JP3583998B2 - Multi-channel echo canceling method, apparatus therefor, and program recording medium - Google Patents

Description

Ｌは適応フィルタ１１_１ ，…，１１_Ｎ のタップ数（フィルタ長）である。
【０００４】
受話信号ベクトルｘ _１（ｋ），…， ｘ _Ｎ（ｋ）を次式のベクトルＸ（ｋ）とする
Ｘ（ｋ）＝［ｘ _１ ^Ｔ（ｋ）， ｘ _２ ^Ｔ（ｋ），…， ｘ _Ｎ ^Ｔ（ｋ）］^Ｔ
各適応フィルタ１１_１ ，…，１１_Ｎ のフィルタ係数ベクトルｈ＾_１ｍ（ｋ）， ｈ＾_２ｍ（ｋ），…， ｈ＾_Ｎｍ（ｋ）を結合して
ｈ＾_ｍ（ｋ）＝［ｈ＾_１ｍ ^Ｔ（ｋ）， ｈ＾_２ｍ ^Ｔ（ｋ），…， ｈ＾_Ｎｍ ^Ｔ（ｋ）］^Ｔ
と適応フィルタ結合係数ベクトルｈ＾_ｍ（ｋ）の更新は、例えばＮＬＭＳアルゴリズム（学習同定法）を用いた場合
ｈ＾_ｍ（ｋ＋１）＝ｈ＾_ｍ（ｋ）＋αｅ_ｍ （ｋ）ｘ（ｋ）／（ｘ ^Ｔ（ｋ） ｘ（ｋ））
αはステップサイズパラメータである。
となる。つまり修正係数計算部１３ではこの式の右辺第２項αｅ_ｍ （ｋ） ｘ（ｋ）／（ｘ ^Ｔ（ｋ） ｘ（ｋ））を計算して、各適応フィルタ１１_１ ，…，１１_Ｎ に対するフィルタ係数の修正係数を得、これら修正係数により適応フィルタ１１_１ ，…，１１_Ｎ の各フィルタ係数をそれぞれ係数更新部１４_１ ，…，１４_Ｎ で更新する。適応フィルタ１１_１ ，…，１１_Ｎ ではそのフィルタ係数を用いてフィルタ処理部１５_１ ，…，１５_Ｎ でフィルタ係数ベクトルと受話信号ベクトルとの内積演算（フィルタ処理）ｙ＾_１ｍ（ｋ）＝ｈ＾_１ｍ ^Ｔ（ｋ） ｘ _１（ｋ），…，ｙ＾_Ｎｍ（ｋ）＝ｈ＾_Ｎｍ ^Ｔ（ｋ） ｘ _Ｎ（ｋ）が行われて、疑似エコー信号が生成される。
【０００５】
各フィルタ係数ベクトルは

であって、前記修正を繰返すことにより、これらフィルタ係数ベクトルを、各スピーカ２_１ ，…，２_Ｎ からマイクロホン３_ｍ への各エコー経路のインパルス応答ｈ_１ｍ（ｋ，ｎ），…，ｈ_Ｎｍ（ｋ，ｎ）の時系列を各要素とするエコー経路ベクトル

と一致させることができ、残留信号ｅ_ｍ（ｋ）を小さくすることができる。
【０００６】
なお、各フィルタ処理部とその係数更新部及び対応する修正係数計算部分を含めて本来の適応フィルタが構成されるが、ここでは便宜上フィルタ処理部と係数更新部を適応フィルタと呼ぶ。
【０００７】
【発明が解決しようとする課題】
多チャネルの再生系及び収音系よりなる音響システムにおいては、スピーカ２_１ ，…，２_Ｎ 中には音量が大きなものと、小さいものとがあり、同様にマイクロホン３_１ ，…，３_Ｍ 中には収音利得が大きいものと、小さいものとがある。音量が小さいスピーカからの再生音が収音利得が小さいマイクロホンに混入したエコー信号のレベルは小さなものとなる。従ってこのエコー信号をある所定レベルまで減算させるには適応フィルタの長さは短かくてよい。逆に音量が大きいスピーカからの再生音が収音利得が大きいマイクロホンに混入したエコー信号のレベルは大きなものとなり、これを前記ある所定レベルまで減衰させるには、適応フィルタ長を長くする必要がある。
【０００８】
しかし、従来においては適応フィルタ１１_１ ，…，１１_Ｎ のフィルタ長は全て同一としていた。このためレベルが小さいエコー信号のエコー経路と対応する適応フィルタには必要以上のフィルタ長を与え、それだけ不必要に多くの演算処理を行い、あるいはハードウェア規模を不必要に大きくすることになっていた。逆にレベルが大きいエコー信号のエコー経路と対応する適応フィルタには不十分なフィルタ長が与えられ、エコー信号を十分抑圧することができないことがあった。この問題はＮＬＭＳアルゴリズムを用いる場合に限らず、適応フィルタのフィルタ係数を、エコー経路のインパルス応答に近ずける他の適応アルゴリズムを用いた場合も同様のことが言える。
【０００９】
この発明の目的は、多チャネルエコー消去系において、全体適応フィルタのフィルタ長が与えられた時、各適応フィルタに適切なフィルタ長を与える多チャネルエコー消去方法、その装置、そのプログラム及びその記録媒体を提供することにある。
【００１０】
【課題を解決するための手段】
この発明によれば、各チャネルについて、そのチャネルの受話信号を再生するスピーカと、収音信号を出力するマイクロホン間、つまり受話信号の再生信号と収音信号との間の間接音による音響結合量を求め、対数領域では、この各チャネルの間接音による音響結合量とそのチャネルの受話信号量との加算と、これらから目標誤差信号量の減算とを行った値を、目標間接音抑圧量として求め、チャネルの目標間接音抑圧量に比例してそのチャネルの適応フィルタ長を決定する。
【００１１】
【発明の実施の形態】
全体の適応フィルタのフィルタ長が与えられた時に、各適応フィルタに適切なフィルタ長を与えるエコー消去として考えられるものを示す図１を参照して説明する。図１で図６と対応する部分に同一参照符号を付けて重複説明を省略する。また修正係数算出部１３を省略してある。
各適応フィルタ１１₁ ,…,１１_N と対応してフィルタ長決定部２１₁ ,…,２１_N が設けられ、フィルタ長決定部２１₁ は間接音音響結合量計算部２２及びフィルタ長計算部２３からなる。フィルタ長決定部２１₂ ,…,２１_N も、図に示していないが、それぞれ同様に間接音音響結合量計算部２２及びフィルタ長計算部２３から構成されている。
【００１２】
この適応フィルタ長を求める原理を以下に説明する。スピーカ２_i からマイクロホン３_m へのエコー経路における残響曲線Ｃ_i(ｋ）はそのエコー経路に対応するインパルス応答ｈ_i を、その最後から自乗積分したものである。
Ｃ_i(ｋ）＝Σｈｊ(ｋ＋ｊ）²，Σは、ｊ＝０，１，２，・・・に関する総和。
Ｃ_i(0)はインパルス応答の自乗積分値であるからスピーカ２_iとマイクロホン３_m 間の音響結合量Ａ_i に等しい。音響結合量Ａ_i のうち直接音によるものをＤ_i 、間接音によるものをＲ_i とするとＡ_i ＝Ｄ_i ＋Ｒ_i である。スピーカが１₁ と１₂ の二つつまりエコー経路が二つの場合における、これらエコー経路の残響曲線Ｃ₁ ,Ｃ₂ は図２に示すようになる。図２において横軸は時間軸であり、縦軸は残響音量である。時刻ｋ＝０における値Ｃ₁(０),Ｃ₂(０)は各エコー経路のインパルス応答の自乗積分値であり、音響結合量Ａ₁ ,Ａ₂ であり、図２においてｋ＝０付近での曲線Ｃ₁ ,Ｃ₂ の平らな部分は、エコーの遅延時間であり、この部分を明示するため、つまり縦軸から離してあるが、実際にはフィルタ長Ｌ₁ ,Ｌ₂ と比べ著しく短かく、例えば数ｍｓから数十ｍｓであって無視できる程度である。なおｋ＝０付近での平らな部分も雑音等のために若干の傾きを持つが、直接音や間接音による傾きに比べると小さく、平らと見なしてよい。図２中のＥは目標音響結合量であって、各エコー経路の残響が目標音響結合量Ｅになるまでのエコー信号を、適応フィルタ１１₁ ,１１₂ でそれぞれ模擬すればよい。つまり、スピーカ２₁ ,２₂ からの再生音が生じてからそのエコー経路の残響のレベルが目標音響結合量Ｅになるまでの時間Ｌ₁ ,Ｌ₂ を適応フィルタ１１₁ ,１１₂ のフィルタ長にすればよい。
【００１３】
残響曲線Ｃ_１ ，Ｃ_２ は音響結合量Ａ_１ ，Ａ_２ からそれぞれ直接音による音響結合量Ｄ_１ ，Ｄ_２ を差し引いた値、つまり間接音による音響結合量Ｒ_１ ，Ｒ_２ （対数値）から時間と共に対数領域で直線的に減少する。また、同一室内では、間接音による残響曲線の減少の傾きがスピーカ、マイクロホンの位置によらず一定である。従って、前述したようにほぼｋ＝０の時刻から残響曲線が直線的に減少するとみなすことができるから、全体のフィルタ長Ｌ＝Ｌ_１ ＋Ｌ_２ とすると、
Ｌ_１ ：Ｌ_２ ≡ｌｏｇ Ｒ_１ −ｌｏｇ Ｅ：ｌｏｇ Ｒ_２ −ｌｏｇ Ｅ
が成立する。よって、フィルタ長Ｌ_ｉ は次式により求まる。
【００１４】

となる。スピーカの数がＮの一般的な場合における各適応フィルタ１１_ｉ のフィルタ長Ｌ_ｉ は次式により求まる。
Ｌ_ｉ ＝［（ｌｏｇ Ｒ_ｉ −ｌｏｇ Ｅ）／Σ（ｌｏｇ Ｒ_ｉ −ｌｏｇ Ｅ）］Ｌ
Σはｉ＝１からＮまでの総和である。
以上の説明から理解されるように、図１において例えばフィルタ長決定部２１_１ においてフィルタ長計算部２３に全体のフィルタ長Ｌと、目標音響結合量Ｅが設定入力され、間接音音響結合量計算部２２からの間接音による音響結合量Ｒ_１ も入力され、その減算部２３ａで目標間接音結合抑圧量ｌｏｇ Ｒ_１ −ｌｏｇ Ｅが計算される。同様に各フィルタ長決定部２１_２，… ，２１_Ｎ でそれぞれ目標間接音結合抑圧量ｌｏｇ Ｒ_２ −ｌｏｇ Ｅ，…，ｌｏｇ Ｒ_Ｎ −ｌｏｇ Ｅが計算される。これら目標間接音結合抑圧量ｌｏｇ Ｒ_１ −ｌｏｇ Ｅ，…，ｌｏｇ Ｒ_Ｎ −ｌｏｇ Ｅにそれぞれ比例して対応する適応フィルタ１１_１ ，…，１１_Ｎ にそのフィルタ長を図中に破線で示すように設定してもよい。
【００１５】
適応フィルタ１１_１ ，…，１１_Ｎ のフィルタ長Ｌ_１ ，…，Ｌ_Ｎの和Ｌが決められている場合は更に次のようにする。目標間接音結合抑圧量ｌｏｇ Ｒ_１ −ｌｏｇ Ｅ，…，ｌｏｇ Ｒ_Ｎ −ｌｏｇ Ｅは加算部２４で加算され、つまりΣ_ｉ＝１ ^Ｎ（ｌｏｇ Ｒ_ｉ −ｌｏｇ Ｅ）＝Ｐ_ＲＥＡ が得られ、この加算値Ｐ_ＲＥＡ が各フィルタ長決定部２１_１ ，…，２１_Ｎ の各フィルタ長計算部２３に入力され、各フィルタ長計算部２３では割算部２３ｂにより（ｌｏｇ Ｒ_ｉ −ｌｏｇ Ｅ）／Ｐ_ＲＥＡ ＝Ｌ_ｉ がそれぞれ計算され、これら計算されたＬ_１ ，…，Ｌ_Ｎ がそれぞれ対応する適応フィルタ１１_１ ，…，１１_Ｎ にそのフィルタ長として設定される。
【００１６】
このようにして適応フィルタ１１_１ ，…，１１_Ｎ はそれぞれそのエコーレベルに応じたフィルタ長に設定される。
所で間接音音響結合量計算部２２における間接音による音響結合量Ｒ_ｉ を求めるには次の手法が考えられる。
図３Ａに示すようにインパルス応答測定部２５において、予め各エコー経路についてインパルス応答を測定し、そのインパルス応答を用いる。つまり各スピーカ２_１ ，…，２_Ｎ の各１つからインパルス音響を再生させ、その時のマイクロホン３_ｍ の出力のパワーの減衰状態を測定してインパルス応答における各時刻のｈ_ｉ を求め、この測定したインパルス応答を間接音音響結合量計算部２２に入力し、この計算部２２において先に述べたようにそのインパルス応答の自乗積分値Ｃ_ｉ（０）を求め、この値Ｃ_ｉ（０）＝Ａ_ｉ から、インパルス応答の時刻ｋ＝０における測定パワーｈ_ｉ（０）^２ 、つまり直接音による音響結合量Ｄ_ｉ を対数領域では差し引いて間接音による音響結合量Ｒ_ｉ を求める。
【００１７】
あるいは図３Ｂに示すように適応フィルタ１１_ｉ からそのフィルタ係数ｈ_ｉ を入力して、図３Ａの場合と同様に自乗積分値Ｃ_ｉ（０）を求め、更にこの値からｈ_ｉ（０）^２ ＝Ｄ_ｉ を対数領域では差し引いて間接音に与える音響結合量Ｒ_ｉ を求める。
図２に示した特性から理解されるように、Ｃ_ｉ（ｋ）／Ｃ_ｉ（ｋ＋τ）＞θであるような最小の整数ｋを直接音の到来時刻と考える。ここでτは数ｍｓに相当する値、θは２〜４程度である。つまり図２において直接音から間接音（残響）に変化する、急激にレベルが下がる時刻ｋを求め、直接音による音響結合をＤ_ｉ＝Ｃ_ｉ（ｋ）／Ｃ_ｉ（ｋ＋τ）とする。図２は対数値であるから引き算となる。よって間接音Ｒ_ｉはＲ_ｉ＝Ｃ_ｉ（０）／Ｂ_ｉにより求まる。直接音が到来してしまっているはずの時刻をｋが過ぎても先の条件が満たされない場合は、間接音が優勢と判断し、Ｄ_ｉは１とθの中間に設定する。
【００１８】
あるいは図３Ｃに示すように各受話信号ｘ_１（ｋ），…，ｘ_Ｎ（ｋ）を共分散計算部２６に入力して、これらを要素とする入力信号列ベクトルをＸとする時、Ｘの共分散を要素とする列ベクトルの時間的平均値ｒ（ｋ）を求める。受話信号がｘ_１（ｍ）とｘ_２（ｍ）の二つの場合は
ｒ（ｋ）＝αｒ（ｋ−１）＋（１−α）［ｘ_１（ｋ）^＊ｘ_１（ｋ），ｘ_１（ｋ）^＊ｘ_２（ｋ），ｘ_１（ｋ）ｘ_２（ｋ），ｘ_２（ｋ）^＊ｘ_２（ｋ）］^Ｔ
Ｔは転置行列、 ^＊は共役を表わす。０≦α＜１
である。マイクロホン３_ｍ の出力ｙ_ｍ（ｋ）を２乗平均値計算部２７に入力して２乗平均値
ｙ^２（ｋ）＝βｙ^２（ｋ−１）＋（１−β）ｙ^２ _ｍ（ｋ），０≦β＜１
を求める。ｒ（ｋ）とｙ^２（ｋ）を音響結合推定部２８に入力して、次式を計算する。
【００１９】
ｍｉｎ（Ｅ［ｙ^２（ｋ）−ｇ^Ｔ（ｋ）ｒ（ｋ）^２］，ｇ^Ｔ）
ここでＥ［ ］は平均をとる演算を表わし、Ｅ［ａ（ｋ）］＝Σｗ（ｉ）ａ（ｋ−ｉ），ｉ＝０，１，…である。窓関数ｗ（ ）のとり方には、ある一定の区間Ｎだけ１／Ｎで他は０という方形窓や、ｗ（ｉ）＝λ^ｉ，（０＜λ＜１）という指数減衰窓がよく用いられる。
ｍｉｎ（Ｊ，ａ）はＪが最小になるようにａを選ぶという演算を意味する。ｙ^２（ｋ）とｇ^Ｔ（ｋ）ｒ（ｋ）の差の２乗が最小になるように、最小２乗法、再帰的最小２乗法、共役勾配法、適応フィルタリング手法などにより列ベクトルｇ^Ｔ（ｋ）を求める。ここでｇはｙ_ｍ（ｋ）とｘ_１ ，…，ｘ_Ｎ を関係づける列ベクトルである。受話信号がｘ_１ とｘ_２ の二つの場合、ｇは次式を表わす。
【００２０】
ｇ＝［Ｇ_１１Ｇ_１１ ^＊ ，Ｇ_１２Ｇ_１１ ^＊ ，Ｇ_１１Ｇ_１２ ^＊ ，Ｇ_１２Ｇ_１２ ^＊ ］^Ｔ
ここでＧ_１１Ｇ^＊ はスピーカ２_１ とマイクロホン３_ｍ 間の音響結合量であり、Ｇ_１２Ｇ_１２ ^＊ はスピーカ２_２ とマイクロホン３_ｍ 間の音響結合量である。
このようにして得られた各スピーカ２_１ ，…，２_Ｎ とマイクロホン３_ｍ 間の音響結合量Ａ_１ ，…，Ａ_Ｎ から減算部２９で直接音による音響結合量とみなす適当な値Ｄ_ｉ′を対数領域では差し引いて間接音による音響結合量Ｒ_ｉ を求める。直接音による音響結合量Ｄ_ｉ はｌｏｇ Ｒ_ｉ −ｌｏｇ Ｅ_ｉ と比較して１桁程度は小さい値であり、経験的に予測できる数ｄＢ程度の値であって、その程度の値をＤ_ｉ′として減算部２９に設定入力すればよい。ｇ^Ｔ（ｋ）をより正しく求めるには、逐次得られるｇ^Ｔ（ｋ）中のＧ_１１Ｇ_１１ ^＊の最小値又は最頻値（例えばヒストグラムを作って）を求め、ｇＧ_１１Ｇ_１１ ^＊のｇ^Ｔ（ｋ）を用いればよい。
【００２１】
この場合は各フィルタ長決定部２１₁ ，…,２１_N に用いる間接音による音響結合量Ｒ₁ ，…,Ｒ_N が１つ１つの間接音音響結合量計算部２２により求まる。
図３Ｂ及び図３Ｃに示した手法によれば、この多チャネルエコー消去装置を使用中に、音響結合量が変化した場合に適応的に各チャネルに適したフィルタ長を変更することができる。
次にこの発明の実施形態を図４に、図１と対応する部分に同一参照符号を付けて示す。この実施形態ではフィルタ長決定部３１₁ ，…,３１_N が適応フィルタ１１₁ ，…,１１_N と対応して設けられる。フィルタ長決定部３１₁ ，…,３１_Nは同一構成であり、フィルタ長決定部３１₁ に示すように間接音音響結合量計算部２２とフィルタ長計算部３２とより構成される。間接音音響結合量計算部２２は図１中のそれと同様であるが、フィルタ長計算部３２には、間接音音響結合量Ｒ₁、全体のフィルタ長Ｌの他に目標誤差信号レベルＥ_eと、対応する受話信号ｘ₁ が入力される。
【００２２】
フィルタ長計算部３２ではその加減算部３２ａで受話信号ｘ_ｉ のパワーＸ_ｉ とＥ_ｅ とＲ_ｉ とから目標間接音抑圧量ｌｏｇ Ｒ_ｉ ＋ｌｏｇ Ｘ_ｉ −ｌｏｇ Ｅ_ｅ が計算される。この場合も図中に波線で示すように目標間接音抑圧量Ｅｌｏｇ Ｒ_ｉ ＋ｌｏｇ Ｘ_ｉ −ｌｏｇ Ｅ_ｅ に比例した値を適応フィルタ１１_１，…，１１_Ｎにそれぞれフィルタ長Ｌ_１，…，Ｌ_Ｎとして設定してもよい。全体のフィルタ長Ｌが決められている場合は更に次のようにする。各フィルタ長決定部３１_１ ，…，３１_Ｎ からの目標間接音抑圧量ｌｏｇ Ｒ_ｉ ＋ｌｏｇ Ｘ_ｉ −ｌｏｇ Ｅ_ｅ が加算部３３で加算され、その加算値Ｐ_ＲＥＡ′が各フィルタ長計算部３２へ供給される。
【００２３】
各フィルタ長計算部３２ではその割算部３２ｂ（ｌｏｇ Ｒ_ｉ ＋ｌｏｇ Ｘ_ｉ −ｌｏｇ Ｅ_ｅ ）／Ｐ_ＲＥＡ′＝Ｌ_ｉを計算し、そのＬ_ｉ を対応する適応フィルタ１１_ｉ のフィルタ長に設定する。
この場合は適応フィルタ長Ｌ_ｉ の決定に、その受話信号ｘ_ｉ も考慮されているため、例えば受話信号ｘ_ｉ が大きなレベルの場合はそれに応じて、対応適応フィルタのフィルタ長が長くされ、そのエコー信号を十分抑圧することができる。間接音による音響結合量Ｒ_ｉ の推定は図３に示した各種手法を用いることができる。
【００２４】
上述では各種計算を対数領域で行ったが、線形領域で行ってもよい。対数領域での引算は線形領域では割算となる。
上述したこの発明の実施形態を、コンピュータによりプログラムを実行させて機能させることができる。つまりそのためのプログラムを記録したＣＤ−ＲＯＭ、フロッピー（登録商標）ディスク、磁気ディスク、あるいは伝送路を通じて別の場所に記憶されている記憶媒体から、コンピュータ内の動作プログラムメモリ内にインストールして、そのプログラムメモリ上のプログラムをコンピュータにより実行させて機能させてもよい。
【００２５】
【発明の効果】
以上述べたようにこの発明によれば、全チャネルに用いられる適応フィルタのフィルタ長が与えられた時、つまりそのため演算量が与えられた時に、各チャネルに適したフィルタが設定され、不必要にフィルタ長が長い適応フィルタが生じたり、適応フィルタ長が短かいためにエコー信号を十分抑圧することができなかったりするおそれがない。
特に受話信号のレベルも考慮するため、話者の発声音量に応じて適切なフィルタ長を設定することができる。
【図面の簡単な説明】
【図１】適切なフィルタ長の設定が可能とされた多チャネルエコー消去装置を示す機能構成図。
【図２】この発明の原理を説明するための残響曲線の例を示す図。
【図３】間接音による音響結合量を求める具体例の機能構成を示す図。
【図４】この発明の実施形態を示す機能構成図。
【図５】多チャネルエコー消去系の一般的な構成を示す図。
【図６】従来の多チャネルエコー消去装置の機能構成を示す図。[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention is applied to, for example, a communication conference system having a multi-channel sound reproduction system, and relates to a multi-channel echo canceling method for canceling acoustic echoes that cause howling and impair hearing, a device therefor, a program therefor, and a recording medium therefor. Things.
[0002]
[Prior art]
2. Description of the Related Art With the recent development of digital networks and high-efficiency audio-video coding technology, a multi-channel loudspeaker communication system that allows a plurality of people to easily participate and provide a more natural communication environment has been studied. To achieve this, it is necessary to consider the technical issues and solutions of multi-channel acoustic echo cancellation that cancels acoustic sneak from multiple speakers into the microphone.
A communication conference system including a reproduction system of N (≧ 2) channels and a sound collection system of M (≧ 1) channel cancels an acoustic echo by a configuration as shown in FIG. That received signal from each receiving terminal ₁ 1 to 1 _N is reproduced as an acoustic signal at the speakers ₂ 1 to 2 _N, the N-number of the acoustic echo path ₁₀ 1 to 10 via the _N respective microphone ₃ m (m = 1 , ..., M). And the reception terminal ₁ 1 to 1 _N for all N channels of the receiving side, an acoustic connect the N-channel echo canceling part ₄ 1 to 4 _M between the transmitter terminal ₅ 1 to 5 _M each of M channel transmitting end Cancel the echo.
[0003]
The N-channel echo canceller 4 _m has a configuration to process the N inputs and one output time-series signal between total N-channel sound collection side of one channel of the reproduction side for each sound collection channel. FIG. 6 shows the configuration of the N-channel echo canceling unit 4 _m (m = 1,..., M).
Each received signal _{x 1 (k), x 2} (k), ..., x N (k) , respectively the adaptive filter _{11 1,} 11 2, ..., 11 are filtered by _N pseudo-echo signal _{y ^} 1m (k) , Y ＾ _2m (k),..., Y ＾ _Nm (k) are generated, and these pseudo echo signals are subtracted from the picked-up signal y _m (k) of the microphone 3 _m by subtraction units 12 ₁ , 12 _{2 ,.} in sequentially subtracted, the echo signal is output to the transmitter terminal 5 _m are suppressed. In correction coefficient calculating unit 13, the received signals _{x 1 (k), ...,} x N (k) and residual echo signal (error signal) and _{e m} is input each adaptive filter output to the transmitter terminal _{5 m} The correction coefficients for the filter coefficients are calculated. That is, the following received signal vector is generated from each received signal.

L is the number of taps (filter length) of the adaptive filters 11 ₁ ,..., 11 _N.
[0004]
The received signal vector x ₁ (k),..., X _N (k) is defined as a vector X (k) of the following equation.
^{_{X (k) = [x 1}} T (k), x 2 T (k), ..., x N T (k)] T
Each adaptive filter ₁₁ 1, ..., _{11 N} of the filter coefficient vector _{h ^ 1m (k), h} ^ 2m (k), ..., to combine _{h ^} Nm (k)
^{_{h ^ m (k) = [}} h ^ 1m T (k), h ^ 2m T (k), ..., h ^ Nm T (k)] T
And the adaptive filter coupling coefficient vector hベ ク ト ル_m (k) are updated, for example, by using the NLMS algorithm (learning identification method).
_{h ^ m (k + 1)} = h ^ m (k) + αe m (k) x (k) / (x T (k) x (k))
α is a step size parameter.
It becomes. That correction coefficient calculation unit 13 in the second term on the right side .alpha.e _m of the equation ^{(k) x (k) /} (x T (k) x (k)) by calculating the respective adaptive filter ₁₁ 1, ..., 11 _N obtain the correction coefficients of the filter coefficients for the adaptive filter ₁₁ 1 these correction factors, ..., 11 each coefficient update unit ₁₄ 1 each filter coefficients of _N, ..., to update at 14 _N. Adaptive filter ₁₁ 1, ..., 11 filter processing unit ₁₅ 1 using the filter coefficients in _N, ..., 15 the inner product calculation _(filtering) of the filter coefficient vector and the reception signal vector _{N y ^ 1m (k) =} h ^{_{^ 1m T (k) x 1}} (k), ..., y ^ Nm (k) = h ^ Nm T (k) x N (k) is performed, the pseudo echo signal is generated.
[0005]
Each filter coefficient vector is

By repeating the above correction, these filter coefficient vectors are converted into the impulse responses h _1m (k, n),..., H _Nm of each echo path from each speaker 2 ₁ ,..., 2 _N to the microphone 3 _m . Echo path vector with (k, n) time series as each element

Can be matched with, it is possible to reduce the residual signal e _{m (k).}
[0006]
Note that the original adaptive filter is configured to include each filter processing unit, its coefficient updating unit, and the corresponding correction coefficient calculation unit. For convenience, the filter processing unit and the coefficient updating unit are referred to as adaptive filters.
[0007]
[Problems to be solved by the invention]
In multi-channel audio system consisting reproduction system and sound collection system of the speaker 2 _1, ..., and volume during a 2 _N is large, there is a small, likewise microphone 3 _1, ..., 3 _M in Some have a large sound collection gain and some have a small sound collection gain. The level of the echo signal in which the reproduced sound from the speaker having a small volume is mixed into the microphone having a small sound collection gain is small. Therefore, in order to subtract this echo signal to a certain predetermined level, the length of the adaptive filter may be short. Conversely, the level of an echo signal in which a reproduced sound from a loud speaker is mixed into a microphone having a large sound pickup gain becomes large, and to attenuate the echo signal to the predetermined level, it is necessary to lengthen the adaptive filter length. .
[0008]
However, conventionally, the filter lengths of the adaptive filters 11 ₁ ,..., 11 _N are all the same. For this reason, an adaptive filter corresponding to the echo path of the echo signal having a small level is provided with an unnecessarily large filter length, so that unnecessarily many computations are performed or the hardware scale is unnecessarily increased. Was. Conversely, an adaptive filter corresponding to the echo path of an echo signal having a large level is given an insufficient filter length, and the echo signal may not be sufficiently suppressed. This problem is not limited to the case where the NLMS algorithm is used, and the same can be said for a case where the filter coefficients of the adaptive filter use another adaptive algorithm that approaches the impulse response of the echo path.
[0009]
SUMMARY OF THE INVENTION It is an object of the present invention to provide a multi-channel echo canceling method, an apparatus, a program, and a recording medium for giving an appropriate filter length to each adaptive filter when a filter length of the entire adaptive filter is given in a multi-channel echo canceling system. Is to provide.
[0010]
[Means for Solving the Problems]
According to the present invention, for each channel, the amount of acoustic coupling between the speaker that reproduces the reception signal of the channel and the microphone that outputs the pickup signal, that is, the indirect sound between the reproduction signal of the reception signal and the pickup signal In the logarithmic domain, a value obtained by adding the acoustic coupling amount due to the indirect sound of each channel and the received signal amount of the channel and subtracting the target error signal amount therefrom is defined as the target indirect sound suppression amount. Then, the adaptive filter length of the channel is determined in proportion to the target indirect sound suppression amount of the channel.
[0011]
BEST MODE FOR CARRYING OUT THE INVENTION
Given the filter length of the entire adaptive filter, reference is made to FIG. 1 which illustrates what may be considered as echo cancellation to provide an appropriate filter length for each adaptive filter. In FIG. 1, portions corresponding to those in FIG. 6 are denoted by the same reference numerals, and redundant description will be omitted. The Ru omitted correction coefficient calculating unit 13 tare.
Each adaptive filter 11 _1, ..., 11 _N corresponds to the filter length determining unit 21 _1, ..., 21 _N are provided, the filter length determining unit 21 ₁ is indirect sound acoustic coupling amount calculating section 22 and the filter length calculating section 23 Consists of The filter length determining units 21 ₂ ,..., 21 _N are each also formed of an indirect sound / acoustic coupling amount calculation unit 22 and a filter length calculation unit 23, although not shown.
[0012]
The principle of obtaining the adaptive filter length will be described below. Reverberation curve C _i in the echo path from the speaker 2 _i to the microphone 3 _m (k) is the impulse response h _i corresponding to the echo path, it is obtained by squaring integration from the last.
C _i (k) = {hj (k + j) ² ,} is the sum of j = 0, 1, ² ,.
Since C _i (0) is the square integral of the impulse response, it is equal to the acoustic coupling amount A _i between the speaker 2 _i and the microphone 3 _m . D _i and by direct sound of the acoustic coupling amount A _i, if the the R _i by indirect sound is A _i = D _i + R _i. When two clogging echo path speakers 1 ₁ and 1 ₂ are two, the reverberation curve C _1, C ₂ of the echo path is as shown in FIG. In FIG. 2, the horizontal axis is the time axis, and the vertical axis is the reverberation volume. The values C ₁ (0) and C ₂ (0) at the time k = 0 are the square integration values of the impulse responses of the respective echo paths, and are the acoustic coupling amounts A ₁ and A ₂ . The flat portions of the curves C ₁ and C ₂ are echo delay times, which are clearly shown, that is, separated from the vertical axis, but are actually significantly shorter than the filter lengths L ₁ and L _2. Thus, for example, several ms to several tens ms, which is negligible. The flat part near k = 0 also has a slight inclination due to noise or the like, but is smaller than the inclination due to a direct sound or an indirect sound and may be regarded as flat. E in FIG. 2 is a target acoustic coupling amount, the echo signal to the reverberation of the echo path becomes the target acoustic coupling amount E, the adaptive filter 11 _1, 11 ₂ may be simulated, respectively. That is, the time L ₁ , L ₂ from when the reproduced sound from the speakers 2 ₁ , 2 ₂ is generated to when the reverberation level of the echo path reaches the target acoustic coupling amount E is defined as the filter length of the adaptive filters 11 ₁ , 11 ₂ . What should I do?
[0013]
The reverberation curves C ₁ and C ₂ are values obtained by subtracting the acoustic coupling amounts D ₁ and D ₂ due to the direct sound from the acoustic coupling amounts A ₁ and A ₂ , that is, the acoustic coupling amounts R ₁ and R ₂ due to the indirect sound (logarithmic values). Decreases linearly with time in the logarithmic domain. In the same room, the inclination of the decrease in the reverberation curve due to the indirect sound is constant regardless of the positions of the speaker and the microphone. Therefore, as described above, since the reverberation curve can be considered to decrease linearly from the time of approximately k = 0, assuming that the entire filter length L = L ₁ + L ₂ ,
L ₁ : L ₂ ≡log R ₁ -log E: log R ₂ -log E
Holds. Therefore, the filter length _Li is obtained by the following equation.
[0014]

It becomes. The filter length L _i of each adaptive filter 11 _i in a general case where the number of speakers is N is obtained by the following equation.
L _i = [(log R _i -log E) / Σ (log R _i -log E)] L
Σ is the sum from i = 1 to N.
As understood from the above description, the overall filter length L to the filter length calculating unit 23 in the filter length determining unit 21 ₁ for example, in FIG. 1, a target acoustic coupling amount E is set input, indirect sound acoustic coupling amount calculated acoustic coupling amount _{R 1} by indirect sound from part 22 is also inputted, the target indirect sound coupling suppressing amount log _R 1 -log E is calculated in the subtraction portion 23a. Similarly the filter length determining unit ₂₁ 2, ..., respectively target indirect sound coupling suppressing amount _{21 N log R 2 -log E,} ..., log R N -log E is calculated. These objectives indirect sound coupling suppressing amount _{log R 1 -log E, ...,} log R N -log adaptive filter ₁₁ 1 corresponding in proportion respectively to E, ..., as indicated by the broken line in FIG its filter length 11 _N May be set.
[0015]
Adaptive filter ₁₁ 1, ..., the filter length _L 1 of 11 _N, ..., further as follows when the sum L of _{L N} are determined. The target indirect sound coupling suppression amounts log R ₁ -log E,..., Log R _N -log E are added by the adding unit 24, that is, _{ｉ i = 1} ^N (log R _i -log E) = P _REA is obtained. This added value P _REA is input to each filter length calculation unit 23 of each filter length determination unit 21 ₁ ,..., 21 _{N. In} each filter length calculation unit 23, (log R _i −log E) / P _REA = _{L i} is calculated respectively, and these calculated _L 1, ..., adaptive filter _{11 1 L N} correspond respectively, ..., it is set as the filter length 11 _N.
[0016]
In this way, each of the adaptive filters 11 ₁ ,..., 11 _N is set to a filter length corresponding to its echo level.
Here, the following method can be considered to obtain the acoustic coupling amount R _i due to the indirect sound in the indirect acoustic coupling amount calculating unit 22.
As shown in FIG. 3A, the impulse response measurement unit 25 measures an impulse response for each echo path in advance, and uses the impulse response. That the speaker 2 _1, ..., 2 from each one of the _N to regenerate the impulse sound, seek h _i at each time in the impulse response by measuring the attenuation state of the power output of the microphone 3 _m at that time, the measurement The impulse response thus obtained is input to the indirect sound / acoustic coupling amount calculation unit 22, and the square integration value C _i (0) of the impulse response is obtained in the calculation unit 22, as described above, and this value C _i (0) = from a _i, measured power h _{i (0)} at time k = 0 of the impulse response ^2, i.e. the acoustic coupling amount D _i by the direct sound by subtracting the logarithmic domain determine the acoustic coupling amount R _i by indirect sound.
[0017]
Or by entering the filter coefficients _{h i} from the adaptive filter 11 _i as shown in FIG. 3B, obtains a square integration value _C i (0) as in the case of FIG. 3A, _h i ⁽⁰⁾ from this value further ² = D _i is subtracted in the logarithmic domain to determine the acoustic coupling amount R _i given to the indirect sound.
As understood from the characteristics shown in FIG. 2, the smallest integer k such that C _i (k) / C _i (k + τ)> θ is considered as the arrival time of the direct sound. Here, τ is a value corresponding to several ms, and θ is about 2 to 4. That is, in FIG. 2, a time k at which the level changes rapidly from a direct sound to an indirect sound (reverberation) and rapidly decreases is obtained, and the acoustic coupling by the direct sound is set to D _i = C _i (k) / C _i (k + τ). Since FIG. 2 is a logarithmic value, it is subtracted. Thus indirect sound _{R i} is determined by _{R i = C i (0)} / B i. If the direct sound is not satisfied arrival also the time of should have gone past k is the previous conditions, to determine the indirect sound is dominant, D _i is set in the middle of 1 and θ.
[0018]
Alternatively, as shown in FIG. 3C, when each received signal x ₁ (k),..., X _N (k) is input to the covariance calculation unit 26 and an input signal sequence vector having these as elements is X, X , The temporal average value r (k) of the column vector having the covariance of If the received signal is a two _x 1 (m) and _{x 2 (m) r (k} ) = αr (k-1) + (1-α) [x 1 (k) * x 1 (k), x ^{_{1 (k) * x 2 (}} k), x 1 (k) x 2 (k), x 2 (k) * x 2 (k)] T
^T represents a transposed matrix, and ^* represents a conjugate. 0 ≦ α <1
It is. Microphone _{3 m} of the output _y m (k) to be inputted to the mean square value calculating section 27 mean square ^{y 2 (k) = βy 2} (k-1) + (1-β) y 2 m (k ), 0 ≦ β <1
Ask for. r (k) and y ² (k) are input to the acoustic coupling estimation unit 28, and the following equation is calculated.
[0019]
^{min (E [y 2 (k} ) -g T (k) r (k) 2], g T)
Here, E [] represents an operation for averaging, and E [a (k)] = Σw (i) a (ki), i = 0, 1,. As the window function w (), a rectangular window of 1 / N for a certain interval N and 0 for the other, or an exponential decay window of w (i) = λ ⁱ , (0 <λ <1) are often used. Can be
min (J, a) means an operation of selecting a so that J is minimized. The column vector g ^{T is determined} by a least squares method, a recursive least squares method, a conjugate gradient method, an adaptive filtering method, or the like so that the square of the difference between y ² (k) and g ^T (k) r (k) is minimized. Find (k). Here, g is a column vector that associates y _m (k) with x ₁ ,..., X _N. If the received signal is a two x ₁ and x _2, g represents the following formula.
[0020]
g = [G ₁₁ G ₁₁ ^* , G ₁₂ G ₁₁ ^* , G ₁₁ G ₁₂ ^* , G ₁₂ G ₁₂ ^* ] ^T
Here _G 11 ^{G *} is the acoustic coupling amount between the speaker _{2 1} and the microphone _{3 m,} the _G 12 _{G 12} ^* is an acoustic coupling amount between the speaker _{2 2} and the microphone _{3 m.}
Thus each loudspeaker 2 _{1 obtained,} ..., 2 acoustic coupling amount A ₁ between _N and the microphone 3 _m, ..., suitable values regarded as acoustic coupling amount by direct sound by the subtraction unit 29 from the A _N D _i Is subtracted in the logarithmic domain to obtain the acoustic coupling amount R _i due to the indirect sound. Acoustic coupling amount _{D i} by the direct sound about one order of magnitude compared to the log _{R i} -log E i is smaller, a value of about several dB can be predicted empirically, the values of the degree _{D i} 'May be set and input to the subtraction unit 29. In order to obtain g ^T (k) more correctly, the minimum value or mode value (for example, by making a histogram) of G ₁₁ G ₁₁ ^* in sequentially obtained g ^T (k) is obtained, and the value of gG ₁₁ G ₁₁ ^* is obtained. g ^T (k) may be used.
[0021]
In this case the filter length determining unit 21 _1, ..., acoustic coupling amount R ₁ by indirect sound used in the 21 _N, ..., R _N is obtained by one single indirect sound acoustic coupling amount calculation unit 22.
According to the method shown in FIGS. 3B and 3C, it is possible to adaptively change the filter length suitable for each channel when the amount of acoustic coupling changes while using the multi-channel echo canceller.
Next, an embodiment of the present invention is shown in FIG. 4 by attaching the same reference numerals to parts corresponding to FIG. Filter length determining unit 31 ₁ in this embodiment, ..., 31 _N adaptive filter 11 _1, ..., provided in correspondence with the 11 _N. Filter length determining unit 31 _1, ..., 31 _N are the same configuration, more composed and indirect sound acoustic coupling amount calculating section 22 and the filter length calculating unit 32 as shown in the filter length determining unit 31 _1. The indirect sound / acoustic coupling amount calculation unit 22 is the same as that in FIG. 1, but the filter length calculation unit 32 includes the indirect sound / acoustic coupling amount R ₁ , the overall filter length L, and the target error signal level E _e . , received signals x ₁ corresponding is input.
[0022]
Target indirect sound reduction amount from the power _{X i} and _{E e} and _{R i} of the received signal _{x i} by the filter length calculating unit 32 in the adder portion 32a log _R i + _log X i -log E _e is calculated. In this case the target indirect sound suppressing amount _{Elog R i + log X i -log} E adapting the value proportional to _e filter ₁₁ 1, as shown in broken line in the figure, ..., 11 respectively to the _N filter length _L 1, ..., L _N may be set. If the entire filter length L is determined, the following is further performed. The target indirect sound suppression amount log R _i + log X _i −log E _e from each filter length determining unit 31 ₁ ,..., 31 _N is added by the adding unit 33, and the added value P _REA ′ is added to each filter length calculating unit 32 Supplied to
[0023]
Each filter length calculating unit 32 calculates the division portion _{32b (log R i + log X} i -log E e) / P REA '= L, sets the _{L i} to the filter length of the corresponding adaptive filter 11 _i I do.
To determine in this case the adaptive filter length L _i, because the have received signals x _i be considered, for example, when the received signal x _i is the large level accordingly, the filter length of the corresponding adaptive filter is long, the The echo signal can be sufficiently suppressed. Various methods shown in FIG. 3 can be used for estimating the acoustic coupling amount R _i by the indirect sound.
[0024]
In the above description, various calculations are performed in the logarithmic domain, but may be performed in the linear domain. Subtraction in the logarithmic domain is division in the linear domain.
The implementation mode of the invention described above can be made to function by executing the program by a computer. That is, a program for that purpose is installed from a CD-ROM, a floppy (registered trademark) disk, a magnetic disk, or a storage medium stored at another location through a transmission line into an operation program memory in a computer. The program on the program memory may be executed by a computer to function.
[0025]
【The invention's effect】
As described above, according to the present invention, when the filter length of the adaptive filter used for all channels is given, that is, when the calculation amount is given, a filter suitable for each channel is set, and unnecessary There is no fear that an adaptive filter having a long filter length is generated or that the echo signal cannot be sufficiently suppressed due to a short adaptive filter length.
In particular, since the level of the received signal is also taken into consideration, an appropriate filter length can be set according to the volume of the utterance of the speaker.
[Brief description of the drawings]
FIG. 1 is a functional configuration diagram showing a multi-channel echo canceller capable of setting an appropriate filter length .
FIG. 2 is a diagram showing an example of a reverberation curve for explaining the principle of the present invention.
FIG. 3 is a diagram showing a functional configuration of a specific example for obtaining an acoustic coupling amount due to an indirect sound.
[4] functional configuration diagram showing an embodiment of the present invention.
FIG. 5 is a diagram showing a general configuration of a multi-channel echo canceling system.
FIG. 6 is a diagram showing a functional configuration of a conventional multi-channel echo canceller.

Claims (8)

The received signal of each channel is input to the adaptive filter, a pseudo echo signal is obtained as an output thereof, the pseudo echo signal is subtracted from the collected signal, the remaining error signal obtained by the subtraction, the received signal of the channel and in the echo canceller filter coefficients of the adaptive filter that corresponds is updated from,
Provided for each of the above channels,
An impulse response between a reproduction signal of a reception signal of the channel and the collected signal or a filter coefficient of a corresponding adaptive filter is input, and an acoustic coupling amount of an indirect sound between the reproduction signal and the collected signal is calculated. Means to
Means for receiving the acoustic coupling amount R _i of the indirect sound, the reception signal X _{i of the} channel thereof, and the target error signal amount E _e, and outputting a result of calculating R _i + X _i −E _e in the logarithmic domain;
The result of R _i + X _i −E _e Means for setting a filter length in proportion to the adaptive filter of the channel to the multi-channel echo canceller. The received signal of each channel is input to the adaptive filter, a pseudo echo signal is obtained as an output thereof, the pseudo echo signal is subtracted from the collected signal, the remaining error signal obtained by the subtraction, the received signal of the channel and in the echo canceller filter coefficients are updated in the corresponding adaptive filter from,
A covariance calculator for receiving a received signal of each channel and calculating a temporal average value r (k) of a column vector having a covariance of a received signal column vector x having the received signal as an element;
A mean-square calculating unit to which the picked-up signal y _m (k) is input and for calculating a temporal average y ² (k) of the squared value;
The average value r (k) and the average value y ² (k) are input, and the column vector g ^T (k) that minimizes the square of the difference between y ² (k) and g ^T (k) r (k) And a sound coupling estimating unit for obtaining a sound coupling amount between the reproduction signal of the reception signal of each channel and the collected sound signal as an element of the column vector g ^T (k);
Means for receiving the acoustic coupling amount of each channel and a predetermined value, calculating and outputting the acoustic coupling amount of the indirect sound between the reproduction signal and the collected sound signal of each channel,
Means for inputting the acoustic coupling amount and the target acoustic coupling amount of each of the indirect sounds, obtaining and outputting the target indirect sound coupling suppression amount from these,
Means for setting a filter length in proportion to the target indirect sound coupling suppression amount of each channel to an adaptive filter of the channel. The received signal of each channel is input to the adaptive filter, a pseudo echo signal is obtained as an output thereof, the pseudo echo signal is subtracted from the collected signal, the remaining error signal obtained by the subtraction, the received signal of the channel and in the echo canceller filter coefficients of the adaptive filter that corresponds is updated from,
A covariance calculator for receiving a received signal of each channel and calculating a temporal average value r (k) of a column vector having a covariance of a received signal column vector x having the received signal as an element;
A mean-square calculating unit to which the picked-up signal y _m (k) is input and for calculating a temporal average y ² (k) of the squared value;
The average value r (k) and the average value y ² (k) are input, and the column vector g ^T (k) that minimizes the square of the difference between y ² (k) and g ^T (k) r (k) And a sound coupling estimating unit for obtaining a sound coupling amount between the reproduction signal of the reception signal of each channel and the collected sound signal as an element of the column vector g ^T (k);
A means for calculating an acoustic coupling amount of the indirect sound between the reproduction signal and the collected sound signal of each channel, wherein the acoustic coupling amount of each channel and a predetermined value are input,
Means for receiving the acoustic coupling amount R _i of the indirect sound, the reception signal X _{i of the} channel thereof, and the target error signal amount E _e, and outputting a result of calculating R _i + X _i −E _e in the logarithmic domain;
The result of R _i + X _i −E _e Means for setting a filter length in proportion to the adaptive filter of the channel to the multi-channel echo canceller. The received signal of each channel is processed by an adaptive filter to generate a pseudo echo signal, the pseudo echo signal is subtracted from the collected signal, and a corresponding adaptive signal is obtained from the remaining error signal and the received signal of the channel. In an echo cancellation method for updating a filter coefficient of a filter,
For each channel, from an impulse response or a corresponding filter coefficient of the adaptive filter between the reproduction signal and the upper KiOsamuoto signal of the receiving signal of the channel, the indirect sound between the reproduced signal and the collected sound signal Find the acoustic coupling amount,
The acoustic coupling amount R _i of the indirect sound R _i + X _i −E _e in the logarithmic domain based on the received signal X _{i of the} channel and the target error signal E _e ,
A multi-channel echo canceling method characterized by updating a filter length of an adaptive filter of the channel in proportion to the calculation result . The reception signals of the channels more processes, each adaptive filter generates a pseudo echo signal, Pull insert the pseudo echo signal from the sound collection signal from the remaining error signal obtained by subtracting the, the received signal of the channel in echo cancellation method to update the filter coefficients of the corresponding adaptive filter,
Temporal average value calculated Me a r (k) of the column vector and covariance elements of the received signal column vector x of the received signal as an element from the received signal of each channel,
The collected sound signal y _m square value temporal average value y ² (k) the determined eye (k),
A column vector g ^T (k) that minimizes the square of the difference between the temporal average y ² (k) of the squared value and g ^T (k) r (k) is obtained, and the column vector g ^T ( k) obtaining the acoustic coupling amount between the reproduction signal of the reception signal of each channel and the collected signal as an element of k) ;
Said from the acoustic coupling amount and the predetermined value for each channel, the acoustic coupling amount determined Me indirect sounds between the reproduced signal and the collected sound signal of each channel,
Acoustic coupling amount of each of the above indirect sound and the target acoustic coupling amount by Ri target indirect sound coupling suppressing the amount of the required Me,
A filter length that is proportional to the target indirect sound coupling suppressing amount of each channel, multi-channel echo cancellation method comprising users update the filter length of the adaptive filter of that channel. The reception signals of the channels more processes, each adaptive filter generates a pseudo echo signal, Pull insert the pseudo echo signal from the sound collection signal from the remaining error signal obtained by subtracting the, the received signal of the channel in echo cancellation method to update the filter coefficients of the adaptive filter that corresponds,
Temporal average value calculated Me a r (k) of the column vector and covariance elements of the received signal column vector x of the received signal as an element from the received signal of each channel,
The collected sound signal y _m square value temporal average value y ² (k) the determined eye (k),
A column vector g ^T (k) that minimizes the square of the difference between the temporal average value y ² (k) of the square value and g ^T (k) r (k) is obtained, and the column vector g ^T As an element of (k), an acoustic coupling amount between a reproduction signal of a reception signal of each channel and the collected signal is obtained,
From the acoustic coupling amount of each channel and a predetermined value , the acoustic coupling amount of the indirect sound between the reproduction signal and the collected signal of each channel is determined ,
And the reception signal X _i of the channel and the acoustic coupling amount R _i of the indirect sound, the target error signal amount E _e, in the log domain to calculate the R _i + X _i -E _e,
A filter length proportional to the calculated result of the R _i + X _i -E _e, multi-channel echo cancellation method comprising the Turkey to update the filter length of the adaptive filter of that channel. Program for executing the respective programs of the methods described in the computer in any one of claims 4 to 6. A computer-readable recording medium on which the program according to claim 7 is recorded.

Images