JP4345208B2 - Reverberation and noise removal device - Google Patents

Description

を計算し、
ＰＸＦ（ｌ）＞Ｖ・ＰＮＦ（ｌ） …（１２）
のとき「音声あり」として検出する。
【００６６】
「音声あり」のとき、周波数軸音声検出部３８はＡＤＦ部３４の係数更新を停止する。
【００６７】
また「音声あり」のとき、ＦＦＴ演算部３２に式（９）を実行するための信号を出力する。
【００６８】
あるいはまた、特定の周波数、または周波数帯を無視して式（１０）、式（１１）を実施するようにしてもよい。これは非定常ノイズが特定の周波数（周波数帯域）に集中して存在することがわかっているときには、これらのノイズを音声と誤検出するのを防止するのに役立つ。即ち、第２の実施形態に加えて更に精度良く音声の検出を行うことで背景ノイズの推定を正確に行うことができる。
【００６９】
たとえば、式（１１）、式（１２）において特定のｋ、またはいくつかのｋの集まりを計算に反映させないようにする。
【００７０】
（Ｃ−３）第３の実施形態の効果
以上、述べたように、第３の実施形態では音声検出器の位置がＦＦＴ演算部３２と逆ＦＦＴ演算部３３との間において、周波数領域で音声の検出を行うようにしたので、より細かな音声検出を行ってノイズ推定を行い、ノイズを除去することができる。
【００７１】
（Ｄ）第４の実施形態
以下、本発明による反響及び雑音除去装置の第４の実施形態を図面を参照しながら詳述する。ここで、図５は、この実施例の構成を示すものであり、上述した図１との同一、対応部分には同一符号を付して示して、その詳細な説明を省略する。
【００７２】
（Ｄ−１）第４の実施形態の構成
図５は、第４の実施形態の反響及び雑音除去装置の構成を示すブロック図である。第４の実施形態は第１の実施形態をもとにしており、第４の実施形態が第１の実施形態と異なるのは、ＡＤＦ部５２の前に加算器１７が設けられた一方、第１の実施形態におけるＡＤＦ部３４、加算器３が削除されていることである。
【００７３】
以下、図５を用いて第４の実施形態の構成を説明する。
【００７４】
マイクロホン１から逆ＦＦＴ演算部３３までの処理は第１の実施形態で説明したものと同じであるのでここでは説明を省略する。
【００７５】
逆ＦＦＴ演算部３３の出力は加算部５３に出力される。加算部５３には受信側入力Ｒin入力端子９からの信号も入力されている。一方でＦＦＴ演算部３２には図示しないエコーキャンセラ部５のダブルトーク検出器から音声検出信号が入力されている。
【００７６】
第４の実施形態ではＡＤＦ部５２、ＦＦＴ演算部３２の動作が異なっている。即ち、ＡＤＦ部５２はダブルトーク検出器からの音声検出結果によって以下のように係数更新を行う。
【００７７】
受信側入力Ｒin側のみに音声信号あり：ＦＦＴ演算部３２のノイズ更新（９）式（即ち停止）、ＡＤＦ部５２の係数更新
マイクロホン入力Ｓin側のみに音声信号あり：ＦＦＴ演算部３２のノイズ更新（９）式、ＡＤＦ部５２の係数更新停止
受信側入力Ｒin側、マイクロホン入力Ｓin側の双方に音声信号あり：ＦＦＴ演算部３２のノイズ更新（９）式（即ち停止）、ＡＤＦ部５２の係数更新停止
受信側入力Ｒin側、マイクロホン入力Ｓin側の双方に音声信号なし：ＦＦＴ演算部３２のノイズ更新（４）式、ＡＤＦ部５２の係数更新停止
即ち、逆ＦＦＴ演算部３３の出力信号をエコーキャンセラ部５の擬似エコー作成のための参照信号と同様に扱ってよく、これによりハードウェアを大幅に小規模化することができる。
【００７８】
この際、逆ＦＦＴ演算部３３は常に動作して時間軸信号を作成し、ＡＤＦ部５２に出力し続ける。
【００７９】
（Ｄ−３）第４の実施形態の効果
以上説明したように、第４の実施形態によれば逆ＦＦＴ演算部３３の出力をエコーキャンセラ用ＡＤＦ部５２に直接入力し、ＡＤＦ部５２の係数更新制御方法を
受信側入力Ｒin側のみに音声信号あり：ＦＦＴ演算部３２のノイズ更新（９）式（即ち停止）、ＡＤＦ部５２の係数更新
マイクロホン入力Ｓin側のみに音声信号あり：ＦＦＴ演算部３２のノイズ更新（９）式、ＡＤＦ部５２の係数更新停止
受信側入力Ｒin側、マイクロホン入力Ｓin側の双方に音声信号あり：ＦＦＴ演算部３２のノイズ更新（９）式（即ち停止）、ＡＤＦ部５２の係数更新停止
受信側入力Ｒin側、マイクロホン入力Ｓin側の双方に音声信号なし：ＦＦＴ演算部３２のノイズ更新（４）式、ＡＤＦ部５２の係数更新停止
とするようにしたので逆ＦＦＴ演算部３３の出力信号をエコーキャンセラ部５の擬似エコー作成のための参照信号と同様に扱ってよく、これによりハードウェアを大幅に小規模化することができる。
【００８０】
（Ｅ）第５の実施形態
以下、本発明による反響及び雑音除去装置の第５の実施形態を図面を参照しながら詳述する。ここで、図６は、この実施例の構成を示すものであり、上述した図１との同一、対応部分には同一符号を付して、その説明を省略する。
【００８１】
（Ｅ−１）第５の実施形態の構成
図１は、第５の実施形態の反響及び雑音除去装置の構成を示すブロック図である。第５の実施形態は第２の実施形態と第４の実施形態をもとにしている。時間軸音声検出器の動作は第２の実施形態で説明したのでここでは説明を省略する。また、ＦＦＴ演算部３２、逆ＦＦＴ演算部３３の動作は第４の実施形態で説明したとおりである。
【００８２】
ＡＤＦ部５２の制御は
受信側入力Ｒin側のみに音声信号あり：ＡＤＦ部５２の係数更新
マイクロホン入力Ｓin側のみに音声信号あり：ＡＤＦ部５２の係数更新停止
受信側入力Ｒin側、マイクロホン入力Ｓin側の双方に音声信号あり：ＡＤＦ部５２の係数更新停止
受信側入力Ｒin側、マイクロホン入力Ｓin側の双方に音声信号なし：ＡＤＦ部５２の係数更新停止
ＦＦＴ演算部３２のノイズ更新制御は
音声検出器からの検出結果 音声信号あり：ＦＦＴ演算部３２のノイズ更新（９）式（即ち停止）
音声検出器からの検出結果 音声信号なし：ＦＦＴ演算部３２のノイズ更新（４）式（即ち更新）
とする点が、第４の実施形態と異なっている。
【００８３】
即ち、逆ＦＦＴ演算部３３の出力信号をエコーキャンセラ部５の擬似エコー作成のための参照信号と同様に扱ってよく、これによりハードウェアを大幅に小規模化することができる。
【００８４】
この際、逆ＦＦＴ演算部３３は常に動作して時間軸信号を作成し、ＡＤＦ部５２に出力し続ける。
【００８５】
また第２の実施形態で述べたように、音声検出をディレイ部３５の前で行えるので、音声の検出がより迅速に行える。結果ＦＦＴ演算部３２のノイズ推定に与えるダメージを小さくすることができるので、より精度の良いノイズ除去を行える。
【００８６】
（Ｅ−３）第５の実施形態の効果
以上説明したように、第５の実施形態によれば、第４の実施形態に加え、より迅速に音声の検出を行い、結果、ＦＦＴ演算部３２のノイズ推定に与えるダメージを小さくすることができるのでより精度の良いノイズ除去を行える。更に第１の実施形態よりもＡＤＦ部３４を省くことでハードウェアの小規模化を実現できる。
【００８７】
（Ｆ）第６の実施形態
以下、本発明による反響及び雑音除去装置の第６の実施形態を図面を参照しながら詳述する。ここで、図７は、この実施例の構成を示すものであり、上述した図１との同一、対応部分には同一符号を付して、その説明を省略する。
【００８８】
（Ｆ−１）第６の実施形態の構成
図７は、第６の実施形態の反響及び雑音除去装置の構成を示すブロック図である。第６の実施形態は第３の実施形態と第４の実施形態をもとにしている。周波数領域音声検出部３８の動作は第３の実施形態で説明したのでここでは説明を省略する。
【００８９】
第６の実施形態は第３の実施形態と異なり、周波数領域音声検出部３８の出力はＦＦＴ演算部３２にのみ出力される。ＦＦＴ演算部３２では第３の実施形態と同様にノイズの更新を行い、逆ＦＦＴ演算部３３と共に動作して時間軸での擬似ノイズを作成する。
【００９０】
エコーキャンセラ部５のＡＤＦ部５２の制御は第４の実施形態で説明したと同じ動作でＡＤＦ部５２の係数を更新する。ＦＦＴ演算部３２とＡＤＦ部５２の更新の制御の方法は第５の実施形態で説明したと同様である。
【００９１】
（Ｆ−３）第６の実施形態の効果
以上説明したように、第６の実施形態によれば、音声検出を迅速に行い、更に周波数領域で音声の検出を行うようにしたのでより細かな音声検出を行ってノイズ推定を行い除去し、かつ第４の実施形態と同様に、ハードウェアを大幅に小規模化することができる。
【００９２】
なお、上記の実施形態ではマイクロホンとスピーカを使用した、いわゆる「スピーカホン」を例に説明したが、本発明はこれに限定されること無く、例えばマイクロホンではなく、電気的な入力であっても良く、また同様にスピーカによる電気／音響変換が不要な系では、電気信号をそのまま出力するような系であってもよいのはもちろんである。
【００９３】
【発明の効果】
以上のように、本発明の反響及び雑音除去装置によれば、ノイズを時間軸で適応的に除去することにより、ノイズ除去に伴なう音質の違和感を生じせしめること無く、かつエコーキャンセラの性能が劣化すること無く、エコーのない自然な音声を送出することができる。
【図面の簡単な説明】
【図１】本発明が適用される第１の実施形態の概略構成を示すブロック図である。
【図２】自動車エンジンの周波数特性を示す周波数特性図である。
【図３】本発明が適用される第２の実施形態の概略構成を示すブロック図である。
【図４】本発明が適用される第３の実施形態の概略構成を示すブロック図である。
【図５】本発明が適用される第４の実施形態の概略構成を示すブロック図である。
【図６】本発明が適用される第５の実施形態の概略構成を示すブロック図である。
【図７】本発明が適用される第６の実施形態の概略構成を示すブロック図である。
【符号の説明】
１…マイクロホン、３…ノイズキャンセラ部、５…エコーキャンセラ部、７…スピーカ、３１…Ａ／Ｄ変換部、３２…ＦＦＴ演算部、３３…逆ＦＦＴ演算部、３４…ＡＤＦ部、３５…ディレイ部、３６…加算器、３７…音声検出部、３８…音声検出部、５１…加算器、５２…ＡＤＦ部、５３…加算器。[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a reverberation and noise removal apparatus, and more particularly to a reverberation and noise removal apparatus suitable for application to an in-vehicle hands-free automobile telephone or the like.
[0002]
[Prior art]
In recent years, in-vehicle hands-free automobile telephones have become widespread, and along with this, various apparatuses for removing echo and noise in such hands-free automobile telephones have been proposed.
[0003]
An echo canceller disclosed in Japanese Patent Laid-Open No. 10-243082 is known as an apparatus for removing such echo and noise. The technology described in this publication is made in order to eliminate the fact that the circuit configuration of the noise canceller disposed in front of the echo canceller is substantially the same as that of the echo canceller, and therefore the overall circuit configuration is complicated. By adding the output signal of the band elimination filter that extracts noise other than the transmission voice band from the transmission signal that transmits the estimated echo in the echo canceller to the received signal, the amount of echo suppression is reduced by the influence of ambient noise. Noise is reduced and communication quality is ensured without lowering.
[0004]
[Problems to be solved by the invention]
However, in the conventional apparatus described above, noise is removed using a band removal filter as pre-processing of the echo canceller. Therefore, when the voice band and the noise band overlap, the noise is removed at the same time as the noise is removed. Therefore, there has been a problem that deterioration of voice is inevitable.
[0005]
In addition, in order to avoid such drawbacks, when a well-known frequency subtraction method (spectral subtraction method) is used instead of a band removal filter, noise subtraction processing in the frequency domain is nonlinear. It may be a process.
[0006]
In general, it is well known that the echo canceller has a characteristic that the amount of echo cancellation greatly deteriorates if a non-linear path is included in the echo path. There has been a problem of deterioration.
[0007]
The present invention has been made in consideration of the above points, and provides an echo and noise removal device that does not cause a sense of incongruity in sound quality due to noise removal by adaptively removing noise on a time axis. It is something to try.
[0008]
[Means for Solving the Problems]
  In order to solve this problem, the first present inventionEcho and noise removal deviceOutputs a received signal as an audio signal from a speaker and outputs an audio signal input from a microphone as a transmission signal., EnterA reverberation and noise removal device for removing reverberant sound and noise mixed in a sound signal to be applied in an echo canceller and a noise canceller, respectively.(1) The noise canceller section includes (1-1) an FFT operation section that converts a time axis signal of an input digital audio signal into a frequency axis signal, and the FFT operation section includes the presence or absence of audio in the transmission input signal. Based on the sound detection signal for detecting the noise, the noise update is controlled in the FFT operation unit, the noise update is stopped when there is a sound, and the noise update is performed when there is no sound. -2) Further, an inverse FFT operation unit that converts the frequency axis signal converted by the FFT operation unit into a time axis signal, and (1-3) to remove noise components using the output of the inverse FFT operation unit as a reference input And an adaptive digital filter unit that outputs a removal signal of the signal, and the digital filter operates for each sample and uses the time-axis noise signal generated by the noise canceler unit as a reference signal. This estimates the pseudo noise and cancels the noise signal of the audio signal input from the microphone. (1-4) The processing timing of the signal delayed by the arithmetic processing in the FFT arithmetic unit and the inverse FFT arithmetic unit and the transmission signal is determined. (2) an echo canceller for canceling and removing the echo by generating a pseudo echo based on the received signal and the received signal; It is characterized by providing.
[0009]
  The reverberation and noise elimination apparatus according to the second aspect of the present invention outputs the received reception signal as an audio signal from the speaker, and mixes the input audio signal when the audio signal input from the microphone is output as a transmission signal. In an echo and noise canceller that removes echo and noise in an echo canceller and noise canceller, respectively. (1) In the noise canceller, (1-1) Converts the time axis signal of the input digital audio signal to a frequency axis signal. An FFT operation unit that controls noise update in the FFT operation unit based on a sound detection signal that detects the presence or absence of sound in the transmission input signal. The update is stopped and the noise is updated when there is no sound. (1-2) Frequency axis signal converted by the FFT calculation unit (1-3) a delay unit for matching the processing timing of the transmission signal and the signal delayed by the arithmetic processing in the FFT arithmetic unit and the inverse FFT arithmetic unit, (1-3) -4) A reception side adder that adds the output from the inverse FFT operation unit and the reference signal from the reception signal, and (2) the echo canceller unit, (2-1) output from the reception side adder and echo cancellation And echo canceling means for generating a pseudo echo based on the output from the transmission adder for canceling and canceling the echo.
[0010]
Thus, in the present invention, the noise component is once converted into a frequency component using an FFT (Fast Fourier Transform) operation unit, and the frequency component is smoothed for each frame to extract a rough characteristic of the frequency. An adaptive digital filter (hereinafter simply referred to as ADF) that has an inverse FFT operation unit that converts it back into a time-axis signal and adaptively removes pseudo-noise using the time signal reflecting the characteristics of the noise output from the inverse FFT operation unit as a reference signal. Therefore, the ADF unit that linearly operates on the time axis performs noise removal so as to ensure the performance of the subsequent echo canceller unit.
[0011]
DETAILED DESCRIPTION OF THE INVENTION
(A) First embodiment
Hereinafter, a first embodiment of a reverberation and noise removal apparatus according to the present invention will be described in detail with reference to the drawings.
[0012]
(A-1) Configuration of the first embodiment
FIG. 1 is a block diagram showing the configuration of the echo and noise removal apparatus of the first embodiment.
[0013]
The reverberation and noise removal apparatus according to the first embodiment includes a microphone 1, a noise canceller unit 3 that removes a noise component included in an audio signal collected by the microphone 1, an echo canceller unit 5 that removes an echo component, It is comprised by the speaker 7 which outputs an audio | voice signal as an audio | voice.
[0014]
The noise canceller unit 3 is connected to the microphone 1 and converts an analog audio signal input from the microphone 1 into a digital signal. The noise canceller unit 3 is connected to the A / D converter 31 and connected to the A / D converter. An FFT calculation unit 32 that converts a digital transmission signal inputted from 31 from a time axis signal to a frequency axis signal for each frame while receiving noise update control by a voice detection signal from an echo canceller unit 5 described later, An inverse FFT operation unit 33 connected to the FFT operation unit 32 to inversely convert the frequency axis signal into a time axis signal, an ADF unit 34 connected to the inverse FFT operation unit 33, and a signal and delay unit from the ADF unit 34 The adder 36 outputs the transmission signal obtained by adding the output signal from the output signal 35 to the echo canceller unit 5.
[0015]
  The echo canceller unit 5 includes an adder 51 that adds a transmission signal from which noise has been removed from the adder 36 and a signal from an ADF unit 52 described later, and an ADF unit 52.
[0016]
For ease of explanation, Sin represents a microphone input, Sout represents a transmission side output, Rin represents a reception side input, and Rout represents a speaker output.
[0017]
(A-2) Operation of the first embodiment
Hereinafter, the operation of the echo and noise elimination apparatus of the first embodiment will be described with reference to FIG.
[0018]
In the present invention, an audio signal including noise or the like is input from the microphone 1, but it is assumed that noise that is ambient noise is input first, and then the voice to be transmitted such as the real voice is input. . That is, when power is turned on, it is normal for noise to enter from the surroundings before the desired sound is input.
[0019]
First, a signal is input from the microphone 1. Here, the microphone input signal input from the microphone 1 is X. The analog signal from the microphone is digitized by an analog-digital (A / D) converter 31 and input to the FFT operation unit 32 in order to calculate a frequency component as a transmission signal. The processing of the FFT operation unit 32 is already well known and will not be described here.
[0020]
The FFT calculation unit 32 uses the input transmission signal as a frame, and converts it into a frequency component Xf (k, l) for each frame. Here, “k” is the k-th frequency, “l” is a number representing a frame number, and one frame is composed of data of 256 samples. That is, in the present invention, 256-point FFT calculation unit processing is used.
[0021]
The FFT operation unit 32 further obtains a smooth value of the power spectrum as shown in the following equation (1).
[0022]
Xaf (k, l + 1)
= Α · | Xf (k, l) | + (1−α) · Xaf (k, l) (1)
Here, Xaf (k, 0) = 0, α is a smoothing constant that determines the smoothness and smoothness, and 1> α> 0. Although the smoothing constant α = 0.5 is used in the present invention, the present invention is not limited to this.
[0023]
As shown in the equation (1), Xaf (k, l + 1) is obtained by adding the smoothed value Xaf (k, l) of the power spectrum up to the present time and the frequency component Xf (k, l) obtained at present to the next smoothing. It is calculated as a value.
[0024]
Therefore, when the smoothing coefficient α is α = 1, Xaf is updated while being sharply tracked with almost no smoothing. As a result, Xaf (k, l + 1) becomes equal to the frequency component Xf (k, l) as shown in the following equation (2).
[0025]
Xaf (k, l + 1) = | Xf (k, l) | (2)
On the other hand, if the smoothing constant α is small, the instantaneous value Xf (k, l) is not reflected much in the smooth value, and Xaf (k, l + 1) is updated slowly. Now, in the state where the noise N is first input to the microphone 1, the output of the FFT operation unit 32 is
Xf (k) = Nf (k) (3)
It becomes. Accordingly, the FFT operation unit 32 calculates the noise smoothing value Naf (k, l).
[0026]
Naf (k, l + 1)
= Α · | Nf (k, l) | + (1−α) · Naf (k, l) (4)
The Naf (k, l) calculated in this way, when viewed in the frequency domain, represents the characteristics of noise-specific frequency components on average.
[0027]
For example, in the case of automobile engine noise, it has a characteristic that the power spectrum has a relatively large power spectrum at a frequency of 1 KHz or less and gradually decreases as the frequency increases.
[0028]
FIG. 2 shows an example of frequency characteristics of automobile engine noise. Needless to say, the actual power spectrum has a rough shape such as Naf (k, l), but slightly fluctuates from frame to frame.
[0029]
The output from the FFT calculation unit 32 is output to the inverse FFT calculation unit 33, and Naf (k, l) is converted again into a time-axis signal Nh. Nh is a time-axis signal having rough characteristics of noise. Nh is input to the ADF unit 34.
[0030]
The ADF unit 34 removes components having a strong correlation with Nh on the time axis. That is, a pseudo noise signal N ′ that cancels the output signal N from the microphone 1 is created using Nh as a reference signal, added by the adder 36, and the residual signal e output from the adder 36 is again ADF unit 34. And the filter coefficient of the ADF unit 34 is updated so that the value of the expression (5) is minimized.
[0031]
e² = (N-Nh)²                     ... (5)
The coefficient updating algorithm in the ADF unit 32 is the same as that of the ADF unit 52 of the echo canceller unit 5 described later. Although the well-known “learning identification method” algorithm is used in the present invention, any algorithm may be used as long as it is a sequential algorithm that minimizes the equation (5). With regard to such a “learning identification method” algorithm, for example, (S Heykin, Introduction to Adaptive Filters, Hyundai Engineering Co., Ltd., Principal Translation, P120) may be referred to, and the description thereof is omitted here.
[0032]
As described above, since the adder 36 subtracts the pseudo noise reflecting the noise characteristics from the input noise, most of the noise input from the microphone 1 is removed. Here, the delay unit 35 is inserted between the microphone 1 and the adder 36 because the signal is delayed because the FFT calculation unit 32 performs frame processing. This is because the timing is matched in response to the transmission signal from 31.
[0033]
As described above, noise characteristics are once extracted in the frequency domain, converted back to a time axis signal, and input to the ADF unit 34. Since the processing of the ADF unit 34 is linear processing, non-linear processing which has been a problem in the conventional frequency subtraction or the like is not performed, and thus the subsequent echo canceller unit 5 is not adversely affected.
[0034]
A voice detection signal is input from the echo canceller 5 to the FFT calculator 32. This voice detection signal is output from, for example, a double talk detector (not shown), and indicates that there is actually a speaker signal instead of noise on each of the microphone input Sin side and the reception side input Rin side. On the time axis, the presence of each speaker signal is detected from the power of the signal of the receiving side input Rin and the power of the signal of the microphone input Sin.
[0035]
Here, the voice detection signal will be described.
[0036]
First, a signal is input to the microphone input Sin and input to the double talk detector via the delay unit 35, the adder 36, and the adder 51. In the double talk detector, parameters having two following characteristics are calculated as shown in equations (6) and (7). Here, n indicates the data of the nth sample.
[0037]
Xshort (n + 1)
= Β · | X (n) | + (1−β) · Xshort (n)
... (6)
Xlong (n + 1)
= Γ · | X (n) | + (1−γ) · Xlong (n)
... (7)
Further, Xshort (0) = 0 and Xlong (0) = 0.
[0038]
Expressions (6) and (7) are expressions representing signal smoothing as described in Expression (4), and β and γ are given coefficients different from β = 0.7 and γ = 0.2, respectively. To calculate.
[0039]
As described above, the smaller the coefficient, the slower the follow-up of the input signal, so Equation (6) quickly follows the signal change, and Equation (7) follows the signal change slowly.
[0040]
Now, when noise is input from the microphone 1 for a while, the values of the equations (6) and (7) are almost the same.
[0041]
Next, when a voice is input, Xshort (n) quickly follows the voice power and rises. However, since Xlong (n) follows slowly, a relation of Xshort (n)> Xlong (n) occurs.
[0042]
In this embodiment, V = 2
Xshort (n)> V · Xlong (n) (8)
When it became, “There is sound”. Here, V is a parameter that determines the sensitivity of voice detection, and may be appropriately determined by the designer depending on the use environment and the like, and is not limited to V = 2 in the present embodiment.
[0043]
The reception side input Rin side is detected in the same manner.
[0044]
The FFT operation unit 32 of the noise canceller unit 3 performs the calculation of Expression (1) when there is no audio signal on the microphone input Sin side or the reception side input Rin side, and when either of them has a result of “with sound” Is
Naf (k, l + 1) = Naf (k, l) (9)
As a result, the update of Naf (k, l) is stopped, and at the same time, the coefficient update of the ADF unit 34 is stopped. At this time, the inverse FFT operation unit 33 continues to operate.
[0045]
That is, Nh creates pseudo-noise Nh using Naf (k, l) that has been suspended. The pseudo noise Nh is not inconvenient because it only needs to have a rough characteristic of the noise component. Further, since the coefficient of the ADF unit 34 is fixed, even if the audio signal is fed back to the ADF unit 34 via the adder 36, the coefficient of the ADF unit 34 is not disturbed, and noise is continuously generated. Can be removed.
[0046]
The output of the noise canceller unit 3 is input to the echo canceller unit 5. The echo canceller is an adaptive echo canceller, which receives the input of the receiving side input Rin input terminal and the output from the noise canceller unit 3 and uses the output el of the adder 51, el²The coefficient of the ADF unit 52 is updated by an adaptive algorithm so as to minimize. As for the algorithm, a known method may be used as described above.
[0047]
However, as is well known, the coefficient update of the ADF unit 52 of the echo canceller unit 5 is performed only when the signal is present only on the reception side input signal Rin side.
[0048]
In the present invention, the coefficient update of the ADF unit 34 of the noise canceller unit 3 is stopped by the voice detection signal. However, the coefficient may be automatically stopped after being driven for a certain time after the operation is started.
[0049]
(A-3) Effects of the first embodiment
All the effects obtained by the above-described configuration and operation will be specifically described.
[0050]
As described above, according to the first embodiment, the input from the microphone 1 is frequency-converted by the FFT operation unit 32 and smoothed at the frame interval to extract noise characteristics, and the extracted signal is returned to the time axis. The noise is input to the ADF unit 34 for noise removal, and noise is adaptively removed on the time axis without being divided into specific frequencies. Furthermore, since the time axis ADF unit 34 naturally removes noise using linear processing, there is no sense of incongruity of sound quality due to noise removal in the noise canceller unit 3 and performance in the echo canceller unit 5 does not deteriorate. Natural sound without echo can be transmitted.
[0051]
(B) Second embodiment
Next, a second embodiment of the echo and noise removal apparatus according to the present invention will be described in detail with reference to the drawings.
[0052]
(B-1) Configuration of the second embodiment
FIG. 3 is a block diagram showing the configuration of the echo and noise elimination apparatus of the second embodiment. The same reference numerals are assigned to the same and corresponding parts as in FIG. 1 according to the first embodiment, and the details thereof are shown. Description is omitted.
[0053]
The second embodiment differs from the first embodiment in that there is a voice detection unit 37 between the microphone 1 and the FFT calculation unit 32, and that the voice detection signal from the echo canceller unit 5 is the FFT calculation unit 32. Is no longer entered.
[0054]
(B-2) Operation of the second embodiment
The second embodiment is different from the first embodiment in the part of the sound detection unit 37. Other parts are the same as those in the first embodiment, and will not be described here.
[0055]
The output from the microphone 1 is input to the sound detection unit 37 via the analog-digital converter 31. The voice detection unit 37 detects the voice of the microphone input Sin terminal by the same processing as the above formulas (6) and (7). When “sound is present”, the smoothing process in the FFT operation unit 32 is stopped, the execution of Expression (4) is stopped, and Expression (9) is executed instead. Further, when “sound is present”, the coefficient update of the ADF unit 34 is stopped.
[0056]
In the second embodiment, the noise canceller 3 does not use a signal on the receiving side input Rin side. For this reason, when the attenuation in the echo path is large and the echo is extremely small, the level of the echo is not significantly different from the level of the noise, so that the audio signal is not detected. However, as described above, since the FFT calculation unit 32 performs noise smoothing processing to extract a rough characteristic of the noise, there is almost no influence of such a small echo.
[0057]
On the other hand, when the echo is large, since the echo has the same property as the voice, it is detected as voice by the voice detection unit 37 and does not adversely affect noise estimation in the FFT calculation unit 32.
[0058]
As described above, in the second embodiment, since the sound detection can be performed in front of the delay unit 35, the sound on the microphone input Sin side can be detected earlier than the first embodiment, and the microphone input Sin side can be detected. When the sound is input to, it is possible to quickly prevent the influence of the sound from being reflected in the smoothing process by the FFT operation unit 32.
[0059]
(B-3) Effects of the second embodiment
All the effects obtained by the above-described configuration and operation will be specifically described.
[0060]
As described above, according to the second embodiment, the sound detector 37 is provided between the microphone 1 and the FFT operation unit 32 to perform sound detection and control the FFT operation unit 32 and the ADF unit 34. As a result, the noise can be estimated more accurately than in the first embodiment, and the noise can be accurately removed.
[0061]
(C) Third embodiment
Hereinafter, a third embodiment of the echo and noise removing apparatus according to the present invention will be described in detail with reference to the drawings. Here, FIG. 4 shows the configuration of this embodiment. The same reference numerals are given to the same and corresponding parts as in FIG. 3 described above, and the detailed description thereof is omitted.
[0062]
(C-1) Configuration of the third embodiment
FIG. 4 is a block diagram illustrating a configuration of the echo and noise removal apparatus according to the third embodiment. The third embodiment is based on the second embodiment described above. The second embodiment is different from the third embodiment in that an audio detection unit 38 that operates in the frequency domain is provided. It is that you are. Since other aspects are the same as those of the second embodiment, the description thereof is omitted here.
[0063]
The FFT operation unit 32 calculates equations (1) and (4) separately. Specifically, α in the formula (1) and α in the formula (4) are set separately as described below.
[0064]
Xaf (k, l + 1)
= Α₁ ・ | Xf (k, l) | + (1-α₁ ) · Xaf (k, l)
... (1) ’
Naf (k, l + 1)
= Α₂ ・ | Nf (k, l) | + (1-α₂ ) ・ Naf (k, l)
... (4) ’
Where 1> α₁ > Α₂ > 0, and in the third embodiment, α₂ = 0.5, α₁ However, the present invention is not limited to this.
[0065]
Moreover,
[Expression 1]

Calculate
PXF (l)> V · PNF (l) (12)
At this time, it is detected as “with sound”.
[0066]
When “sound is present”, the frequency axis sound detection unit 38 stops the coefficient update of the ADF unit 34.
[0067]
When “sound is present”, a signal for executing Expression (9) is output to the FFT operation unit 32.
[0068]
Or you may make it ignore Formula (10) and Formula (11) ignoring a specific frequency or a frequency band. This helps to prevent false detection of non-stationary noise as speech when it is known that non-stationary noise is concentrated on a specific frequency (frequency band). That is, the background noise can be accurately estimated by detecting the voice with higher accuracy in addition to the second embodiment.
[0069]
For example, a specific k or a set of several k in Expression (11) and Expression (12) is not reflected in the calculation.
[0070]
(C-3) Effects of the third embodiment
As described above, in the third embodiment, since the position of the sound detector is detected in the frequency domain between the FFT operation unit 32 and the inverse FFT operation unit 33, the details are more detailed. Noise can be estimated by performing speech detection and noise can be removed.
[0071]
(D) Fourth embodiment
Hereinafter, a fourth embodiment of the echo and noise removing apparatus according to the present invention will be described in detail with reference to the drawings. Here, FIG. 5 shows the configuration of this embodiment, and the same and corresponding parts as those in FIG. 1 described above are denoted by the same reference numerals, and detailed description thereof is omitted.
[0072]
(D-1) Configuration of the fourth embodiment
FIG. 5 is a block diagram illustrating a configuration of the echo and noise removal apparatus according to the fourth embodiment. The fourth embodiment is based on the first embodiment. The fourth embodiment is different from the first embodiment in that the adder 17 is provided in front of the ADF unit 52, while the fourth embodiment is different from the first embodiment. That is, the ADF unit 34 and the adder 3 in the first embodiment are deleted.
[0073]
The configuration of the fourth embodiment will be described below using FIG.
[0074]
Since the processing from the microphone 1 to the inverse FFT operation unit 33 is the same as that described in the first embodiment, the description thereof is omitted here.
[0075]
The output of the inverse FFT operation unit 33 is output to the addition unit 53. The adder 53 also receives a signal from the receiving side input Rin input terminal 9. On the other hand, an audio detection signal is input to the FFT operation unit 32 from a double talk detector (not shown) of the echo canceller unit 5.
[0076]
In the fourth embodiment, the operations of the ADF unit 52 and the FFT operation unit 32 are different. That is, the ADF unit 52 updates the coefficient as follows according to the sound detection result from the double talk detector.
[0077]
There is an audio signal only on the input side Rin on the receiving side: noise update (9) equation (that is, stop) of the FFT calculation unit 32, coefficient update of the ADF unit 52
There is an audio signal only on the microphone input Sin side: noise update equation (9) in the FFT operation unit 32, coefficient update stop in the ADF unit 52
There is an audio signal on both the receiving side input Rin side and the microphone input Sin side: noise update (9) expression (that is, stop) of the FFT calculation unit 32, coefficient update stop of the ADF unit 52
No audio signal on both the receiving side input Rin side and the microphone input Sin side: Noise update (4) formula in the FFT calculation unit 32, coefficient update stop in the ADF unit 52
That is, the output signal of the inverse FFT operation unit 33 may be handled in the same manner as the reference signal for creating the pseudo echo of the echo canceller unit 5, thereby making it possible to greatly reduce the hardware scale.
[0078]
At this time, the inverse FFT operation unit 33 always operates to create a time axis signal and continue to output it to the ADF unit 52.
[0079]
(D-3) Effect of the fourth embodiment
As described above, according to the fourth embodiment, the output of the inverse FFT operation unit 33 is directly input to the echo canceller ADF unit 52, and the coefficient update control method of the ADF unit 52 is changed.
There is an audio signal only on the input side Rin on the receiving side: noise update (9) equation (that is, stop) of the FFT calculation unit 32, coefficient update of the ADF unit 52
There is an audio signal only on the microphone input Sin side: noise update equation (9) in the FFT operation unit 32, coefficient update stop in the ADF unit 52
There is an audio signal on both the receiving side input Rin side and the microphone input Sin side: noise update (9) equation (that is, stop) of the FFT calculation unit 32, coefficient update stop of the ADF unit 52
No audio signal on both receiving side input Rin side and microphone input Sin side: Noise update (4) equation of FFT operation unit 32, coefficient update stop of ADF unit 52
Thus, the output signal of the inverse FFT operation unit 33 may be handled in the same manner as the reference signal for creating the pseudo echo of the echo canceller unit 5, and the hardware can be greatly reduced in scale.
[0080]
(E) Fifth embodiment
Hereinafter, a fifth embodiment of the echo and noise removal apparatus according to the present invention will be described in detail with reference to the drawings. Here, FIG. 6 shows the configuration of this embodiment. The same reference numerals are given to the same and corresponding parts as in FIG.
[0081]
(E-1) Configuration of the fifth embodiment
FIG. 1 is a block diagram illustrating a configuration of an echo and noise removal apparatus according to the fifth embodiment. The fifth embodiment is based on the second embodiment and the fourth embodiment. Since the operation of the time axis sound detector has been described in the second embodiment, the description thereof is omitted here. The operations of the FFT operation unit 32 and the inverse FFT operation unit 33 are as described in the fourth embodiment.
[0082]
The control of the ADF unit 52 is
There is an audio signal only on the receiving side input Rin side: coefficient update of the ADF unit 52
There is an audio signal only on the microphone input Sin side: coefficient update of the ADF unit 52 is stopped
There is an audio signal on both the receiving side input Rin side and the microphone input Sin side: coefficient update of the ADF unit 52 is stopped
No audio signal on both the receiving side input Rin side and the microphone input Sin side: coefficient update of the ADF unit 52 stopped
The noise update control of the FFT operation unit 32 is
Detection result from voice detector With voice signal: Noise update (9) expression of FFT operation unit 32 (ie, stop)
Detection result from voice detector No voice signal: Noise update (4) formula (that is, update) of FFT operation unit 32
This is different from the fourth embodiment.
[0083]
In other words, the output signal of the inverse FFT operation unit 33 may be handled in the same manner as the reference signal for creating the pseudo echo of the echo canceller unit 5, thereby greatly reducing the hardware scale.
[0084]
At this time, the inverse FFT operation unit 33 always operates to create a time axis signal and continue to output it to the ADF unit 52.
[0085]
Further, as described in the second embodiment, since the sound detection can be performed in front of the delay unit 35, the sound can be detected more quickly. As a result, the damage given to the noise estimation of the FFT operation unit 32 can be reduced, so that more accurate noise removal can be performed.
[0086]
(E-3) Effects of the fifth embodiment
As described above, according to the fifth embodiment, in addition to the fourth embodiment, voice can be detected more quickly, and as a result, damage to the noise estimation of the FFT operation unit 32 can be reduced. Therefore, more accurate noise removal can be performed. Further, by omitting the ADF unit 34 as compared with the first embodiment, the hardware can be reduced in size.
[0087]
(F) Sixth embodiment
Hereinafter, a sixth embodiment of the echo and noise removing apparatus according to the present invention will be described in detail with reference to the drawings. Here, FIG. 7 shows the configuration of this embodiment. The same reference numerals are given to the same and corresponding parts as in FIG.
[0088]
(F-1) Configuration of the sixth embodiment
FIG. 7 is a block diagram showing the configuration of the echo and noise removal apparatus of the sixth embodiment. The sixth embodiment is based on the third embodiment and the fourth embodiment. Since the operation of the frequency domain sound detection unit 38 has been described in the third embodiment, the description thereof is omitted here.
[0089]
Unlike the third embodiment, the sixth embodiment outputs the output of the frequency domain sound detection unit 38 only to the FFT operation unit 32. The FFT calculation unit 32 updates the noise in the same manner as in the third embodiment, and operates with the inverse FFT calculation unit 33 to create pseudo noise on the time axis.
[0090]
The control of the ADF unit 52 of the echo canceller unit 5 updates the coefficient of the ADF unit 52 by the same operation as described in the fourth embodiment. The update control method of the FFT operation unit 32 and the ADF unit 52 is the same as that described in the fifth embodiment.
[0091]
(F-3) Effect of the sixth embodiment
As described above, according to the sixth embodiment, voice detection is performed quickly, and further, voice detection is performed in the frequency domain, so that finer voice detection is performed and noise estimation is performed and removed. As in the fourth embodiment, the hardware can be greatly reduced in size.
[0092]
In the above-described embodiment, a so-called “speaker phone” using a microphone and a speaker has been described as an example. However, the present invention is not limited to this example. Of course, in a system that does not require electrical / acoustic conversion by a speaker, it is a matter of course that a system that outputs an electrical signal as it is may be used.
[0093]
【The invention's effect】
As described above, according to the echo and noise removal device of the present invention, the noise is adaptively removed on the time axis, so that the sound quality does not cause a sense of incongruity associated with noise removal and the performance of the echo canceller. Natural sound without echo can be transmitted without deterioration.
[Brief description of the drawings]
FIG. 1 is a block diagram showing a schematic configuration of a first embodiment to which the present invention is applied.
FIG. 2 is a frequency characteristic diagram showing a frequency characteristic of an automobile engine.
FIG. 3 is a block diagram showing a schematic configuration of a second embodiment to which the present invention is applied.
FIG. 4 is a block diagram showing a schematic configuration of a third embodiment to which the present invention is applied.
FIG. 5 is a block diagram showing a schematic configuration of a fourth embodiment to which the present invention is applied.
FIG. 6 is a block diagram showing a schematic configuration of a fifth embodiment to which the present invention is applied.
FIG. 7 is a block diagram showing a schematic configuration of a sixth embodiment to which the present invention is applied.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 1 ... Microphone, 3 ... Noise canceller part, 5 ... Echo canceller part, 7 ... Speaker, 31 ... A / D conversion part, 32 ... FFT operation part, 33 ... Inverse FFT operation part, 34 ... ADF part, 35 ... Delay part, 36 ... adder, 37 ... voice detector, 38 ... voice detector, 51 ... adder, 52 ... ADF unit, 53 ... adder.

Claims (6)

When the received reception signal is output as an audio signal from the speaker and the audio signal input from the microphone is output as a transmission signal, echo sound and noise mixed in the input audio signal are reflected in the echo canceller unit and the noise canceller unit. In the reverberation and noise removal device to remove each,
In the noise canceller,
An FFT operation unit that converts a time axis signal of an input digital audio signal into a frequency axis signal ;
The FFT operation unit controls the noise update in the FFT operation unit based on a sound detection signal that detects the presence or absence of sound in the transmission input signal, and stops noise update when there is sound, The noise is updated when there is none,
Further, an inverse FFT operation unit that converts the frequency axis signal converted by the FFT operation unit into a time axis signal;
And a adaptive digital filter unit for outputting a removal signal for removing a noise component as a reference input the output of the inverse FFT operation section,
The digital filter operates for each sample and estimates the pseudo noise using the time axis noise signal created by the noise canceller as a reference signal, and cancels the noise signal of the audio signal input from the microphone,
A delay unit for adjusting the processing timing of the signal delayed by the arithmetic processing in the FFT arithmetic unit and the inverse FFT arithmetic unit and the transmission signal;
In prior Symbol echo canceller,
An echo and noise removal apparatus comprising: an output from the noise canceller unit; and an echo removal means for generating a pseudo echo based on the received signal and canceling and removing the echo. When the noise removal unit controls the noise update in the FFT operation unit by the voice detection signal from the echo canceller unit, there is a result of “with voice” in either the microphone input signal or the reception side input signal. Is
When the noise is N and the smoothing value of the noise is Naf (k, l) (where k is the k-th frequency, l is a number representing the frame number),
Naf (k, l + 1) = Naf (k, l)
Stop updating Naf (k, l) as
Otherwise, update the coefficient of the adaptive digital filter unit Naf (k, l + 1)
= Α · | Nf (k, l) | + (1−α) · Naf (k, l)
The reverberation and noise elimination apparatus according to claim 1, wherein the background noise estimation value is updated as follows. When the received reception signal is output as an audio signal from the speaker and the audio signal input from the microphone is output as a transmission signal, echo sound and noise mixed in the input audio signal are reflected in the echo canceller unit and the noise canceller unit. In the reverberation and noise removal device to remove each,
The noise canceller unit includes an FFT operation unit that converts a time axis signal of an input digital audio signal into a frequency axis signal ,
The FFT operation unit controls the noise update in the FFT operation unit based on a sound detection signal that detects the presence or absence of sound in the transmission input signal, and stops noise update when there is sound, The noise is updated when there is none,
And inverse FFT operation section that converts the converted frequency axis signal by the FFT calculation unit on the time axis signal,
A delay unit for matching processing timings of the signal delayed by the arithmetic processing in the FFT arithmetic unit and the inverse FFT arithmetic unit and the transmission signal;
And a receiving-side adder for adding a reference signal from the output and the reception signal from the inverse FFT operation section,
In the echo canceller,
Echo and the noise removal device, characterized in that the output from the previous SL recipient adder output and the transmission side adder for echo cancellation and generates a pseudo echo on the basis and a echo removing means for removing cancel the echo .   4. The echo according to claim 1, wherein the voice detection signal is obtained by a voice detection unit that performs voice detection based on a difference between two types of smooth values having different reaction speeds. 5. Noise removal device. The echo and noise removal apparatus according to any one of claims 1 to 4 , wherein the voice detection signal is input from an echo canceller or a voice detector. The speech detection section, echo and noise removal device according to any one of claims 1 to 5, characterized in that operating in the time domain or frequency domain.

Images