当前位置：文档库 › Speech Enhancement Based on Spectral Estimation from Higher-lag Autocorrelation

Speech Enhancement Based on Spectral Estimation from Higher-lag Autocorrelation

Speech Enhancement Based on Spectral Estimation from Higher-lag

Autocorrelation

Benjamin J.Shannon,Kuldip K.Paliwal and Climent Nadeu?

School of Engineering,Grif?th University

Brisbane,QLD4111,Australia

Ben.Shannon@https://www.wendangku.net/doc/c75949967.html,.au,K.Paliwal@https://www.wendangku.net/doc/c75949967.html,.au,climent@talp.upc.es

Abstract

In this paper,we propose a unique approach to enhance speech signals that have been corrupted by non-stationary noises.This ap-proach is not based on a spectral subtraction algorithm,but on an algorithm that separates the speech signal and noise signal contri-butions in the autocorrelation domain.We call this technique the AR-HASE speech enhancement algorithm.

In this initial study,we evaluate the performance of the new algorithm using the average PESQ score computed from10 male utterances and10female utterances taken from the TIMIT database as a measure of speech quality.We test the algorithm using one broadband stationary noise and two non-stationary noises.We will show that the AR-HASE enhancement algorithm produces near transparent quality for clean speech,gives poor enhancement performance for broadband stationary noises,and gives signi?cantly enhanced quality for the two non-stationary noises.

Index Terms:speech enhancement,autocorrelation,impulsive noise.

1.Introduction

Many of the state-of-the-art speech enhancement algorithms use the analysis-modi?cation-synthesis framework[1]in their opera-tion.In this framework,the corrupted speech signal is broken up into short-time segments,which are transformed to the frequency domain where only the spectral magnitude is modi?ed.The speech signal is then reconstructed with an inverse short-time Fourier transform followed by an overlap-add operation.This structure is used by the popular spectral subtraction algorithm,originally proposed by Boll[2]in1979,and also by techniques related to Wiener?ltering,such as Ephraim-Malah’s method[3]and all its more recent variants.

These spectral enhancement algorithms require an estimate of the noise spectrum,which can be obtained from non-speech seg-ments indicated by a voice activity detector or,alternatively,with a minimum statistics approach[4],i.e.by tracking spectral minima in each frequency band.In consequence,they are effective only when the noise signals are stationary or at least do not show rapidly varying statistical characteristics.The worst type of noise for these systems is when the noise signal is typically coincident with the ?This work was performed while Climent Nadeu was on leave from the Signal Theory and Communications Department,Universitat Politec-

nica de Catalunya,08034Barcelona,Spain.speech signal,and absent at other times.This situation,for exam-ple,could arise with an impulsive noise.In this case,most of the non-speech frames could be completely devoid of impulsive noise, but the speech frames could contain a large amount of this noise. To handle these situations,noise reduction techniques that oper-ate intra-frame(within the current frame)are required;these tech-niques cannot use the noise power spectrum estimate from other non-speech frames.

In previous work,we have proposed a noise robust spectral estimation technique for short-time speech signals that operates intra-frame.This method uses the periodic correlation property of short-time speech signals and the autocorrelation domain to per-form noise reduction.It is well known that the pitch period of human speech is typically constrained to values between2ms and 12ms.This means that in the autocorrelation domain,we will have large magnitude coef?cients at these periods.This property, conversely,is generally not true for noise signals.By computing a spectral estimate using only the higher-lag autocorrelation coef-?cients,we have a way of separating the speech and noise signal without having to estimate the noise signal directly.We call this method,Higher-lag Autocorrelation Spectral Estimation(HASE) [5][6].

The HASE method was motivated by the large volume of pre-vious work on noise robust Automatic Speech Recognition ASR feature extraction based on autocorrelation domain processing[7] [8][9][10].This method has been successfully applied to the noise robust ASR problem,particularly where the noise signal had rapidly changing characteristics.The goal of ASR feature extrac-tion is to produce features that have a low dimensionality,are in-sensitive to speaker and environmental changes and are effective in discriminating the linguistic units.These goals have little in common with the goals of speech enhancement.

In this paper,we investigate the HASE algorithm for speech enhancement.We show that this algorithm has some inherent lim-itations for enhancement applications.We propose to overcome these limitations by using an Auto-Regressive(AR)model of high order.We refer to this extended HASE algorithm as the AR-HASE algorithm.It is our aim in this work to explore the potential of this technique for the enhancement of speech signals corrupted by both stationary and non-stationary disturbances.

2.Speech Enhancement using Higher-lag

Autocorrelation Spectral Estimation

A brief description of the previously proposed Higher-lag Auto-correlation Spectral Estimation(HASE)technique proceeds as fol-lows.The short-time speech segment(approx.32ms)is

?rst

windowed using a Hamming window.Following this,a biased estimate of the autocorrelation sequence is made.Once the auto-correlation sequence is computed,the higher-lag range(2ms to 32ms)of one-side of the autocorrelation sequence is windowed using a high dynamic range window function.The Double Dy-namic Range(DDR)window function design method[5]is used to compute this window.The magnitude spectrum of the windowed higher-lag autocorrelation sequence is then computed as an esti-mate of the short-time power spectral density.

The speech enhancement framework that we?rst used to eval-uate the performance of the HASE algorithm for speech enhance-ment is shown in Fig.1.Here,we have taken the typical spectral subtraction algorithm and modi?ed it.We have substituted the en-hanced short-time power spectrum estimate in the spectral subtrac-tion framework with the power spectral estimate computed using the HASE algorithm.

As mentioned previously,the spectral subtraction algorithm requires an estimate of the noise power spectrum.In the proposed framework,this estimate is not required.The speech signal en-hancement is performed based on prior knowledge of the auto-correlation sequences of typical speech and noise signals.Speech signals(particularly voiced)have autocorrelation sequences with large magnitude coef?cients at higher-lag values.This property is not typically observed in noise signals.Therefore,by using only the higher-lag portion of the autocorrelation sequence to compute a spectral estimate,the noise contribution is reduced.

The?rst problem we encountered in applying the HASE al-gorithm in this framework is the Fourier phase spectrum and the HASE magnitude spectrum are not well matched.To achieve good results in the synthesis stage,the pitch harmonic features in the phase spectrum and magnitude spectrum need to match well.This problem is demonstrated in the analysis shown in Fig.2.This?g-ure shows the Fourier power spectrum of a32ms frame containing an/iy/sound(plot(a)dashed line)and the group delay sequence computed from the Fourier phase spectrum(plot(b)).Wherever a pitch harmonic is present in the Fourier power spectrum,the corre-sponding group delay sequence shows a near constant value.How-ever,the low power regions between the pitch harmonics give spu-rious values in the group delay sequence.Figure2(a)also shows the HASE spectral estimate(solid line)for the same frame.Due to the extra windowing steps in the HASE algorithm,the band-widths of the pitch harmonics are larger than in the direct case. This means that in the synthesis stage,relatively high magnitude spectral coef?cients are matched with spurious phase coef?cients. This is the cause of noticeable distortion in the output speech.

There are several ways to reduce the problem of pitch har-monic bandwidth mismatch between the magnitude and phase spectrum.The approach we have chosen for this study is to in-crease the number of samples used in estimating the magnitude spectrum.For example,in the case of the HASE algorithm,a 32ms frame is processed.This allows a one-sided biased auto-correlation sequence to be computed that has a lag range of up to32ms.To reduce the bandwidth of the pitch harmonics,we need a one-sided autocorrelation sequence with a lag range greater than32ms.Extension of this sequence can be achieved with the aid of Auto-Regressive(AR)modelling.Rather than computing the biased estimate of the autocorrelation sequence using the FFT algorithm,we propose to compute it as the inverse Fourier trans-form of a high order AR power spectral estimate,thus extending the non-zero lag range beyond32ms.This approach also provides a further degree of freedom.By manipulating the order of the

Figure2:Comparison of Fourier power spectrum and the HASE power spectrum.(a)Power spectrum estimate of a32ms frame containing an/iy/sound using Fourier transform(dashed line)and the HASE algorithm(solid line).(b)Group delay sequence com-puted from the Fourier phase spectrum.

AR model,we can tune the performance of the enhancement algo-rithm.A brief evaluation of the proposed HASE and AR-HASE based enhancement algorithm are now presented.

3.Experimental Evaluation

In this section,we evaluate the performance of both the HASE and the AR-HASE algorithm.We?rst explore the performance of the HASE algorithm in clean conditions to determine how signi?cant the pitch pulse bandwidth mismatch problem discussed in section 2is.We then go on and test the enhancement potential of the AR-HASE algorithm using three types of noise.One of the noises is a stationary type noise and the other two are non-stationary.

To evaluate the performance of the proposed speech enhance-ment algorithms,we took20speech?les from the TIMIT database and down-sampled them to a sampling frequency of8kHz.The 20utterances came from10different male and10different female https://www.wendangku.net/doc/c75949967.html,ing these20samples,the average PESQ[11]score was computed as a measure of performance.PESQ stands for ”Perceptual Evaluation of Speech Quality”.This algorithm was designed to provide a way to estimate the subjective quality of speech.The output from the algorithm is an estimate of the Mean Opinion Score(MOS),which is a number between1and5.The meanings assigned to the scores in relation to the speech quality are:1-Bad2-Poor3-Fair4-Good5-Excellent.

The three noise samples used in the evaluation are theoreti-cally ideal for the HASE algorithm.That is,for an analysis frame size of32ms,the theoretical autocorrelation sequence has high

magnitude coef?cients for time lags between0and2ms and zero value coef?cients for time lags greater than2ms.These three noises are white Gaussian noise,repeating impulse noise and re-peating chirp noise.

The three noises were created using the following steps.The arti?cial white noise was obtained using a Gaussian random num-ber generator.To create the arti?cial impulsive noise,we?rst be-

Enhanced Speech

Speech

Corrupt Figure 1:Block diagram of the proposed AR-HASE based speech enhancement algorithm.

Model Order

Male Female Combined 32 2.73 2.50 2.6164 3.29 3.23 3.2696 3.25 3.85 3.55128 3.13 3.83 3.48160 3.11 3.79 3.45192 3.06 3.72 3.39224 2.97 3.65 3.31255

2.89

3.59

3.24

Table 1:Mean PESQ scores of AR-HASE algorithm with different AR model orders tested on clean speech.

gan with a 32ms block of zeros.To this block,we added a unit pulse of 2ms duration.The starting position of the 2ms pulse was randomly selected between 0and 30ms using a uniform random number generator.We then concatenated this block with another 32ms block that contained only zeros.These two steps were then repeated,but this time the sign of the 2ms pulse was reversed to maintain zero mean.These four steps were then repeated contin-uously to get a suf?ciently long sequence of the impulsive noise.Thus,for this noise,the separation between successive pulses ran-domly varies between 32to 92ms.Finally,the arti?cial chirp noise was created by de?ning one period of the chirp as a sinusoidal sig-nal whose frequency changes linearly from 0kHz to 4kHz (half of the sampling frequency)over a period of 32ms.This period was then repeated to give a sequence of suf?cient length.3.1.HASE enhancement

Using the HASE algorithm in the proposed modi?ed analysis -modi?cation -synthesis speech enhancement framework gave a mean PESQ score of 2.85for clean speech.This is considered a low score for clean speech.As expected,distortion was also noted during listening.

3.2.AR-HASE enhancement

The ?rst evaluation of the AR-HASE algorithm is performed on clean speech for different AR model orders.A high model order is expected to give better performance;therefore,we start at a model order of 32and increase it by 32until all the frame data is used in the AR modelling.These results are shown in Table 1.

Since a model order of 96gave the best performance in clean

conditions,this model order is used in the enhancement evaluation.The results comparing the AR(96)-HASE enhanced speech with unenhanced speech is given in Fig.3.

4.Discussion

When we apply the HASE algorithm to clean speech utterances and listen to the HASE enhanced utterances,the speech is easily understood,but it sounds like the speakers pitch has been distorted.We get an average PESQ score of 2.85which is approximately equivalent in speech quality to speech corrupted with white Gaus-sian noise at a global SNR of 20dB.Thus,the HASE enhancement algorithm reduces the speech quality signi?cantly for clean speech signals.Therefore,we have disregarded this algorithm.

When we apply the AR-HASE algorithm to clean speech sig-nals and investigate its performance as a function of AR model order,the peak in speech quality occurs at a model order of 96.To compute the AR model of order 96,autocorrelation lags up to 12ms are used.This is suf?cient to cover the pitch period of most human speakers.For example,if we take a voiced speech frame from a speaker that has a pitch of 100Hz,then compute a Fourier spectrum from 0to 4kHz,we expect to see 40peaks.To make an AR spectrum match well with each of the 40peaks in the Fourier spectrum,we would require a minimum of 80poles in the AR model.Therefore,an order of 96makes intuitive sense.

The AR-HASE algorithm is nearly transparent for clean speech.Where there is noticeable distortion,it sounds more like a reverberant distortion than an additive background distortion.The average PESQ score for clean speech was 3.55.This was equiva-lent to a speech quality of speech corrupted with white Gaussian noise at a global SNR >30dB.

The enhancement properties of the AR-HASE algorithm were dependent on the corrupting additive noise.For the broadband white Gaussian noise,no enhancement in quality was achieved us-ing a model order of 96.We attribute the poor performance for this case to the estimate of the short-time autocorrelation sequence.Over a short analysis frame,the autocorrelation estimate of a white broadband noise is far from the asymptotic estimate.In informal testing,it was found that by using a very low model order (12-24),the white Gaussian noise could be

eliminated from the speech,but this was at the expense of signi?cant speech distortion.

The AR-HASE algorithm worked very well for the non-stationary noises.For the impulsive noise and repeating chirp noise at 5dB SNR,the average PESQ scores were 0.87and 0.97

Figure3:PESQ performance of AR-HASE speech enhancement compared to unenhanced speech.(a)White Gaussian noise.(b) Repeating impulse noise.(c)Repeating chirp noise.

higher than the unenhanced scores respectively.This is equivalent to a listener’s opinion of the speech quality moving from poor to fair.

Since the AR-HASE algorithm gave good enhancement per-formance for the non-stationary noises and poor performance for the broadband stationary noise,it could be possible to get bet-ter performance by combining this algorithm with an existing en-hancement algorithm such as spectral subtraction.For this type of approach,the contributions from both algorithms may be compli-mentary.That is,if we use the spectral subtraction and the AR-HASE algorithm in cascade,the spectral subtraction algorithm could remove the stationary noise,and the following AR-HASE algorithm could reduce any residual non-stationary noise.

5.Conclusion

In this paper,we have proposed a new approach to the enhance-ment of speech signals that have been corrupted by non-stationary, additive and uncorrelated noise signals.This approach was not based on a spectral subtraction algorithm,but on an algorithm that separates the speech signal and noise signal contributions in the autocorrelation domain.This technique was called the AR-HASE algorithm.

The AR-HASE algorithm was?rst tested on clean speech sig-nals.It was shown that after choosing an appropriate AR model order,near transparent quality could be achieved for clean speech. The algorithm was then tested on three types of noise signals using the average PESQ score as a speech quality measure.

For broadband stationary noise,little enhancement of the speech quality was gained using the AR-HASE algorithm.For the other two noises tested,repeating chirp and impulsive noise,a large improvement in speech quality was measured.

6.References

[1]J.B Allen,“Short Term Spectral Analysis,Synthesis,and

Modi?cation by Discrete Fourier Transform,”IEEE Trans.

ASSP,vol.25,no.3,pp.235–238,1977.

[2]S.F.Boll,“Suppression of Acoustic Noise in Speech Using

Spectral Subtraction,”Acoustics,Speech and Signal Process-ing,vol.ASSP-27,no.2,pp.113–120,1979.

[3]Y.Ephraim and D.Malah,“Speech enhancement using a

minimum mean-square error short-time spectral amplitude estimator,”IEEE Trans.ASSP,vol.32,pp.1109–1121,1984.

[4]R.Martin,“Noise power spectral density estimation based

on optimal smoothing and minimum statistics,”IEEE Trans.

SAP,vol.9,no.5,pp.504–512,2001.

[5] B.J.Shannon and K.K.Paliwal,“MFCC Computation from

Magnitude Spectrum of Higher Lag Autocorrelation Coef-?cients for Robust Speech Recognition,”in Proc.ICSLP, 2004.

[6] B.J.Shannon and K.K.Paliwal,“Spectral Estimation using

Higher-lag Autocorrelation Coef?cients with Applications to Speech Recognition,”in Proc.ISSPA,2005,pp.599–602.

[7]J.A.Cadzow,“Spectral Estimation:An overdetermined

rational model equation approach,”in Proc.IEEE,1982, vol.70,pp.907–939.

[8]Y.T.Chan and https://www.wendangku.net/doc/c75949967.html,ngford,“Spectral Estimation via the

High-Order Yule-Walker Equations,”IEEE Trans.on ASSP, ,no.5,pp.689–698,1982.

[9] D.Mansour and B.H.Juang,“The Short-Time Modi?ed

Coherence Representation and Noisy Speech Recognition,”

IEEE Transactions on ASSP,vol.37,no.6,pp.795–804, 1989.

[10]J.Hernando and C.Nadeu,“Linear Prediction of the One-

Sided Autocorrelation Sequence for Noisy Speech Recogni-tion,”IEEE Transactions

on Speech and Audio Processing, vol.5,no.1,pp.80–84,1997.

[11] A.W Rix,J.G.Beerends,M.P.Hollier,and A.P.Hek-

stra,“Perceptual evaluation of speech quality(PESQ)-a new method for speech quality assessment of telephone networks and codecs,”in Proc.ICASSP,2001,vol.2,pp.749–752.

合工大研究生英语翻译

汉翻英第一单元 1、要善于恭维他人，重要的一步就是要懂得为什么恭维会有助于你建立更好的人际关系。 An important step in becoming an effective flatterer is to understand why flattery helps you establish better relationships with others. 2、恭维之所以奏效，最根本的原因是恭维符合了人类行为的一个基本原则：人们渴望得到赏识。 The root cause of the power of flattery gets at(达到) a basic principle of human behavior：People crave being appreciated. Power of flattery lies in the fact that it fulfils a basic principle of human behavior: 3、尽管文化背景各不相同，但绝大多数人都有类似的想法。 The vast majority of people are of the similar idea despite different cultures. 4、在亚洲文化中，人们对群体赏识的渴求一般要强于对个体赏识的渴求。但不管怎样，人们渴望赏识是普遍存在的。 In Asian cultures the desire for group recognition is generally stronger than the desire for individual recognition. Nevertheless, the need for recognition is present/universal. 5、很多人认为，工作本身带来的乐趣要比外界赏识包括恭维更为重要。Many people hold that the joy of work itself is more important than external recognition, including flattery. 6、工作的乐趣也许是一种巨大的动力，但是即使是那些从工作中得到极大乐趣的人如科学家、艺术家、摄影师也渴望得到恭维和认可，否则他们就不会去竞争诺贝尔奖或在重要的展览会上展示他们的作品了。 The joy of work may be a powerful motivator, but even those who get the biggest joy from their work —such as scientists, artists, and photographers —crave flattery and recognition. 7、否则他们就不会去竞争诺贝尔奖或在重要的展览会上展示他们的作品了。Otherwise they wouldn’t compete for Nobel Prizes or enter their work in important exhibitions. 8、恭维之所以奏效，还因为它与人们对认可的正常需要有关。 Another reason flattery is so effective relates to the normal need to be recognized. 9、尽管有一些关于恭维的书和文章问世，并对恭维极力进行宣扬，但是大多数人还是没有得到应有的赏识。 Although some articles and books have been written and preached zealously about flattery, most people receive less recognition than they deserve. 10、很多人无论在工作上或在家里都很少收到赞美，所以对认可的渴求就更加强烈了。 Many people hardly ever receive compliments either on the job or at home, thus intensifying their demand for flattery. 汉翻英第二单元 1、鲜花是最常送的礼物之一。 Flowers are among the most frequently given gifts. 2、有一种传统的用鲜花表达的语言。精心挑选的一束花卉可以传达多种不同的情感和祝福。 There’s a traditional floral language, and a carefully selected bouquet or plant can convey a wide range of emotions （love, affection, pity）and sentiments 、 3、红玫瑰象征着爱情也象征着新事业充满希望的开端； Red roses symbolize love as well as the hopeful beginning of a new enterprise; 4、紫罗兰是祈求受花人不要忘却送花人。 violets beseech the recipient not to forget the donor; 5、兰花以及其他精美的花卉则表示（你希望）受花人认为你情调高雅（decent, graceful, elegant）、受人尊重(esteemed, respectable)、出类拔萃(outstanding, distinguished, celebrated)。 orchids and other exquisite blooms indicate that the recipient regards you as exotic（珍奇的）, precious（讲究的）and rare（珍贵的）. 6、送一束鲜花能唤起温馨的回忆，比那些仅仅显示炫耀和奢华的礼物更为珍贵。 A floral gift that evokes warm recollections will be prized more than one that is

合工大研究生英语课后汉翻英中英对照版考试必备(精)

要善于恭维他人,重要的一步就是要懂得为什么恭维会有助于你建立更好的人际关系。 An important step in becoming an effective flatterer is to understand why flattery helps you establish better relationships with others. 恭维之所以奏效,最根本的原因是恭维符合了人类行为的一个基本原则:人们渴望得到赏识。 The root cause of the power of flattery gets at a basic principle of human behavior: People crave being appreciated., 尽管文化背景各不相同,但绝大多数人都有类似的想法。 he vast majority of people are of the similar idea despite different cultures. 在亚洲文化中,人们对群体赏识的渴求一般要强于对个体赏识的渴求。 In Asian cultures the desire for group recognition is generally stronger than the desire for individual recognition. 但不管怎样,人们赏识是普遍存在的。 Nevertheless, the need for recognition is present. 很多人认为,工作本身带来的乐趣要比外界赏识包括恭维更为重要。工作的乐趣也许是一种巨大的动力,但是即使是那些从工作中得到极大乐趣的人,如科学家、艺术家、摄影师也渴望得到恭维和认可,否则他们就不会去竞争诺贝尔或在重要的展览会上展示他们的作品了。 Many people hold that the joy of work itself is more important than external recognition, including flattery. The joy of work may be a powerful motivator, but even those who get the biggest joy from their work--- such as scientists, artists, and photographers --- crave flattery and recognition. Otherwise they wouldn’t compete for Nobel Prizes or enter their work in important exhibitions.

2017合工大研究生英语全缩印版

2017合工大研究生英语全缩印版第五章： 1.The serious part of my life ever since boyhood has been devoted to two different objects which for a long time remained separate and have only in recent years united into a single whole. 自少年时代起，我一生的重要时光都致力于实现两个不同的目标。这两个目标长久以来一直分开的，只是近几年才合为一体。 2. I was troubled by scepticism and unwillingly forced to the conclusion that most of what knowledge is open to reasonable doubt. 我被怀疑论干扰着，不愿意被迫下这个结论，那就是过去被迫接受的东西都可以公开的，合理地质疑。 3. But I discovered that many mathematical demonstrations, which my teachers expected me to accept, were full of , and that, if certainty were indeed discoverable in mathematics, it would be in a new kind of mathematics, with more solid foundations than those that had 但是我发现老师们期望我接受的许多数学论证都充满了谬误：假如确切性果真能发现于数学，它也是一种新型的数学，比迄今为止被认为是无懈可击的数学的原理更为牢固。 4. there was nothing more that I could do in the way of making mathematical knowledge . 在使数学知识不容怀疑这个方面，我已尽了全力。 5. And I am convinced that intelligence, patience, and can, sooner or later, lead the human its self-imposed itself meanwhile. 我相信：只要人类不自相残杀，智慧、忍耐和雄辩迟早会指引人类走出自找的磨难。 6 But I remain completely incapable of agreeing with those who accept fatalistically the view that man is born to trouble. 但是我仍然不能完全认同那些拥有宿命论观点认为人生下来就是麻烦的观点。 7. And there have been morbid miseries fostered by gloomy creeds,which have led men into profound inner discords that made all outward prosperity of avail . 还有一种就是由悲观的信念滋生的病态的痛苦，它使人类的内心极不和谐，从而使外部的繁荣毫无成果。 8.To preserve hope in our world our intelligence and our energy. In those who despair it is

研究生英语课件

The Hidden Side of Happiness 1.激流漂流：an recreational activity using an inflatable (可充气的）raft to navigate a river or other bodies of water. hurricane: a storm that has very strong fast winds and that moves over water 飓风 e.g. In America, Hurricane Katrina turned New Orleans from a stable, wealthy and vibrant city into a wasteland in a few days. typhoon 台风 monsoon 季风与季风相伴的雨季 trade wind 信风 vicious: violent and cruel in a way that hurts someone physically e.g. a vicious murder/killer/cycle ? A tiger wearing a monk's beads/a vicious person pretending to be benevolent 老虎戴素珠——假充善人 very unkind in a way that is intended to hurt someone's feelings or make their character seem bad, e.g. She was shocked by the vicious tone in his voice. similar words: malicious, evil, violent, savage, cruel, ferocious, fierce Their refrain might go sth. like this: refrain: n. a remark or idea that is often repeated 重复 e.g. （1）Their proposal met with constant refrain that it was impractical. （2）I made a terrific effort to refrain from tears. refrain: v. to hold oneself back; forbear e.g. refrain your tongue from backbiting. 不要在背后中伤人家 2 tribulation: a serious trouble or a serious problem 苦难，困苦，考验 ?Rejoicing in hope, patient in tribulation. (John Kennedy, American president) 从希望中得到欢乐，在苦难中保持坚韧。(美国总统肯尼迪) ?Even at this point in our awful tribulation under the German scourge we were quite friendly. 此刻，即便是处在德军蹂躏的极端痛苦中，我们仍然是十分友好的。

合肥工业大学研究生英语考试(上)课文复习翻译

HFUT研究生英语上考试课文重点翻译 Unit One 1. I have never cultivated a mustache, though I’m sure one would enhance my distinguished looks and cause women to giggle as I passed along the boulevard. 尽管我确信蓄胡子会使我更加气度不凡，走在大街上会使女性发笑，但我从不留胡子。 2. I wouldn’t know how to respond.I might be thrown into such a panic that I’d blurt out, ‖I like yours, too.‖ 我会无所适从，不知怎样回答才好。我可能会惊慌得脱口而出：―我也喜欢您的胡子。‖ 3. It is one of the paradoxes of social intercourse that a compliment is much harder to respond to than an insult. Here is an area of small talk where most of use act awkwardly. 在社会交往中，应对恭维比对付辱骂要艰难得多，这话听起来有点矛盾，却有一定的道理。闲聊时来句恭维话，往往让我们大多数人不知所措。 4. Someone utters a pleasing, praise remark in our direction and we grow inarticulate and our kneecaps begin to vibrate. 有人对我们说上一句动听、赞美的话，我们就慌得说不出话来，膝盖开始瑟瑟发抖。 5. I can’t even accept with grace a compliment bestowed upon me for a thing that isn’t really mine. 如果别人称赞不是真正属于我自己的东西时，我根本无法欣然接受。 6．The nearest I ever came to downright acceptance of this particular compliment was the time I said, ―Well , we like it.‖ This is a response that should be used with caution. 我在接受这种特定的恭维时，表示最能完全接受的说法就是―嗯，我们喜欢。‖采用这种答话必须得小心谨慎。 7. Not long ago I was in a group where a geophysicist form Australia was talking eloquently about the wonders of the universe. 不久前，我和一批人在一起时，其中有位来自澳大利亚的地球物理学家在滔滔不绝地谈论宇宙中的奇观。 8．There was a long pause, and then, carried away by the vastness of his complimentary remark, a woman said, ―Well, we like it.‖ 随后便是长时间的停顿。这时，一位被他的这种极度夸张的恭维话所吸引的妇女，禁不住说道，―嗯，我们喜欢这个地球。‖ 9. I think we make a mistake when we react to a compliment with denial and derogation. 我认为，对待恭维采取否定和贬低的态度是错误的。

Speech Enhancement Based on Spectral Estimation from Higher-lag Autocorrelation

最新合肥工业大学研究生英语复习课本重点句

合工大研究生英语翻译

合工大研究生英语课后汉翻英中英对照版考试必备(精)

2017合工大研究生英语全缩印版

研究生英语课件

合肥工业大学研究生英语考试(上)课文复习翻译