is employed to sparsely decompose the original signal, thus the high correlation components are retained to reconstruct in the first stage. Then, according to the frequency distribution of both foreground sound and background noise, the signal can be compensated by the residual components in the second stage. Via the two-stage reconstruction, high non-stationary noises are ef- fectively reduced, and the reconstruction precision of foreground sound is improved. At recognition stage, we employ deep belief networks to model the composite feature sets extracted from reconstructed signal. The experimental results show that the proposed approach achieved superior recognition per- formance on 60 classes of ecological sounds in different environments under different Signal-to-Noise Ratio （SNR）, compared with the existing method.     

Improved perceptually non-uniform spectral compression for robust speech recognition

According to the decline of recognition rate of speech recognition system in the noise environments, an improved perceptually non-uniform spectral compression feature extraction algorithm is put forward in this paper. This method can realize an effective compression of the speech signals and make the training and recognition environments more matching, so the recognition rate can be improved in the noise environments. By experimenting on the intelligent wheelchair platform, the result shows that the algorithm can effectively enhance the robustness of speech recognition, and ensure the recognition rate in the noise environments.     

Study on feature extraction method in border monitoring system using optimum wavelet packet decomposition

Border monitoring plays a key role in the national defense. In this study, we applied the sound identification technology on the border monitoring, and assumed an ideal border monitoring sound target recognition system. Feature extraction is a crucial step in this recognition system. As the border sounds are of non-stationary signals, the traditional methods failed to extract such kind of features. Fortunately, wavelet packet transform (WPT) can provide an arbitrary time-frequency decomposition for the signals. Based on WPT, a novel feature extraction method using optimum wavelet packet decomposition (OWPD) was proposed. According to the characteristics analysis of the border monitoring sounds using WPT, the signals were analyzed by selective multi-scale wavelet packet decomposition (i.e. OWPD), and then we built the meaningful and compact energy feature vectors as the input vectors of the BP neural network, in order to recognize the border monitoring sound. Extensive experimental results showed that this feature extraction method has convincing recognition efficiency.     

基于光纤光栅传感器的心音信号提取与处理算法研究

本文研究了一种用于光纤光栅智能服装的心音信号提取与处理算法,实现异常心音的初步识别。提出基于希尔伯特-黄变换（HHT）和小波阈值消噪相结合的心音提取算法,对波长解调信号进行消噪,提取有用的心音信号。采用数学形态学进行心音包络提取,提出基于直线结构元素和余弦结构元素相结合的心音处理算法,准确获取心音峰值点和起止点位置并计算心音特征值,根据心音特征值的临床意义,判断心音是否正常。实验结果表明该算法能够有效消除波长解调信号中的呼吸干扰与噪声,对20例实测正常心音和8类常见异常心音均能正确识别。该算法具有易实现、识别率高的特点,对光纤传感智能服装的研发和心脏疾病的早期诊断具有重要意义。      

Robust Voice Activity Detection Using the Spectral Peaks of Vowel Sounds

This letter proposes the use of vowel sound detection for voice activity detection. Vowels have distinctive spectral peaks. These are likely to remain higher than their surroundings even after severe corruption. Therefore, by developing a method of detecting the spectral peaks of vowel sounds in corrupted signals, voice activity can be detected as well even in low signal‐to‐noise ratio (SNR) conditions. Experimental results indicate that the proposed algorithm performs reliably under various noise and low SNR conditions. This method is suitable for mobile environments where the characteristics of noise may not be known in advance.     

卷积神经网络在异常声音识别中的研究

卷积神经网络(CNNs)已广泛应用于语音识别领域中以改善传统声学模型存在的鲁棒性弱、实时性差、识别性能低等缺点。本文对卷积神经网络在异常声音识别任务中的适用性及其识别性能进行了研究,针对日常常见的6种不同异常声音样本,分析了不同声音特征的维度对卷积神经网络识别性能的的影响,还将卷积神经网络分别与高斯混合模型、BP神经网络进行比较。实验结果表明,无噪声条件下,一维特征在卷积神经网络中的平均识别率比二维特征相对提升了2.91%,且误差收敛速度更快,但在有噪声条件下,二维特征的平均识别率比一维特征相对提升了3.41%。同时卷积神经网络比其它两种识别模型在对噪声的鲁棒性和误差收敛速度等方面均有明显的优势。      

基于时频特征的跳频信号调制识别

跳频信号在抗干扰方面具有良好的性能。准确识别跳频信号的调制方式,能够为判断敌我目标属性、干扰敌方信号等军事信息战提供有力支撑,但国内外对于跳频信号的调制识别仍存在很大空缺。本文提出一种基于时频特征的跳频信号调制识别方法,通过平滑伪魏格纳-维利分布(SPWVD)时频变换获取不同调制类型的跳频信号时频图像,将时频图像送入卷积神经网络(CNN)中进行特征提取及分类识别。仿真实验证明,本文CNN在低信噪比下取得了较好的识别效果。     

A comparative approach between cepstral features for human authentication using heart sounds

The main objective of this paper is to provide a comparative study between different cepstral features for the application of human recognition using heart sounds. In the past 10 years, heart sound, which is known as phonocardiogram, has been adopted for human biometric authentication tasks. Most of the previously proposed systems have adopted mel-frequency and linear frequency cepstral coefficients as features for heart sounds. In this paper, two more cepstral features are proposed. The first one is based on wavelet packet decomposition where a new filter bank structure is designed to select the appropriate bases for extracting discriminant features from heart sounds. The other is based on nonlinear modification for mel-scaled cepstral features. The four cepstral features are tested and compared on two databases: One consists of 21 subjects, and the other consists of 206 subjects. Based on the achieved results over the two databases, the two proposed cepstral features achieved higher correct recognition rates and lower error rates in identification and verification modes, respectively.