JPS6033752A - Noise level reduction system - Google Patents
Noise level reduction systemInfo
- Publication number
- JPS6033752A JPS6033752A JP14196783A JP14196783A JPS6033752A JP S6033752 A JPS6033752 A JP S6033752A JP 14196783 A JP14196783 A JP 14196783A JP 14196783 A JP14196783 A JP 14196783A JP S6033752 A JPS6033752 A JP S6033752A
- Authority
- JP
- Japan
- Prior art keywords
- noise
- time
- waveform
- waveforms
- periodic
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 230000009467 reduction Effects 0.000 title claims description 5
- 230000000737 periodic effect Effects 0.000 claims abstract description 27
- 238000005311 autocorrelation function Methods 0.000 claims abstract description 12
- 238000000034 method Methods 0.000 claims description 32
- 230000008569 process Effects 0.000 claims description 6
- 238000003672 processing method Methods 0.000 claims description 6
- 238000012545 processing Methods 0.000 abstract description 10
- 238000005070 sampling Methods 0.000 abstract description 5
- 230000003247 decreasing effect Effects 0.000 abstract 1
- 238000004364 calculation method Methods 0.000 description 3
- 238000004891 communication Methods 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 3
- 230000005236 sound signal Effects 0.000 description 3
- 238000012935 Averaging Methods 0.000 description 2
- JOJYUFGTMHSFEE-YONYXQDTSA-M Cytarabine ocfosphate Chemical compound [Na+].O[C@H]1[C@H](O)[C@@H](COP([O-])(=O)OCCCCCCCCCCCCCCCCCC)O[C@H]1N1C(=O)N=C(N)C=C1 JOJYUFGTMHSFEE-YONYXQDTSA-M 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000010354 integration Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000002238 attenuated effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000005314 correlation function Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 230000014509 gene expression Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L21/00—Speech or voice signal processing techniques to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
- G10L21/02—Speech enhancement, e.g. noise reduction or echo cancellation
- G10L21/0208—Noise filtering
Landscapes
- Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Quality & Reliability (AREA)
- Signal Processing (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- Acoustics & Sound (AREA)
- Multimedia (AREA)
- Noise Elimination (AREA)
Abstract
Description
【発明の詳細な説明】
本発明は、音声波形に重なった雑音のレベルを低減する
方式に関するものである。DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a method for reducing the level of noise superimposed on a speech waveform.
工場、工事現場などの騒音や、列車、航空機、自動車な
どの交通騒音の激しい環境で話された音声は、騒音のた
めに聞き取り難く、円滑な会話を行うことは困難である
。BACKGROUND ART Voices spoken in environments with heavy traffic noise, such as those of factories, construction sites, and trains, airplanes, and automobiles, are difficult to hear due to the noise, making it difficult to have a smooth conversation.
一方、無線通信による音声の伝送では、伝送途中におけ
る障害のためにＳ／Ｎが低下し、通話ができなくなるこ
とがある。On the other hand, when audio is transmitted by wireless communication, the S/N ratio may drop due to a failure during the transmission, making it impossible to make a call.
騒音環境でＳ／Ｎのよい音声を得るため、従来は主とし
て接話マイク、咽喉マイクなどのようにマイクロホンや
ピックアップ方式の開発に頼ってきた。しかし、これら
は騒音レベルの高いときに効果が不充分であったシ、原
理的によい品質の音声が得られないなどの問題がある。In order to obtain sound with a good S/N ratio in a noisy environment, conventional methods have mainly relied on the development of microphones and pickup methods, such as close-talk microphones and throat microphones. However, these methods have problems such as insufficient effectiveness when the noise level is high and, in principle, good quality audio cannot be obtained.
既に雑°音が重畳したり、ひずみの多い音声のＳ／Ｎを
改善して会話に対する疲労度を減らし、了解性を高める
ためには、信号源が音声であることを利用した信号処理
が必要であり、これらに関しては「鈴木誠史 ＳＳＳ／
Ｈの低い音声信号から雑音を減らす最近の信号処理技術
”１日経エレクトロニクス、Ｎｏ、２８１． 昭和５７
年１月４日」で紹介されている。しかし、紹介されてい
る多くの手法も、動作、性能、価格などに一長一短があ
るため、実用化されているものはない。In order to improve the S/N ratio of speech that is already overlaid with noise or has a lot of distortion, reduce conversational fatigue, and improve intelligibility, signal processing that takes advantage of the fact that the signal source is speech is necessary. Regarding these, “Masashi Suzuki SSS/
Recent signal processing technology to reduce noise from low H audio signals” 1 Nikkei Electronics, No. 281. 1977
It was introduced on January 4, 2017. However, many of the methods that have been introduced have advantages and disadvantages in terms of operation, performance, price, etc., so none have been put into practical use.
これらの中で、入力音声信号を短時間自己相関関数に変
換する音声処理方式Ｓ Ｐ Ａ Ｃ（ＳｐｅｅｃｈＰｒ
ｏｃｅｓｓｉｎｇ ｓｙｓｔｅｍ ｂｙ ｕｓｅ ｏｆ
Ａｕｔｏ−Ｃｏｒｒｅ−１ａ日ｏｎ ｆｕｎｃｔｉｏ
ｎ 、特許第１０４５１０２号、ＸＸ音声処理方式“、
昭和５６年５月２８日；鈴木誠史、５自己相関関数を利
用した音声処理方式Ｓ Ｐ Ａ Ｃ／／。Among these, the speech processing method SPAC (Speech Pr
ocessing system by use of
Auto-Corre-1a day on function
n, Patent No. 1045102, XX audio processing method",
May 28, 1981; Masashi Suzuki, Speech processing method using 5 autocorrelation functions SPAC//.
電子通信学会、技術研究報告ＥＡ７５−２５．昭和５０
年７月）は、評価実験も行われ、ハードウェアも試作さ
れていて有望々方法である。しかし、５ＰＡＣは、ラン
ダム性雑音の低減能力は大きいが、周期波形を強調する
性質をもっているため、周期性雑音に対しては効果がな
い。Institute of Electronics and Communication Engineers, Technical Research Report EA75-25. Showa 50
(July 2016), evaluation experiments have been conducted, and hardware has been prototyped, so it is a promising method. However, although 5PAC has a large ability to reduce random noise, it has a property of emphasizing periodic waveforms and is therefore ineffective against periodic noise.
本発明は、音声波形の１周期ごとに短時間自己相関関数
をめ、その相関関数の１周期の波形を接続して音声波形
を合成する過程において、短時間自己相関関数を平均し
た波形を差し引くことが特徴であって、その目的は雑音
が加わった音声波形信号からランダム性雑音とともに周
期性雑音のレベルを低減することである。The present invention calculates a short-time autocorrelation function for each period of a speech waveform, and in the process of connecting the waveforms of one period of the correlation function to synthesize a speech waveform, subtracts the waveform obtained by averaging the short-time autocorrelation functions. Its purpose is to reduce the level of both random noise and periodic noise from a speech waveform signal to which noise has been added.
以下、図に従って本発明を説明する。第１図は本発明の
方式の原理を説明するだめの、種々の波形とその短時間
自己相関数の関係の例、第２図は本発明の方式の処理過
程の流れ図、第３図は第２図の各段階の波形を示した図
である。第１図で１は音声波形（母音）、３はランダム
雑音波形、５最初に、第１図によって種々の波形とその
短時間自己相関関数Ｓの関係を説明する。■は母音のよ
うガ周期的音声波形であり、１のＳは２のようになり、
１と２の周波数成分は等しい。力お、標本化周期Ｉで標
本化された入力信号をａｌとすると、Ｓは（１）式で計
算される。The present invention will be explained below with reference to the drawings. Fig. 1 is an example of the relationship between various waveforms and their short-time autocorrelation numbers, which is useful for explaining the principle of the method of the present invention, Fig. 2 is a flowchart of the processing process of the method of the present invention, and Fig. 3 is a flowchart of the processing process of the method of the present invention. FIG. 2 is a diagram showing waveforms at each stage in FIG. 2; In FIG. 1, 1 is a speech waveform (vowel), 3 is a random noise waveform, and 5 is a random noise waveform. First, the relationship between various waveforms and their short-time autocorrelation function S will be explained with reference to FIG. ■ is a periodic speech waveform like a vowel, and S in 1 becomes like 2,
The frequency components of 1 and 2 are equal. Furthermore, if the input signal sampled at the sampling period I is set to al, then S is calculated using equation (1).
Ｊは遅延時間に対応する標本番号、積分の時間はＮ１で
ある。ところで、ランダム雑音３のＳは４のようになり
、Ｎを大きくするとインパルスに近づく。また、周期性
雑音を５とすると、どの時刻に関してもそのＳは６であ
る。J is the sample number corresponding to the delay time, and the integration time is N1. By the way, S of random noise 3 becomes 4, and as N becomes larger, it approaches an impulse. Further, if the periodic noise is 5, its S is 6 at any time.
ところで、Ｓ ＰＡＣでは、準周期的とみなされる音声
波形のＳは２のような波形になるが、ランダム雑音のＳ
が４のようなインパルス状になることに着目し、」＝０
近傍を除いてＳの１周期の波形を切り出して出力信号と
している。その結果、１に３が加わった波形の場合、ラ
ンダム雑音成分を大幅に減少した出力信号を得ている。By the way, in S PAC, the S of the speech waveform that is considered to be quasi-periodic is a waveform like 2, but the S of random noise is
Paying attention to the fact that it becomes an impulse like 4, ``=0
One cycle of the waveform of S is cut out, excluding the vicinity, and used as an output signal. As a result, in the case of a waveform in which 3 is added to 1, an output signal with significantly reduced random noise components is obtained.
一方、１に５が加わった場合、そのＳは２と６の和にな
り、５の成分を除くことはできない。しかし、音声信号
な長時間観測すると定常な部分はわずかであり、波形を
区切って重ね合わせ平均すると、３のようなランダム雑
音に近くなることはよく知られている。しだがって、音
声のＳを数多く平均すると、４に近くなることが期待で
きる。一方、５のような周期信号のＳは、常に同じ形を
しているので、平均してもその形は変わらない。そこで
、１周期ごとに計算されるＳから、Ｓを平均した波形Ｓ
を差し引くと、Ｓに含まれている周期性雑音成分を大幅
にとり除くことができる。On the other hand, when 5 is added to 1, S becomes the sum of 2 and 6, and the 5 component cannot be removed. However, it is well known that when an audio signal is observed for a long time, there are only a few stationary parts, and when the waveforms are segmented and averaged over each other, it becomes close to random noise like 3. Therefore, if you average S of many voices, you can expect it to be close to 4. On the other hand, since S of a periodic signal such as 5 always has the same shape, its shape does not change even if it is averaged. Therefore, from the S calculated every cycle, the waveform S which is the average of S
By subtracting , the periodic noise component included in S can be largely removed.
本発明の方式を第２図の流れ図と、第３図の波形によシ
説明する。第２図で、入力には音声波形１と周期性雑音
波形５が加わった波形１０が、標本化周期■の時系列信
号として加えられる。１０に関して、時刻ｔ１を基準に
して（１）式により短時間自己相関関数Ｓ＋（Ｊ）１１
を計算する。なお、（１）式では、ｔｌに対応する標本
を２１としている。次に １１から５（Ｊ）を長時間に
わたって平均した５１（Ｄ １２を減じ、Ｇｌ （ｊ）
１３をめる。この処理の結果、１３では１１に含まれ
ていた周期性雑音成分がほとんど取り除かれている。次
に１３の極大値をめ、その標本の遅延時間から周期Ｉｌ
ｌ、を決定し、１３から１゛１に相当する標本を切り出
して出力信号１６とする。なお、この標本の切出しに際
しては、」＝０近傍の標本は使用しない。第２図では、
ＪがＯに近く、振幅が０から立ち上る標本から、１周期
の標本を切り出している。これは、１０に含捷れるラン
ダム雑音成分が、４で示したようにＪ−０の付近に集ま
り、その影響が１３にも残る可能性があるので行う処理
である。The method of the present invention will be explained with reference to the flowchart of FIG. 2 and the waveforms of FIG. 3. In FIG. 2, a waveform 10 in which a speech waveform 1 and a periodic noise waveform 5 are added is added to the input as a time-series signal with a sampling period ■. 10, the short-time autocorrelation function S+(J)11 is calculated using equation (1) based on time t1.
Calculate. Note that in equation (1), the sample corresponding to tl is set to 21. Next, 51 (D, which is the average of 5 (J) from 11 over a long period of time, subtract 12, and Gl (j)
Get 13. As a result of this processing, in 13, most of the periodic noise components contained in 11 have been removed. Next, find the maximum value of 13 and calculate the period Il from the delay time of that sample.
1 is determined, and a sample corresponding to 1゛1 is cut out from 13 and used as an output signal 16. Note that when cutting out this sample, samples near "=0" are not used. In Figure 2,
A sample of one period is cut out from a sample where J is close to O and the amplitude rises from 0. This process is performed because the random noise components included in 10 may gather near J-0 as shown in 4, and their influence may remain on 13 as well.
次に、１０について、時刻ｔ２（ｔ２＝’ｌ ＋ＴＩ
）を基準にして短時間自己相関関数８２（Ｊ）１４を計
算する。１４からＳ （Ｄを長時間にわたって平均した
５２（Ｄ ； （１２とほとんど同じである。）を減じ
、Ｇｚ（Ｄ１５をめる。１５から周期Ｔ２を決定し、Ｔ
２に相当する標本を切り出し、１３がら切り出しだ標本
に接続して出力信号１６とする。次に、１ｏについて、
時刻ｔ３（ｔ３＝ ｔ２＋Ｔ２ ）を基準にして同様の
処理を行う。この操作を反復することにより、１６には
１０の中に含まれているランダム雑音と周期性雑音成分
を大幅に低減した信号を得ることができる。Next, for 10, time t2 (t2='l +TI
) is used as a reference to calculate the short-time autocorrelation function 82(J)14. From 14, subtract 52 (D; which is almost the same as 12), which is the average of D over a long period of time, and subtract Gz (D15. From 15, determine the period T2, and calculate T
A sample corresponding to 2 is cut out and connected to the cut out sample 13 to provide an output signal 16. Next, regarding 1o,
Similar processing is performed using time t3 (t3=t2+T2) as a reference. By repeating this operation, it is possible to obtain a signal in which the random noise and periodic noise components contained in signal 10 are significantly reduced.
以上の説明は有声音に対するものであるが、無声音の場
合は５（Ｊ）の極大値から適当に周期を定め、有声音と
同様の処理を行えばよい。Although the above explanation is for voiced sounds, in the case of unvoiced sounds, the period may be determined appropriately from the maximum value of 5 (J) and the same processing as for voiced sounds may be performed.
なお、本発明の方式において、■は標本化定理を満足す
るように設定する。まだ、積分の時間Ｎ１’ ｕ ２０
ｍｓ前後、ＬＩモｚＱｍｓ程ｉＫ＆ル、Ｊ：つＫ、Ｎ
やＬを設定すればよい。Note that in the method of the present invention, (2) is set so as to satisfy the sampling theorem. There is still time for integration N1' u 20
Around ms, LIMozQms, iK&L, J:tsuK,N
or L may be set.
一方、Ｓ （Ｄの長時間平均値Ｓ （Ｄには幾つかの計
算法が考えられる。第１の方式は、時刻ｔｋに関して（
２）式でＳ （Ｄをめ、処理を行う。On the other hand, several calculation methods can be considered for the long-term average value S (D of S (D). The first method is that (
2) Determine S (D) in the equation and perform the process.
この方式は、常にｔｋよりも過去のＭ個のＳ （ｊ）の
平均値を使用する。これは移動平均であり、周期性雑音
がゆるやかに変動する場合でも、比較的容易に追従する
ことができる。ただし、平均操作により音声のみのＳを
４のような形にするためにはＭを大きくとる必要がある
。例えばＭ＝ ２００ とすると、常に２００のＳ （
Ｄを記憶するメモリが必要である。This method always uses the average value of M S (j) past tk. This is a moving average and can be tracked relatively easily even if the periodic noise fluctuates slowly. However, in order to make the voice-only S into a shape like 4 by averaging, it is necessary to set M large. For example, if M = 200, then S (
A memory is required to store D.
ところで、周期性雑音が定常な場合、あるいは極めてゆ
るやかに変化する場合は、Ｍ個の平均のＳ （Ｄを、次
のＭ周期の間使用しても実用上差し支えない。換言する
と、時刻ｉｐに関して （３）式％式％（
を計算する。５ｋ（Ｊ）からは、ｋに最も近いｐ（ｋ≧
ｐ）のＳｐ （Ｄを減じて処理を行う。これを第２の方
式とする。この方式は、第１の方式の場合よシも、周期
性雑音の変動に対する追従が悪く、また、処理を開始し
て最初のＭ周期の時間が経過するまで、周期性雑音の低
減は行われ々いが、２種類の５ｐ（ｊ） ；現在使って
いるものと次に使うために計算中のものを記憶しておく
だけでよい。By the way, if the periodic noise is stationary or changes very slowly, there is no practical problem in using M averages of S(D) for the next M periods.In other words, with respect to time ip, (3) Formula % Formula % ( Calculate. From 5k (J), p (k≧
Processing is performed by subtracting Sp (D) of p).This is the second method.This method has poor tracking of periodic noise fluctuations than the first method. Periodic noise is not reduced until the first M periods have elapsed, but two types of 5p(j) are used: the one currently used and the one being calculated for next use. Just remember it.
第３の方式として、（２）式と（３）式の茗（」）の中
間的な平均値として、時定数をもった荷重平均が考えら
れる。これを（４）式で示す。As a third method, a weighted average with a time constant can be considered as an intermediate average value between the expressions (2) and (3). This is shown by equation (4).
Ｓｋ （ｉ） ＝ Ｂ ：Ａ −８ｋ−ｍ （Ｄｍに１
−Ｂ （５ｋ−１（Ｄ＋Ａ−百に−＋ （Ｄ ） （４
）ここで、Ａは減衰定数（０＜Ａ＜ １ ）、Ｂは周期
信号のときに、Ｓ （ＤとＳ （Ｄを等しくするだめの
定数である。このＳ （Ｄでは、時間の経過とともに古
いＳ （Ｄの寄与が小さくなり、周期性雑音の変動にも
追従しゃすい。（４）式の計算には、毎回積和計算を必
要とするが、メモリは（３）式を使用する方式と同様で
少ない。Sk (i) = B :A -8k-m (1 to Dm -B (5k-1(D+A-100-+ (D) (4
) Here, A is the attenuation constant (0<A<1), and B is a constant to make S (D and S (D) equal when the signal is periodic. In this S (D, as time passes, The contribution of the old S (D becomes smaller and it is easier to follow the fluctuations of periodic noise. Calculation of equation (4) requires sum-of-products calculation every time, but the memory method uses equation (3). Same as and less.
次に本発明の実施例について述べる。３．４ ｋＨｚに
帯域制限した男声に、妨害波として８００Ｈｚの正弦波
を加え、Ｓ／Ｎがは′ぼＯｄＢになるような資料を作成
した。この資料を標本化周期０．１ ｍｓで標本化し、
第２図の方法で処理を行った。Next, examples of the present invention will be described. I created a material in which an 800Hz sine wave was added as an interference wave to a male voice whose band was limited to 3.4kHz, and the S/N ratio was almost OdB. This material was sampled at a sampling period of 0.1 ms,
The treatment was carried out using the method shown in FIG.
（１）式のＮは２００、Ｌは１７０である。（２）式を
使用する第１の方式で、Ｍを３０から２５０″！！で変
えて出力信号をめたところ、いずれの場合も妨害波の８
００Ｉ−１ｚはほとんど完全に除去され検知できなくな
った。ただ、習には音声のＳの平均値が成分として含ま
れ、これはＭを無限大にすれば０になるものである。本
実施例のようにＭが有限の場合は雑音となり、検知され
るが入力信号に比較するとはるかにききやすい。Ｍが２
００程度（平均の周期はｓｍｓ位）のとき、出力音声信
号に対してこの雑音レベルは−１３〜−１４ｄＢであり
、実質的な妨害の度合は少ない。なお、更にＭを大きく
とれば雑音レベルは低くなるが、妨害波の周期波を除去
するまでに時間がかかるため、通常は１〜３秒の区間で
百をめるようにＭを設定すればよい。In formula (1), N is 200 and L is 170. In the first method using formula (2), we measured the output signal by changing M from 30 to 250''!!
00I-1z was almost completely removed and could no longer be detected. However, Xi includes the average value of S of the voice as a component, and this becomes 0 if M is set to infinity. When M is finite as in this embodiment, it becomes noise and is detected, but it is much easier to hear than the input signal. M is 2
00 (average period is about SMS), this noise level is -13 to -14 dB with respect to the output audio signal, and the degree of substantial interference is small. Note that if M is made larger, the noise level will be lowered, but since it takes time to remove the periodic wave of the interference wave, normally it is best to set M to the nearest hundred in an interval of 1 to 3 seconds. good.
次に（３）式を使用する第２の方式について同様の実験
を行った。この場合、最初のＳρが計算される（Ｍ周期
の時間が経過する。）丑で、周期性雑音の減少は行われ
ないが、その後は第１の方式と同様の効果をあげた。Next, a similar experiment was conducted regarding the second method using equation (3). In this case, the periodic noise was not reduced when the first Sρ was calculated (time of M periods elapsed), but thereafter the same effect as the first method was obtained.
次に（４）式を使用する第３の方式について同様の実験
を行った。ここで、Ａ二０．９８、Ｂ＝１７４９とした
ところ、周期性雑音は徐々に減衰し、約２秒経過してか
らは、はとんど検知できなくなった。その後は、第１の
方式と同様の効果をあげた。Next, a similar experiment was conducted regarding a third method using equation (4). Here, when A2 was set to 0.98 and B was set to 1749, the periodic noise gradually attenuated and became almost undetectable after approximately 2 seconds. After that, the same effect as the first method was obtained.
以上のように、本発明によれば、従来のランダム性雑音
の低減に加えて、音声信号に加わった周期性雑音を大幅
に低減することができるから、本発明は騒音環境下の音
声通信や、無線通信における通話品質の改善に広く利用
することができる。As described above, according to the present invention, in addition to the conventional reduction of random noise, periodic noise added to voice signals can be significantly reduced. , can be widely used to improve call quality in wireless communications.
第１図と第３図は本発明の詳細な説明する波形、第２図
は本発明の実施例の流れ図である。
特許出願人 郵政省電波研究所長
…７１桔号
第 ２ 図FIGS. 1 and 3 are waveforms illustrating detailed explanations of the present invention, and FIG. 2 is a flowchart of an embodiment of the present invention. Patent applicant Director of Radio Research Institute, Ministry of Posts and Telecommunications…71 No. 2 Figure
Claims (3)
Ｓに変換し、Ｓの１周期の波形を逐次接続して出力信号
とする音声処理方式において、過去のＭ（Ｍ≧２の整数
）個のＳの平均値をＳから減じてから波形を接続するこ
とによシ、入力音声波形信号に重なったランダム雑音成
分と周期性雑音成分を大幅に減少した音声波形信号を得
ることを特徴とする雑音レベル低減方式。(1) In an audio processing method in which an audio waveform signal is converted into a short-time autocorrelation function S every cycle, and the waveforms of one cycle of S are successively connected to produce an output signal, the past M (an integer of M≧2) is used. ) by subtracting the average value of S from S and then connecting the waveforms, a voice waveform signal with significantly reduced random noise components and periodic noise components superimposed on the input voice waveform signal is obtained. Noise level reduction method.
Ｓに変換し、Ｓの１周期の波形を逐次接続して出力信号
とする音声処理方式において、Ｍ個のＳの平均値をめ、
この平均値を次のＭ個のＳからそれぞれ減じてから波形
を接続して出力信号とし、この過程をＭ周期ごとに反復
することにより、入力音声波形信号に重なったランダム
雑音成分と周期性雑音成分を大幅に減少した音声波形信
号を得ることを特徴とする雑音レベル低減方式。(2) In an audio processing method in which an audio waveform signal is converted into a short-time autocorrelation function S every cycle, and the waveforms of one cycle of S are successively connected to produce an output signal, the average value of M S is calculated. ,
This average value is subtracted from each of the next M S's, and the waveforms are connected to obtain the output signal. By repeating this process every M periods, random noise components and periodic noise superimposed on the input speech waveform signal are A noise level reduction method characterized by obtaining an audio waveform signal with significantly reduced components.
Ｓに変換し、Ｓの１周期の波形を逐次接続して出力信号
とする音声処理方式において、ｍ周期（ｍ≧１）前のＳ
に、（ｍ−１）回係数Ａ（０＜Ａ＜１ ）を乗じて累積
加算してめたＳの荷重平均値を、Ｓがら減じてから波形
を接続することによシ、入力音声波形に重なったランダ
ム雑音成分と周期性雑音成分を大幅に減少した音声波形
信号を得ることを特徴とする雑音レベル低減方式。(3) In an audio processing method in which an audio waveform signal is converted into a short-time autocorrelation function S every cycle, and the waveforms of one cycle of S are sequentially connected to produce an output signal, S
By multiplying (m-1) times by a coefficient A (0<A<1) and cumulatively adding the weighted average value of S, subtracting it from S and then connecting the waveforms, the input speech waveform can be obtained. A noise level reduction method characterized by obtaining a speech waveform signal with significantly reduced random noise components and periodic noise components superimposed on the noise.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| JP14196783A JPS6033752A (en) | 1983-08-04 | 1983-08-04 | Noise level reduction system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| JP14196783A JPS6033752A (en) | 1983-08-04 | 1983-08-04 | Noise level reduction system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| JPS6033752A true JPS6033752A (en) | 1985-02-21 |
| JPH0336346B2 JPH0336346B2 (en) | 1991-05-31 |
Family
ID=15304287
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| JP14196783A Granted JPS6033752A (en) | 1983-08-04 | 1983-08-04 | Noise level reduction system |
Country Status (1)
| Country | Link |
|---|---|
| JP (1) | JPS6033752A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2016099312A1 (en) * | 2015-02-10 | 2016-06-23 | Szechniuk Sławomir | Phase filter and method for interference and noise reduction in systems with two signal paths. |
| US9384756B2 (en) | 2014-03-07 | 2016-07-05 | JVC Kenwood Corporation | Cyclic noise reduction for targeted frequency bands |
-
1983
- 1983-08-04 JP JP14196783A patent/JPS6033752A/en active Granted
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9384756B2 (en) | 2014-03-07 | 2016-07-05 | JVC Kenwood Corporation | Cyclic noise reduction for targeted frequency bands |
| WO2016099312A1 (en) * | 2015-02-10 | 2016-06-23 | Szechniuk Sławomir | Phase filter and method for interference and noise reduction in systems with two signal paths. |
| US10158386B2 (en) | 2015-02-10 | 2018-12-18 | Szechniuk Sławomir | Phase filter and method for interference and noise reduction in systems with two signal paths |
Also Published As
| Publication number | Publication date |
|---|---|
| JPH0336346B2 (en) | 1991-05-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109065067B (en) | Conference terminal voice noise reduction method based on neural network model | |
| US8010355B2 (en) | Low complexity noise reduction method | |
| KR0175965B1 (en) | Transmitted noise reduction in communications systems | |
| US7231347B2 (en) | Acoustic signal enhancement system | |
| Macho et al. | Evaluation of a noise-robust DSR front-end on Aurora databases. | |
| CN101027719B (en) | Noise suppressor | |
| US20050288923A1 (en) | Speech enhancement by noise masking | |
| Lin et al. | Adaptive noise estimation algorithm for speech enhancement | |
| US20050108004A1 (en) | Voice activity detector based on spectral flatness of input signal | |
| KR20030074762A (en) | Noise reduction method and device | |
| US20100169082A1 (en) | Enhancing Receiver Intelligibility in Voice Communication Devices | |
| CN109102823B (en) | Speech enhancement method based on subband spectral entropy | |
| US8165872B2 (en) | Method and system for improving speech quality | |
| Sakhnov et al. | Dynamical energy-based speech/silence detector for speech enhancement applications | |
| JPS6033752A (en) | Noise level reduction system | |
| Abutalebi et al. | A hybrid subband adaptive system for speech enhancement in diffuse noise fields | |
| JP3310225B2 (en) | Noise level time variation calculation method and apparatus, and noise reduction method and apparatus | |
| Habets et al. | Speech dereverberation using backward estimation of the late reverberant spectral variance | |
| JP2001249676A (en) | Method for extracting fundamental period or fundamental frequency of periodical waveform with added noise | |
| CN114341978A (en) | Noise reduction in headset using voice accelerometer signals | |
| Erkelens et al. | Single-microphone late-reverberation suppression in noisy speech by exploiting long-term correlation in the DFT domain | |
| Abutalebi et al. | Speech dereverberation in noisy environments using an adaptive minimum mean square error estimator | |
| Yegnanarayana | Effect of noise and distortion in speech on parametric extraction | |
| Shindo et al. | Impulsive noise reduction in speech acquisition based on throat vibration measurement with laser microphone | |
| KELAGADI et al. | REDUCTION OF ENERGY FOR IOT BASED SPEECH SENSORS IN NOISE REDUCTION USING MACHINE LEARNING MODEL. |