JP2004274412A - Echo canceler, canceling method, and canceling program - Google Patents

Echo canceler, canceling method, and canceling program Download PDF

Info

Publication number
JP2004274412A
JP2004274412A JP2003062656A JP2003062656A JP2004274412A JP 2004274412 A JP2004274412 A JP 2004274412A JP 2003062656 A JP2003062656 A JP 2003062656A JP 2003062656 A JP2003062656 A JP 2003062656A JP 2004274412 A JP2004274412 A JP 2004274412A
Authority
JP
Japan
Prior art keywords
sequence
signal
simulation
reverberation
conversion
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
Application number
JP2003062656A
Other languages
Japanese (ja)
Other versions
JP3920795B2 (en
Inventor
Suehiro Shimauchi
末廣 島内
Yoichi Haneda
陽一 羽田
Akitoshi Kataoka
章俊 片岡
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nippon Telegraph and Telephone Corp
Original Assignee
Nippon Telegraph and Telephone Corp
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Nippon Telegraph and Telephone Corp filed Critical Nippon Telegraph and Telephone Corp
Priority to JP2003062656A priority Critical patent/JP3920795B2/en
Publication of JP2004274412A publication Critical patent/JP2004274412A/en
Application granted granted Critical
Publication of JP3920795B2 publication Critical patent/JP3920795B2/en
Anticipated expiration legal-status Critical
Expired - Fee Related legal-status Critical Current

Links

Images

Landscapes

  • Cable Transmission Systems, Equalization Of Radio And Reduction Of Echo (AREA)
  • Telephone Function (AREA)

Abstract

<P>PROBLEM TO BE SOLVED: To quicken convergence of a filter coefficient being determined without using a conversion region echo simulation filter coefficient restricting means requiring a large operational amount including inverse linear transform, windowing, and linear transform processing. <P>SOLUTION: The echo canceler comprises a means for obtaining a reproduction signal sequence by storing sound reproduction signals for a predetermined time, a means for obtaining a sound collection signal sequence by storing sound collection signals for a predetermined time, a means for obtaining a reproduction signal conversion sequence by performing linear conversion on the reproduction signal sequence, a means for generating a simulation echo signal conversion sequence simulating the linear conversion sequence of an echo signal leaking from the sound signal reproducing means to the sound signal collecting means by receiving the reproduction signal conversion sequence, a means for receiving the simulation echo signal conversion sequence and the sound collection signal sequence and outputting the simulation difference of the simulation echo signal conversion sequence, and a means for generating an updated sequence of a conversion region echo simulation filter coefficient having any one or both of a means for providing nonlinear processing to a reproduction signal conversion sequence being fed to a conversion region echo simulation filter coefficient updating means and a simulation error nonlinear processing means for providing nonlinear processing to a simulation error. <P>COPYRIGHT: (C)2004,JPO&NCIPI

Description

【0001】
【発明の属する技術分野】
本発明は、相手話者からの受話再生音声を送話音声と重畳して収音することによって生じる音響エコー(受話再生音声)を除去するために、模擬誤差信号に基づいて擬似反響路を適応的に推定し、擬似反響路に基づく擬似反響を収音信号から除去して送話音声とする反響消去装置、方法、及び反響消去プログラムに関する。
【0002】
【従来の技術】
スピーカ1からマイクロホン2へ回り込む反響を消去する反響消去装置は、図1のように接続される。
従来、装置内では、スピーカとマイクロホン間の反響路のインパルス応答hを推定し、推定したインパルス応答h’と再生信号xの畳み込み演算h’*xにより模擬反響信号y’を生成し、実際の反響信号yから減算することで、反響消去信号eを得る。しかし、推定したインパルス応答と再生信号の畳み込み演算には、多くの演算量を必要とし、実装上の問題となっている。近年、この問題を解決するために、再生信号や反響信号を一旦線形変換し、反響路のインパルス応答の線形変換に対応したパラメータを推定し、畳み込みの代わりに乗算処理を用いたり(非特許文献1 参照)、あるいはより小さい畳み込み演算に分割したりする(特許文献1 参照)などをして、演算量を削減する方法が提案されている。線形変換の例としては、(高速)離散フーリエ変換、(高速)離散コサイン変換、(高速)ハートレー変換などがある。
図1で音響信号再生手段としてスピーカを挙げているが、音響信号再生手段としては、再生前段の増幅器やバッファも含む。また、同様に音響信号収音手段は、マイクロホンの後段の増幅器やバッファも含む。
【0003】
【非特許文献1】
Simon Haykin著「適応フィルタ理論 科学技術出版、2001年1月10日 p.500−541
【特許文献1】
特開平9−116472号公報(図6)
【0004】
【発明が解決しようとする課題】
従来の技術で挙げた非特許文献1に記載されている2つの構成を図2、図3に示す。図2の構成は拘束付の構成と呼ばれ、図3の構成は拘束の無い構成と呼ばれている。
図2、図3の構成は、ともに以下の手段を含んでいる。つまり、音響信号再生手段1へと出力される再生信号を入力し一定時間蓄積し再生信号列を得る再生信号入力手段101と、前記音響信号再生手段1と同一空間に存在する音響信号収音手段2から収音信号を入力し一定時間蓄積し収音信号列を得る収音信号入力手段102と、前記再生信号列を線形変換し再生信号変換列を得る再生信号変換手段103と、前記再生信号変換列を入力し前記音響信号再生手段1から前記音響信号収音手段2へと回り込む反響信号の線形変換列を模擬する模擬反響信号変換列を生成する模擬反響生成手段(変換領域反響模擬フィルタ)104と、前記模擬反響信号変換列と前記収音信号列とを入力し模擬反響信号変換列の模擬誤差を出力する模擬誤差出力手段105と、前記再生信号変換列と前記模擬誤差を入力し前記模擬反響生成手段(変換領域反響模擬フィルタ)104の変換領域反響模擬フィルタ係数の更新列(変換領域反響模擬フィルタ係数の配列の線形(周波数)変換に相当する配列)を生成する変換領域反響模擬フィルタ係数更新手段106を含む。
模擬反響生成手段104は、変換領域反響模擬フィルタ係数更新手段106から得られる変換領域反響模擬フィルタ係数更新列をもとに、変換領域反響模擬フィルタ係数を書き換える手段を含む。変換領域反響模擬フィルタ係数更新手段106に入力されるために模擬誤差出力手段105から出力される模擬誤差は、線形変換領域で評価できる形式となっている。
【0005】
図2(従来例1)と図3(従来例2)の構成の違いは、従来例1の構成が変換領域反響模擬フィルタ係数拘束手段107を有する点である。
変換領域反響模擬フィルタ係数拘束手段107の詳細は図4に示される。
非特許文献1では、図4の逆線形変換手段71を(高速)逆離散フーリエ変換で実現し、線形変換手段73を(高速)離散フーリエ変換で実現している。また、窓掛け手段72では、変換領域反響模擬フィルタ係数更新列の逆線形変換列の半分を強制的に零としている。この拘束付けの根拠は、求めるべき変換領域反響模擬フィルタ係数が、真の反響路の長さLのインパルス応答と同一の長さLの零信号列を結合した長さ2Lの信号列の線形変換に相当することを、更新時に反映させることにある。非特許文献1で、変換領域反響模擬フィルタ係数拘束手段の無い従来例2の構成では、変換領域反響模擬フィルタ係数の求めるべき値への収束が遅くなることが指摘されている。
しかしながら、演算量の観点からは、線形逆変換、窓掛け、線形変換の一連の処理は、望ましいものではない。そこで、本発明は、図4に示すような演算量の多い変換領域反響模擬フィルタ係数拘束手段を用いずに、求めるべき変換領域反響模擬フィルタ係数への収束を速めることを課題とする。
【0006】
【課題を解決するための手段】
非特許文献1で図2,図3の反響消去装置に適用されている高速LMS(Least mean Square)アルゴリズムでは、参照信号である再生信号と、真の反響に対する模擬反響の模擬誤差が、無相関となるように変換領域反響模擬フィルタ係数を更新する。
ここで、変換領域反響模擬フィルタ係数拘束手段が無い場合、変換領域反響模擬フィルタ係数の逆線形変換の信号列の半分が零になっている保障はない。この非零の影響は、模擬誤差に巡回畳み込みによる折り返し成分として現れるが、この折り返し成分と、再生信号とは、通常無相関とみなせる。従って、無相関化、すなわち2次の統計量の評価により、変換領域反響模擬フィルタ係数を決定する高速LMSアルゴリズムでは、積極的に折り返しの影響を除去することができない。すなわち、非拘束なフィルタ更新による反響消去では、周波数領域変換において巡回畳み込み処理による折り返し成分が反響路成分の推定に影響し収束性能の低下を招く。
【0007】
従来例の高速LMSは、受話信号の変換係数の複素共役と、誤差信号の変換係数との積を用いて上記変換領域反響模擬フィルタ係数を更新するものであるのに対して、本発明では少なくとも、受話信号か、誤差信号の一方に、線形変換の前後いずれかで、非線形処理を施し、上記の更新用の積を求める。また、少なくとも一方に非線形処理を施した受話信号と、誤差信号との、変換(周波数)領域での積は、受話信号と誤差信号の高次の相関(高速LMSでは、2次(通常の相関))量を含む。つまり本発明は、変換領域反響模擬フィルタ係数の更新において非線形処理を施した受話信号と誤差信号との変換領域での積を用いるアルゴリズムであり、高速LMSの変形により可能になる。
【0008】
本発明では、模擬誤差に含まれる折り返し成分を、参照信号(再生信号)と独立化、すなわち高次の統計量の評価により、変換領域反響模擬フィルタ係数を決定することにより、変換領域反響模擬フィルタ係数拘束手段を用いずに、求めるべき変換領域反響模擬フィルタ係数への収束を速める。具体的には、図5に示すように、図3(従来例2)の構成に、変換領域反響模擬フィルタ係数更新手段106へ入力する前記再生信号変換列に非線形処理を与える再生信号非線形処理手段108と、前記変換領域反響模擬フィルタ係数更新手段106へ入力する前記模擬誤差に非線形処理を与える模擬誤差非線形処理手段109のうち、少なくともどちらかを含む構成である。再生信号非線形処理手段108、または、模擬誤差非線形処理手段109により、処理された信号列に対して、従来技術と同様にして、変換領域反響模擬フィルタ係数更新手段106により、変換領域反響模擬フィルタ係数の更新分に相当する更新列を計算する。但し、非特許文献1にも記載のある更新列の相対的大きさを調節するステップサイズパラメータの値は、必ずしも同じではない。
【0009】
本発明は、高次の統計量を用い、模擬誤差に含まれる折り返し成分を積極的に評価し、変換領域反響模擬フィルタ係数の更新列を決定できる。
【0010】
【発明の実施の形態】
(実施例1)
本発明の基本構成は、図5の構成となるが、容易に類推できる代替構成として、再生信号非線形処理手段108を、再生信号変換手段103よりも前段に配置する構成、すなわち再生信号列を線形変換する前に、非線形処理を施す構成がある。また、模擬誤差出力手段105において、模擬誤差を線形変換前の信号列に対して求める手段と、この線形変換前の模擬誤差列を線形変換する手段と、線形変換後の模擬誤差列を出力する手段を有する場合、模擬誤差非線形処理手段109は、模擬誤差出力手段105内の線形変換前の模擬誤差列を線形変換する手段の前段に配置する構成、すなわち、線形変換前の模擬誤差列に対して、非線形処理を施す構成も採り得る。
再生信号変換手段103の線形変換としてDFTのみならずDCT、ハートレイ変換が用いられる。なお、変換後の周波数領域係数で複素数となった場合での非線形処理の一例として、振幅を正規化して、位相項のみ取り出し、これにより信号の大きさによらず、信号の位相レベルで独立性や、相関を(正規化して)評価することができる。
【0011】
図5における再生信号非線形処理手段108、または模擬誤差非線形処理手段109における、非線形処理は、例えば、サイン関数や、双曲線正接関数が利用できる。
サイン関数は、ある入力要素zに対して、
【数1】

Figure 2004274412
で与えられる。また、双曲線正接関数は、zが複素数のとき、位相歪みが生じるため、サイン関数と組み合わせて
sgn(z)・tanh(|z|)
を用いることもできる。また、zのある範囲の値に対して任意のa(n=0,・・・,∞)によって、
【数2】
Figure 2004274412
と表せる任意の関数を用いることもできる。
再生信号非線形処理手段108と模擬誤差非線形処理手段109とは、同じ非線形関数により処理する場合と、異なる非線形関数により処理する場合とがある。
また、図5における再生信号非線形処理手段108、または模擬誤差非線形処理手段109において、入力される前記再生信号変換列あるいは前記模擬誤差の列の各要素ごとに、上に示した非線形関数から異なる関数を選択し、各要素ごとに異なる非線形処理を施す構成も採り得る。さらに、前記再生信号変換列あるいは前記模擬誤差の列の各要素の一部に対して、非線形処理を施さない構成も採り得る。例えば、線形変換が(高速)離散フーリエ変換によって実現される場合には、音声成分の少ない高い周波数に対応する線形変換列の要素に対する非線形処理を省略することなどが考えられる。
【0012】
(実施例2)
図6は、図5の再生信号非線形処理手段108と模擬誤差非線形処理手段109とにおいて、同じ非線形関数f(z)を用い、かつf(z)がf(a)f(b)=f(ab)を満足する場合、例えばサイン関数などを用いた場合の図5の代替実施例である。図5の再生信号非線形処理手段108と模擬誤差非線形処理手段109は、変換領域反響模擬フィルタ係数の更新列に対してまとめて非線形処理、すなわち、非線形処理後の受話信号(再生信号)と、非線形処理後の誤差信号との積が、処理前の受話信号と誤差信号の積に非線形処理したものと等しい場合は、後段にまとめて非線形処理を施す、例えば受話、誤差信号共にサイン関数を施す場合等、を行う更新列非線形処理手段110により代替されている。
【0013】
(実施例3)
図7は、図5の構成に対し、上記収音信号の短時間平均レベルの上記再生信号の短時間平均レベルに対する比として音響結合量を求める音響結合量測定手段111を追加した構成である。この音響結合量測定手段111により測定された音響結合量の大小に対応して、前記変換領域反響模擬フィルタ係数更新手段106では、前記変換領域反響模擬フィルタ係数の更新量の大小を調節するステップサイズパラメータを決定する。これは、例えば、再生信号非線形処理手段108と模擬誤差非線形処理手段109とにおいて、ともにサイン関数を採用した場合などには、変換領域反響模擬フィルタ係数の更新列の大きさが、再生信号や模擬誤差の大きさに依存しなくなる。
【0014】
(高速)LMS等では、誤差が小さくなるほど、適応フィルタの更新量(フィルタを修正するベクトルの大きさ)は小さくなり、収束が進むにつれて、より小差な修正が可能となる。その反面、本発明のように、サイン関数(出力が±1)を位相に適用した場合、適応フィルタの更新量は、誤差の大小に関わらず一定で、ある誤差範囲より精度良くは収束しない。音響結合利得が1のときは、ステップサイズを0.01としておけば、+0.01,−0.01で更新することになり、誤差1%の範囲で収束させることができる。逆に利得が10000の場合に、ステップサイズを0.01としたら、収束速度が非常に遅くなるので、この場合はステップサイズが100でも構わない。このように、利得に応じてステップサイズを調整する。その例として測定された音響結合量に推定誤差許容率を掛け合わせる等の方法により、上記の例では、音響結合10000×誤差許容率0.01(1%)=ステップサイズ100とする。
【0015】
このように図7の構成では、音響結合量が大きい場合は、相対的に大きなステップサイズパラメータを与え、音響結合量が小さい場合には、相対的に小さなステップサイズパラメータを与える。なお、音響結合量の測定は、図7においては、線形変換前の再生信号、収音信号を用いて実施しているが、線形変換後の再生信号、収音信号を用いた実施も容易に類推できる。このとき、例えば、(高速)離散フーリエ変換により線形変換を行った場合、各周波数成分ごとに音響結合量を測定し、これら各測定値に基づき、各周波数要素毎にステップサイズパラメータを与えることもできる。また、音響結合量の情報が測定により得られない場合であっても、高い周波数成分のステップサイズパラメータを小さくするなど、線形変換の各要素毎に異なるステップサイズパラメータ(事前に固定パラメータとして設定しておいても、音響結合量が周波数毎に得られた場合には、その都度算出される)を与える構成も採り得る。ここで、測定された音響結合量をステップサイズパラメータに反映させること自体は、一般に図1の構成をとる反響消去装置において、反響路のインパルス応答の推定に有効な方法といえる。
また、前記模擬反響生成手段104に過去の前記再生信号変換列を記憶する手段と前記記憶された再生信号変換列に対応した変換領域反響模擬フィルタ係数を有する場合には、前記再生信号変換列の記憶時期に従い記憶時期の古い前記再生信号変換列に対応する前記変換領域反響模擬フィルタ係数ほど更新量を小さくするなどの調節ができるように変換領域反響模擬フィルタ係数更新手段106を図8のような構成により実施することが考えられる。
【0016】
(変換領域反響模擬フィルタ係数更新手段(実施例1))
図8の構成では、1ステップ過去までの再生信号変換列を記憶できる構成となっているが、この構成は容易に複数ステップ過去の再生信号変換列を記憶した場合に拡張できる。図8において、現時刻での(非線形処理後の)再生信号変換列は信号列保持手段61に保持され、信号列保持手段62により保持されている(非線形処理後の)模擬誤差列と、信号列乗算手段63において、列の要素毎に乗算され、変換領域反響模擬フィルタ係数更新列が得られる。この変換領域反響模擬フィルタ係数更新列は、利得調節手段631により適切なステップサイズパラメータを個々の更新列要素毎に乗じられ、再度、変換領域反響模擬フィルタ係数更新列として、更新列保持手段65により保持され、模擬反響生成手段104に渡される。さらに1ステップ過去の(非線形処理後の)再生信号変換列は、信号列遅延器67を通して、信号列保持手段68に保持されており、信号列保持手段62により保持されている(非線形処理後の)模擬誤差列と、信号列乗算手段63において、列の要素毎に乗算され、1ステップ過去の(非線形処理後の)再生信号変換列に対応する変換領域反響模擬フィルタ係数更新列を得られる。この変換領域反響模擬フィルタ係数更新列は、利得調節手段691により利得調節手段631とは異なるステップサイズパラメータを個々の更新列要素毎に乗じられ、再度、変換領域反響模擬フィルタ係数更新列として、更新列保持手段65により、先に得られた現時刻の(線形処理後の)再生信号変換列に対応する変換領域反響模擬フィルタ係数とは独立して保持され、模擬反響生成手段104に渡される。ここで、利得調節手段691のステップサイズパラメータは、実際の反響路とインパルス応答の指数減衰性を考慮に入れると、利得調節手段631のステップサイズパラメータよりも小さくするほうがよい。また、図7のように、音響結合量測定手段111からの情報が得られる場合は、利得調節手段631と利得調節手段691とにおいて、ステップサイズパラメータの相対関係(大小関係)は保持したまま、音響結合量評価手段66により、各々のステップサイズパラメータをスケーリングさせることもできる。音響結合量測定手段111からの情報が得られない場合は、音響結合量評価手段66は省略できる。ここで、図8の構成は、非線形処理手段を含まない図2や図3の構成などにおいても、反響路のインパルス応答の特性推定の高速化または高精度化に効果があるといえる。
【0017】
(変換領域反響模擬フィルタ係数更新手段(実施例2))
さらに、前記模擬反響生成手段に過去の前記再生信号変換列を記憶する手段と前記記憶された再生信号変換列に対応した変換領域反響模擬フィルタ係数を有する場合において、前記再生信号変換列の記憶時期によって、異なる非線形処理を実現するために、再生信号非線形処理手段108と模擬誤差非線形処理手段109を有する図5の基本構成の代わりに、従来技術の図3の構成において、変換領域反響模擬フィルタ係数更新手段106の内部に、図9に示す非線形処理手段601、非線形処理手段602、非線形処理手段603、非線形処理手段604を配置した構成が考えられる。非線形処理手段601は、現時刻の再生信号変換列に非線形処理を実施し、非線形処理手段602は、1ステップ過去の再生信号変換列に非線形処理を実施し、非線形処理手段603は、現時刻の再生信号変換列に対応した模擬誤差の列に非線形処理を実施し、非線形処理手段604は、1ステップ過去の再生信号変換列に対応した模擬誤差の列に非線形処理を実施する。これら、4個の非線形処理手段は、異なる非線形関数により実現され得る。また、1部の非線形処理手段を省略した構成も採り得る。これにより、反響路のインパルス応答においてエネルギーの集中する前半と、あまりエネルギーの無い後半の推定精度、推定速度を個別に調節可能となる。また、図3の構成に対し、図7と同様に、音響結合量測定手段111を設ければ、図9内の音響結合量評価手段66によりステップサイズパラメータの制御も可能となるが、音響結合量測定手段111を設けられない場合は、音響結合量評価手段66は省略できる。図9の構成では、1ステップ過去までの再生信号変換列を記憶できる構成となっているが、この構成は容易に複数ステップ過去の再生信号変換列を記憶した場合に拡張できる。
【0018】
本発明の反響消去装置は、CPUやメモリ等を有するコンピュータと、利用者端末と、CD−ROM、磁気ディスク装置、半導体メモリ等の記録媒体とから構成することができる。
記録媒体に記録された反響除去プログラム、あるいは回線を介して伝送された反響除去プログラムは、コンピュータに読み取られ、コンピュータ上に前述した各構成要素を実現し、各処理を実行する。
【0019】
【発明の効果】
本発明による反響消去装置は、線形変換領域で真の反響路のインパルス応答に対応する変換領域反響模擬フィルタ係数を、演算量の増大する変換領域反響模擬フィルタ係数更新の拘束付けを省略しながらも、高速に求めることができる。また、非線形関数の導入により、高次統計量を評価して、再生信号と模擬誤差の独立性に基づき、変換領域反響模擬フィルタ係数を更新しているため、マイクロホンに混入する雑音や音声信号などが、再生信号と独立と見なせる場合は、それらの影響も受けにくい。
【図面の簡単な説明】
【図1】反響消去装置の概要構成を示す図。
【図2】変換領域反響模擬フィルタ係数拘束手段を有する反響消去装置(従来例1)の構成を示す図。
【図3】変換領域反響模擬フィルタ係数拘束手段の無い反響消去装置(従来例2)の構成を示す図。
【図4】従来例1の変換領域反響模擬フィルタ係数拘束手段の構成を示す図。
【図5】本発明の反響消去装置(実施例1)の構成を示す図。
【図6】本発明の反響消去装置(実施例2)の構成を示す図。
【図7】本発明の反響消去装置(実施例3)の構成を示す図。
【図8】変換領域反響模擬フィルタ係数更新手段(実施例1)の構成を示す図。
【図9】変換領域反響模擬フィルタ係数更新手段(実施例2)の構成を示す図。
【符号の説明】
100・・・反響消去装置、101・・・再生信号入力手段、102・・・収音信号入力手段、103・・・再生信号変換手段、104・・・模擬反響生成手段、105・・・模擬誤差出力手段、106・・・変換領域反響模擬フィルタ係数更新手段、108・・・再生信号非線形処理手段、109・・・模擬誤差非線形処理手段、110・・・更新列非線形処理手段、111・・・音響結合量測定手段[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention adapts a pseudo echo path based on a simulated error signal in order to remove an acoustic echo (received reproduction sound) generated by superimposing a reception reproduction sound from a partner speaker on a transmission sound and collecting the sound. The present invention relates to a reverberation elimination apparatus, method, and reverberation elimination program, which estimates a pseudo reverberation based on a pseudo reverberation path, and removes the pseudo reverberation from a picked-up signal to obtain a transmitted voice.
[0002]
[Prior art]
A reverberation canceling device for canceling reverberation circulating from the speaker 1 to the microphone 2 is connected as shown in FIG.
Conventionally, in a device, an impulse response h of a reverberation path between a speaker and a microphone is estimated, and a simulated reverberation signal y ′ is generated by a convolution operation h ′ * x of the estimated impulse response h ′ and a reproduction signal x. By subtracting from the echo signal y, an echo cancellation signal e is obtained. However, the convolution operation of the estimated impulse response and the reproduced signal requires a large amount of operation, which is a problem in mounting. In recent years, in order to solve this problem, a reproduction signal or a reverberation signal is once linearly transformed, a parameter corresponding to a linear transformation of an impulse response of a reverberation path is estimated, and multiplication processing is used instead of convolution (Non-patent Documents). 1) or a method of dividing into smaller convolution operations (see Patent Document 1), and the like, to reduce the amount of operation has been proposed. Examples of linear transformations include (fast) discrete Fourier transform, (fast) discrete cosine transform, and (fast) Hartley transform.
FIG. 1 shows a speaker as the sound signal reproducing means, but the sound signal reproducing means also includes an amplifier and a buffer in a stage before reproduction. Similarly, the sound signal collecting means also includes an amplifier and a buffer at the subsequent stage of the microphone.
[0003]
[Non-patent document 1]
Simon Haykin, “Adaptive Filter Theory, Science and Technology Publishing, January 10, 2001, pp. 500-541.
[Patent Document 1]
JP-A-9-116472 (FIG. 6)
[0004]
[Problems to be solved by the invention]
FIGS. 2 and 3 show two configurations described in Non-Patent Document 1 cited in the prior art. The configuration in FIG. 2 is called a configuration with constraints, and the configuration in FIG. 3 is called a configuration without constraints.
2 and 3 both include the following means. That is, a reproduction signal input unit 101 that inputs a reproduction signal output to the audio signal reproduction unit 1 and accumulates the reproduction signal for a certain period of time to obtain a reproduction signal sequence, and an audio signal sound collection unit that exists in the same space as the audio signal reproduction unit 1 2, a sound pickup signal input means 102 for inputting a sound pickup signal and accumulating it for a certain period of time to obtain a sound pickup signal sequence, a reproduction signal conversion means 103 for linearly converting the reproduction signal sequence to obtain a reproduction signal conversion sequence, Simulated reverberation generating means (conversion area reverberation simulating filter) for generating a simulated reverberation signal conversion sequence that simulates a linear conversion sequence of a reverberation signal that enters a conversion sequence and circulates from the sound signal reproducing means 1 to the sound signal collecting means 2. 104, simulation error output means 105 for inputting the simulated echo signal conversion sequence and the collected sound signal sequence and outputting a simulation error of the simulated echo signal conversion sequence, and inputting the reproduced signal conversion sequence and the simulation error. A conversion region echo that generates an updated sequence of the conversion region reflection simulation filter coefficients (an array corresponding to a linear (frequency) conversion of an array of the conversion region reflection simulation filter coefficients) of the simulation echo generation means (conversion region reflection simulation filter) 104. A simulation filter coefficient updating unit 106 is included.
The simulated reverberation generating means 104 includes means for rewriting the transformed area reverberation simulation filter coefficients based on the transformed area reverberation simulation filter coefficient update sequence obtained from the transformed area reverberation simulation filter coefficient updating section 106. The simulation error output from the simulation error output unit 105 to be input to the transformation region echo simulation filter coefficient updating unit 106 has a format that can be evaluated in the linear transformation domain.
[0005]
The difference between the configuration of FIG. 2 (conventional example 1) and that of FIG. 3 (conventional example 2) is that the configuration of conventional example 1 has a transform domain echo simulation filter coefficient constraining means 107.
FIG. 4 shows the details of the transform region echo simulation filter coefficient restricting means 107.
In Non-Patent Document 1, the inverse linear transform means 71 of FIG. 4 is realized by (high-speed) inverse discrete Fourier transform, and the linear transform means 73 is realized by (high-speed) discrete Fourier transform. Also, the windowing means 72 forcibly sets half of the inverse linear transformation sequence of the transformation region echo simulation filter coefficient update sequence to zero. The reason for this constraint is that the conversion region echo simulation filter coefficient to be obtained is a linear transformation of a 2L-length signal train obtained by combining a zero-signal train of the same length L with the impulse response of the true echo path length L. Is to be reflected at the time of updating. Non-Patent Document 1 points out that in the configuration of Conventional Example 2 without the transform region echo simulation filter coefficient restricting means, the convergence of the transform domain echo simulation filter coefficient to the value to be obtained becomes slow.
However, from the viewpoint of the amount of computation, a series of processes of linear inverse transformation, windowing, and linear transformation is not desirable. Therefore, an object of the present invention is to speed up the convergence to the transform region reverberation simulation filter coefficient to be obtained without using the transform region reverberation simulation filter coefficient restricting means as shown in FIG.
[0006]
[Means for Solving the Problems]
In the high-speed LMS (Least Mean Square) algorithm applied to the echo canceller shown in FIGS. 2 and 3 in Non-Patent Document 1, the reproduction error as a reference signal and the simulation error of the simulation echo with respect to the true echo are uncorrelated. The transform domain echo simulation filter coefficient is updated so that
Here, when there is no transform region echo simulation filter coefficient constraining means, there is no guarantee that half of the signal sequence of the inverse linear conversion of the transform domain echo simulation filter coefficients is zero. This non-zero effect appears in the simulation error as a folded component due to cyclic convolution, and the folded component and the reproduced signal can be generally regarded as uncorrelated. Therefore, the high-speed LMS algorithm that determines the transform-domain reverberation simulation filter coefficient by decorrelation, that is, evaluation of the second-order statistic, cannot positively remove the influence of aliasing. That is, in the echo cancellation by the unconstrained filter update, the aliasing component by the cyclic convolution process in the frequency domain transform affects the estimation of the echo path component, and causes a reduction in convergence performance.
[0007]
The conventional high-speed LMS updates the above-mentioned transform area reverberation simulation filter coefficient by using the product of the complex conjugate of the transform coefficient of the received signal and the transform coefficient of the error signal. , The received signal or the error signal is subjected to non-linear processing either before or after the linear transformation, and the above-mentioned updating product is obtained. In addition, the product of the received signal subjected to nonlinear processing on at least one side and the error signal in the conversion (frequency) domain is a higher-order correlation between the received signal and the error signal (second-order (normal correlation in a high-speed LMS)). )) Including quantity. In other words, the present invention is an algorithm that uses the product in the conversion domain of the received signal subjected to the non-linear processing and the error signal in updating the conversion domain reverberation simulation filter coefficient, and is enabled by the modification of the high-speed LMS.
[0008]
According to the present invention, the aliasing component included in the simulation error is made independent of the reference signal (reproduced signal), that is, the conversion region echo simulation filter coefficient is determined by evaluating a higher-order statistic. The convergence to the transform domain reverberation simulation filter coefficient to be obtained is expedited without using the coefficient constraint means. Specifically, as shown in FIG. 5, a reproduced signal non-linear processing means for applying a non-linear processing to the reproduced signal conversion sequence inputted to the conversion area reverberation simulating filter coefficient updating means 106 in the configuration of FIG. 3 (conventional example 2). 108 and a simulation error non-linear processing means 109 for applying a non-linear process to the simulation error input to the transform domain echo simulation filter coefficient updating means 106. For the signal sequence processed by the reproduced signal nonlinear processing means 108 or the simulated error nonlinear processing means 109, the conversion domain echo simulating filter coefficient An update column corresponding to the update of is calculated. However, the value of the step size parameter for adjusting the relative size of the update sequence described in Non-Patent Document 1 is not necessarily the same.
[0009]
According to the present invention, it is possible to positively evaluate the aliasing component included in the simulation error by using a higher-order statistic, and determine an update sequence of the transform area echo simulation filter coefficient.
[0010]
BEST MODE FOR CARRYING OUT THE INVENTION
(Example 1)
Although the basic configuration of the present invention is the configuration shown in FIG. 5, as an alternative configuration that can be easily analogized, a configuration in which the reproduced signal non-linear processing means 108 is arranged at a stage preceding the reproduced signal conversion means 103, that is, the reproduced signal sequence is linear. There is a configuration in which nonlinear processing is performed before conversion. The simulation error output unit 105 outputs a simulation error for the signal sequence before the linear conversion, a unit for linearly converting the simulation error sequence before the linear conversion, and outputs the simulation error sequence after the linear conversion. In the case where the simulation error non-linear processing means 109 is provided, the simulation error non-linear processing means 109 is arranged before the means for linearly converting the simulation error sequence before linear conversion in the simulation error output means 105, that is, for the simulation error sequence before linear conversion. Thus, a configuration for performing non-linear processing may be employed.
As the linear transformation of the reproduction signal transformation means 103, not only DFT but also DCT and Hartley transformation are used. As an example of nonlinear processing when a complex number is obtained using the frequency domain coefficient after conversion, the amplitude is normalized, and only the phase term is extracted, so that the independence is obtained at the phase level of the signal regardless of the signal size. Alternatively, the correlation can be evaluated (normalized).
[0011]
For the nonlinear processing in the reproduced signal nonlinear processing means 108 or the simulated error nonlinear processing means 109 in FIG. 5, for example, a sine function or a hyperbolic tangent function can be used.
The sine function is, for a certain input element z,
(Equation 1)
Figure 2004274412
Given by In addition, since the hyperbolic tangent function causes phase distortion when z is a complex number, sgn (z) · tanh (| z |) is combined with a sine function.
Can also be used. Also, any a n for a range of values with a z (n = 0, ···, ∞) by,
(Equation 2)
Figure 2004274412
Any function that can be expressed can also be used.
The reproduced signal non-linear processing means 108 and the simulation error non-linear processing means 109 may be processed by the same non-linear function or may be processed by different non-linear functions.
In the reproduced signal non-linear processing means 108 or the simulated error non-linear processing means 109 in FIG. 5, a function different from the above-described non-linear function is provided for each element of the input reproduced signal conversion sequence or the simulated error sequence. May be adopted to perform different nonlinear processing for each element. Further, a configuration may be adopted in which non-linear processing is not performed on a part of each element of the reproduced signal conversion sequence or the simulation error sequence. For example, when the linear transformation is realized by a (fast) discrete Fourier transform, it is conceivable to omit the non-linear processing on the elements of the linear transformation sequence corresponding to a high frequency with few voice components.
[0012]
(Example 2)
FIG. 6 shows that the reproduced signal nonlinear processing means 108 and the simulated error nonlinear processing means 109 of FIG. 5 use the same nonlinear function f (z), and f (z) is f (a) f (b) = f ( FIG. 5 is an alternative embodiment of FIG. 5 when ab) is satisfied, for example, when a sine function or the like is used. The reproduction signal non-linear processing means 108 and the simulation error non-linear processing means 109 of FIG. 5 collectively perform the non-linear processing on the updated sequence of the transform domain echo simulation filter coefficients, that is, the reception signal (reproduction signal) after the non-linear processing and the non-linear processing. When the product of the processed error signal and the product of the received speech signal and the error signal before processing is equal to the product obtained by performing non-linear processing on the product, the non-linear processing is collectively performed at the subsequent stage. , Etc., is replaced by the updated column non-linear processing means 110.
[0013]
(Example 3)
FIG. 7 shows a configuration in which an acoustic coupling amount measuring means 111 for obtaining an acoustic coupling amount as a ratio of the short-time average level of the collected signal to the short-time average level of the reproduced signal is added to the configuration of FIG. In accordance with the magnitude of the acoustic coupling amount measured by the acoustic coupling amount measuring unit 111, the transform region echo simulation filter coefficient updating unit 106 adjusts the size of the update amount of the transform domain echo simulation filter coefficient. Determine the parameters. This is because, for example, when the sine function is adopted in both the reproduced signal nonlinear processing means 108 and the simulated error nonlinear processing means 109, the size of the update sequence of the transform area reverberation simulating filter coefficient becomes larger than the reproduced signal or the simulated error. It no longer depends on the magnitude of the error.
[0014]
In a (high-speed) LMS or the like, the smaller the error is, the smaller the update amount of the adaptive filter (the size of the vector for correcting the filter) becomes, and the smaller the correction becomes, the more the convergence proceeds. On the other hand, when the sine function (output is ± 1) is applied to the phase as in the present invention, the update amount of the adaptive filter is constant regardless of the magnitude of the error, and does not converge more accurately than a certain error range. When the acoustic coupling gain is 1, if the step size is set to 0.01, the update is performed at +0.01, -0.01, and the error can be converged within a range of 1%. Conversely, if the step size is 0.01 when the gain is 10000, the convergence speed becomes very slow. In this case, the step size may be 100. Thus, the step size is adjusted according to the gain. In the above example, the acoustic coupling amount is set to 10000 × the allowable error rate 0.01 (1%) = step size 100 by a method such as multiplying the measured acoustic coupling amount by the estimated error allowable rate.
[0015]
As described above, in the configuration of FIG. 7, a relatively large step size parameter is given when the acoustic coupling amount is large, and a relatively small step size parameter is given when the acoustic coupling amount is small. In FIG. 7, the measurement of the acoustic coupling amount is performed using the reproduced signal and the collected signal before the linear conversion, but the measurement using the reproduced signal and the collected signal after the linear conversion can be easily performed. I can analogy. At this time, for example, when a linear transformation is performed by a (high-speed) discrete Fourier transform, the acoustic coupling amount is measured for each frequency component, and a step size parameter may be given to each frequency element based on each of the measured values. it can. Further, even when information on the amount of acoustic coupling cannot be obtained by measurement, a step size parameter that differs for each element of the linear transformation (for example, is set as a fixed parameter in advance), such as reducing the step size parameter of a high frequency component. However, if the acoustic coupling amount is obtained for each frequency, it is calculated each time). Here, reflecting the measured acoustic coupling amount in the step size parameter itself can be said to be an effective method for estimating the impulse response of the reverberation path in the reverberation canceller generally having the configuration of FIG.
Further, when the simulated reverberation generating means 104 has means for storing the past reproduced signal conversion sequence and a conversion area reverberation simulation filter coefficient corresponding to the stored reproduced signal conversion sequence, The conversion area reverberation simulation filter coefficient updating means 106 is configured as shown in FIG. 8 so that the conversion area reverberation simulation filter coefficient corresponding to the reproduced signal conversion sequence having an earlier storage time can be adjusted to have a smaller update amount. It is conceivable that the configuration is implemented.
[0016]
(Transform area echo simulation filter coefficient updating means (Example 1))
Although the configuration of FIG. 8 has a configuration in which the reproduced signal conversion sequence up to one step in the past can be stored, this configuration can be easily extended to the case where the reproduced signal conversion sequence in a plurality of steps in the past is stored. In FIG. 8, the reproduced signal conversion sequence at the current time (after the non-linear processing) is held in the signal sequence holding unit 61, and the simulation error sequence (after the non-linear processing) held by the signal sequence holding unit 62 and the signal The column multiplication means 63 multiplies each element of the column to obtain a transformed domain echo simulation filter coefficient update sequence. The update region reverberation simulation filter coefficient update sequence is multiplied by an appropriate step size parameter for each individual update column element by the gain adjusting unit 631, and is again converted by the update column holding unit 65 as a conversion region reverberation simulation filter coefficient update sequence. It is held and passed to the simulated reverberation generating means 104. Further, the reproduced signal conversion sequence of one step past (after the non-linear processing) is held in the signal sequence holding means 68 through the signal sequence delay device 67 and held by the signal sequence holding means 62 (after the non-linear processing). ) The simulated error sequence and the signal sequence multiplying means 63 are multiplied for each element of the sequence to obtain a conversion region reverberation simulated filter coefficient update sequence corresponding to the reproduced signal conversion sequence one step past (after the non-linear processing). The conversion region echo simulation filter coefficient update sequence is multiplied by a step size parameter different from that of the gain adjustment unit 631 for each update sequence element by the gain adjustment unit 691, and updated again as the conversion region echo simulation filter coefficient update sequence. The column holding means 65 independently holds the conversion area reverberation simulating filter coefficient corresponding to the previously obtained current signal (after linear processing) conversion signal reverberation sequence, and passes it to the simulated reverberation generating means 104. Here, the step size parameter of the gain adjusting means 691 is preferably smaller than the step size parameter of the gain adjusting means 631 in consideration of the actual reverberation path and the exponential decay of the impulse response. In addition, as shown in FIG. 7, when information from the acoustic coupling amount measuring unit 111 is obtained, the gain adjusting unit 631 and the gain adjusting unit 691 maintain the relative relationship (the magnitude relationship) of the step size parameter, The acoustic coupling amount evaluation means 66 can also scale each step size parameter. When the information from the acoustic coupling amount measuring unit 111 cannot be obtained, the acoustic coupling amount evaluating unit 66 can be omitted. Here, it can be said that the configuration of FIG. 8 is effective for speeding up or estimating the characteristic of the impulse response of the echo path even in the configurations of FIGS. 2 and 3 that do not include the non-linear processing means.
[0017]
(Transform area echo simulation filter coefficient updating means (second embodiment))
Further, in a case where the simulated reverberation generating means has a means for storing the past reproduced signal conversion sequence and a conversion area reverberation simulated filter coefficient corresponding to the stored reproduced signal conversion sequence, the storage time of the reproduced signal conversion sequence In order to realize different nonlinear processing, instead of the basic configuration of FIG. 5 having the reproduced signal nonlinear processing means 108 and the simulated error nonlinear processing means 109, in the configuration of FIG. A configuration in which the non-linear processing unit 601, the non-linear processing unit 602, the non-linear processing unit 603, and the non-linear processing unit 604 shown in FIG. The non-linear processing means 601 performs non-linear processing on the reproduced signal conversion sequence at the current time, the non-linear processing means 602 performs non-linear processing on the reproduced signal conversion sequence one step past, and the non-linear processing means 603 performs the non-linear processing on the current time. The nonlinear processing is performed on the simulation error sequence corresponding to the reproduced signal conversion sequence, and the non-linear processing unit 604 performs the nonlinear processing on the simulation error sequence corresponding to the reproduction signal conversion sequence one step past. These four non-linear processing means can be realized by different non-linear functions. Further, a configuration in which a part of the non-linear processing means is omitted may be employed. This makes it possible to individually adjust the estimation accuracy and estimation speed of the first half where energy is concentrated in the impulse response of the echo path and the second half where there is not much energy. If the acoustic coupling amount measuring means 111 is provided in the configuration of FIG. 3 similarly to FIG. 7, the step size parameter can be controlled by the acoustic coupling amount evaluating means 66 in FIG. If the amount measuring unit 111 cannot be provided, the acoustic coupling amount evaluating unit 66 can be omitted. Although the configuration of FIG. 9 is configured to be able to store the reproduced signal conversion sequence up to one step past, this configuration can be easily extended to the case where the reproduced signal conversion sequence of a plurality of steps past is stored.
[0018]
The echo canceller of the present invention can be constituted by a computer having a CPU and a memory, a user terminal, and a recording medium such as a CD-ROM, a magnetic disk device, and a semiconductor memory.
The echo elimination program recorded on the recording medium or the echo elimination program transmitted via a line is read by a computer, and realizes each of the above-described components on the computer and executes each process.
[0019]
【The invention's effect】
The reverberation canceling device according to the present invention, while omitting the constraint of updating the transform region reverberation simulating filter coefficient that increases the amount of calculation, corresponds to the transform region reverberation simulating filter coefficient corresponding to the impulse response of the true reverberation path in the linear transform region. , Can be found at high speed. In addition, the introduction of a nonlinear function evaluates higher-order statistics and updates the transform-domain echo simulation filter coefficients based on the independence of the reproduced signal and simulation error. However, when it can be regarded as independent from the reproduction signal, they are hardly affected.
[Brief description of the drawings]
FIG. 1 is a diagram showing a schematic configuration of an echo canceller.
FIG. 2 is a diagram showing a configuration of an echo canceller (conventional example 1) having a transform domain echo simulation filter coefficient constraining unit.
FIG. 3 is a diagram showing a configuration of an echo canceller (conventional example 2) having no transform domain echo simulation filter coefficient restricting means.
FIG. 4 is a diagram showing a configuration of a transform-domain reverberation simulating filter coefficient constraining means according to Conventional Example 1.
FIG. 5 is a diagram showing a configuration of an echo canceller (Example 1) of the present invention.
FIG. 6 is a diagram showing a configuration of an echo canceller (Example 2) of the present invention.
FIG. 7 is a diagram showing a configuration of an echo canceller (third embodiment) of the present invention.
FIG. 8 is a diagram showing a configuration of a transform area echo simulation filter coefficient updating unit (first embodiment).
FIG. 9 is a diagram showing a configuration of a transform area echo simulation filter coefficient updating unit (second embodiment).
[Explanation of symbols]
Reference numeral 100: echo canceller, 101: reproduction signal input means, 102: sound pickup signal input means, 103: reproduction signal conversion means, 104: simulated echo generation means, 105: simulation Error output means, 106: Transformation area echo simulation filter coefficient updating means, 108: Reproduction signal non-linear processing means, 109: Simulated error non-linear processing means, 110: Updated column non-linear processing means, 111 ...・ Acoustic coupling amount measurement means

Claims (18)

音響信号再生手段へと出力される再生信号を入力し、一定時間蓄積し再生信号列を得る再生信号入力手段と、
前記音響信号再生手段と同一空間に存在する音響信号収音手段から収音信号を入力し一定時間蓄積し収音信号列を得る収音信号入力手段と、
前記再生信号列を線形変換し再生信号変換列を得る再生信号変換手段と、
前記再生信号変換列を入力し前記音響信号再生手段から前記音響信号収音手段へと回り込む反響信号の線形変換列を模擬する模擬反響信号変換列を生成する模擬反響生成手段と、
前記模擬反響信号変換列と前記収音信号列とを入力し模擬反響信号変換列の模擬誤差を出力する模擬誤差出力手段と、
前記再生信号変換列と前記模擬誤差を入力し前記模擬反響生成手段の変換領域反響模擬フィルタ係数の更新列を生成する変換領域反響模擬フィルタ係数更新手段とを有する反響消去装置において、
前記変換領域反響模擬フィルタ係数更新手段へ入力する前記再生信号変換列に非線形処理を与える再生信号非線形処理手段と、前記変換領域反響模擬フィルタ係数更新手段へ入力する前記模擬誤差に非線形処理を与える模擬誤差非線形処理手段のうち何れか、または両方の手段を備えたことを特徴とする反響消去装置。
A reproduction signal input unit that inputs a reproduction signal output to an acoustic signal reproduction unit and accumulates the reproduction signal for a certain period to obtain a reproduction signal sequence;
Sound collection signal input means for inputting a sound collection signal from the sound signal sound collection means existing in the same space as the sound signal reproduction means and accumulating for a certain period of time to obtain a sound collection signal sequence;
Reproduction signal conversion means for linearly converting the reproduction signal sequence to obtain a reproduction signal conversion sequence,
A simulated reverberation generating means for inputting the reproduced signal conversion sequence and generating a simulated reverberation signal conversion sequence simulating a linear conversion sequence of a reverberation signal wrapping from the acoustic signal reproducing means to the acoustic signal collecting means,
Simulation error output means for inputting the simulation echo signal conversion sequence and the sound collection signal sequence and outputting a simulation error of the simulation echo signal conversion sequence,
An echo canceller having a conversion area reverberation simulation filter coefficient updating unit that receives the reproduction signal conversion sequence and the simulation error and generates an update sequence of the conversion region reverberation simulation filter coefficient of the simulation reverberation generation unit,
Reproduction signal nonlinear processing means for applying non-linear processing to the reproduced signal conversion sequence input to the conversion area reverberation simulation filter coefficient updating means, and simulation for applying non-linear processing to the simulation error input to the conversion area reverberation simulation filter coefficient updating means An echo canceller comprising one or both of error non-linear processing means.
請求項1に記載の反響消去装置において、
再生信号非線形処理手段あるいは模擬誤差非線形処理手段の少なくとも一方の非線形処理対象を非線形処理対象の大きさで除する、又は、非線形処理対象に双曲線正接関数を作用することにより非線形処理を実現することを特徴とする反響消去装置。
The echo canceller according to claim 1,
Dividing at least one of the reproduced signal nonlinear processing means and the simulated error nonlinear processing means by the size of the nonlinear processing object, or realizing the nonlinear processing by applying a hyperbolic tangent function to the nonlinear processing object. Characteristic echo canceller.
請求項1に記載の反響消去装置において、
再生信号非線形処理手段あるいは模擬誤差非線形処理手段の少なくとも一方に入力される前記再生信号変換列あるいは前記模擬誤差の列の各要素毎に異なる非線形関数により非線形処理を実現することを特徴とする反響消去装置。
The echo canceller according to claim 1,
Echo cancellation characterized by realizing non-linear processing with a different non-linear function for each element of the reproduced signal conversion sequence or the simulated error sequence input to at least one of the reproduced signal non-linear processing means and the simulated error non-linear processing means. apparatus.
請求項1に記載の反響消去装置において、
再生信号非線形処理手段あるいは模擬誤差非線形処理手段の少なくとも一方に入力される前記再生信号変換列あるいは前記模擬誤差の列の各要素毎の一部には非線形処理を施さないことを特徴とする反響消去装置。
The echo canceller according to claim 1,
A non-linear process is not performed on a part of each element of the reproduction signal conversion sequence or the simulation error sequence input to at least one of the reproduction signal nonlinear processing unit and the simulation error nonlinear processing unit. apparatus.
請求項1に記載の反響消去装置において、
上記収音信号の短時間平均レベルの上記再生信号の短時間平均レベルに対する比として音響結合量を求める音響結合量測定手段を備え、
前記変換領域反響模擬フィルタ係数更新手段は、前記音響結合量の大小に対応して前記変換領域反響模擬フィルタ係数の更新量の大小を決定する手段を有することを特徴とする反響消去装置。
The echo canceller according to claim 1,
Acoustic coupling amount measuring means for determining an acoustic coupling amount as a ratio of the short-time average level of the collected signal to the short-time average level of the reproduction signal,
The reverberation canceling apparatus according to claim 1, wherein the conversion area reverberation simulation filter coefficient updating means includes means for determining the magnitude of the update amount of the conversion domain reverberation simulation filter coefficient in accordance with the magnitude of the acoustic coupling amount.
請求項1に記載の反響消去装置において、
前記模擬反響生成手段は、過去の前記再生信号変換列を記憶する手段と前記記憶された再生信号変換列に対応した変換領域反響模擬フィルタ係数を有し、
前記変換領域反響模擬フィルタ係数更新手段は、前記再生信号変換列の記憶時期に従い記憶時期の古い前記再生信号変換列に対応する前記変換領域反響模擬フィルタ係数ほど更新量を小さくする手段を有することを特徴とする反響消去装置。
The echo canceller according to claim 1,
The simulated reverberation generating means includes means for storing the past reproduced signal conversion sequence and a conversion area reverberation simulated filter coefficient corresponding to the stored reproduced signal conversion sequence,
The conversion area reverberation simulation filter coefficient updating means includes means for reducing the update amount as the conversion area reverberation simulation filter coefficient corresponding to the reproduction signal conversion string having an earlier storage time in accordance with the storage time of the reproduction signal conversion string. Characteristic echo canceller.
再生される音響信号を入力し、一定時間蓄積し再生信号列を得る手順と、
前記音響信号が再生される空間と同一空間から収音された収音信号を入力し一定時間蓄積し収音信号列を得る手順と、
前記再生信号列を線形変換し再生信号変換列を得る手順と、
前記再生信号変換列を入力し再生された音響信号が収音信号へと回り込む反響信号の線形変換列を模擬する模擬反響信号変換列を生成する手順と、
前記模擬反響信号変換列と前記収音信号列とを入力し模擬反響信号変換列の模擬誤差を出力する手順と、
前記再生信号変換列と前記模擬誤差を入力し前記模擬反響信号変換列を生成する変換領域反響模擬フィルタ係数の更新列を生成する手順と、
を有する反響消去方法において、
前記変換領域反響模擬フィルタ係数の更新に用いる、前記再生信号変換列に非線形処理を与える再生信号非線形処理手順と前記模擬誤差に非線形処理を与える模擬誤差非線形処理手順のうち何れか、または両方の手順を備えたことを特徴とする反響消去方法。
A step of inputting an acoustic signal to be reproduced, accumulating the audio signal for a certain period of time, and obtaining a reproduced signal sequence;
A step of inputting a collected sound signal collected from the same space as the space in which the acoustic signal is reproduced and accumulating the collected sound signal for a certain period of time to obtain a collected sound signal sequence;
A step of linearly converting the reproduction signal sequence to obtain a reproduction signal conversion sequence;
A step of generating a simulated reverberation signal conversion sequence that simulates a linear conversion sequence of a reverberation signal in which the reproduced sound signal is input to the reproduction signal conversion sequence and wraps around to the collected signal,
A step of inputting the simulated reverberation signal conversion sequence and the collected sound signal sequence and outputting a simulated error of the simulated reverberation signal conversion sequence,
A step of inputting the reproduction signal conversion sequence and the simulation error and generating an update sequence of a transform domain echo simulation filter coefficient for generating the simulation echo signal conversion sequence;
The echo cancellation method having
One or both of a reproduction signal non-linear processing procedure for applying non-linear processing to the reproduction signal conversion sequence and a simulation error non-linear processing procedure for applying non-linear processing to the simulation error, which is used for updating the conversion area echo simulation filter coefficient. An echo cancellation method characterized by comprising:
請求項7に記載の反響消去方法において、
再生信号非線形処理手順あるいは模擬誤差非線形処理手順の少なくとも一方の非線形処理対象を非線形処理対象の大きさで除する、あるいは、非線形処理対象に双曲線正接関数を作用することにより非線形処理を実現することを特徴とする反響消去方法。
The echo cancellation method according to claim 7,
It is necessary to divide at least one of the reproduced signal non-linear processing procedure and the simulated error non-linear processing procedure by the size of the non-linear processing target, or to realize the non-linear processing by applying a hyperbolic tangent function to the non-linear processing target. Characteristic echo cancellation method.
請求項7に記載の反響消去方法において、
再生信号非線形処理手順あるいは模擬誤差非線形処理手順の少なくとも一方に入力される前記再生信号変換列あるいは前記模擬誤差の列の各要素毎に異なる非線形関数により非線形処理を実現することを特徴とする反響消去方法。
The echo cancellation method according to claim 7,
Echo cancellation characterized by realizing non-linear processing by a different non-linear function for each element of the reproduction signal conversion sequence or the simulation error sequence input to at least one of the reproduction signal non-linear processing procedure and the simulation error non-linear processing procedure. Method.
請求項7に記載の反響消去方法において、
再生信号非線形処理手順あるいは模擬誤差非線形処理手順の少なくとも一方に入力される前記再生信号変換列あるいは前記模擬誤差の列の各要素毎の一部には非線形処理を施さないことを特徴とする反響消去方法。
The echo cancellation method according to claim 7,
The echo canceller characterized in that non-linear processing is not performed on a part of each element of the reproduction signal conversion sequence or the simulation error sequence input to at least one of the reproduction signal nonlinear processing procedure and the simulation error nonlinear processing procedure. Method.
請求項7に記載の反響消去方法において、
上記収音信号の短時間平均レベルの上記再生信号の短時間平均レベルに対する比として音響結合量を求める手順と、
前記音響結合量の大小に対応して前記変換領域反響模擬フィルタ係数の更新量の大小を決定する前記変換領域反響模擬フィルタ係数を更新する手順と、を有することを特徴とする反響消去方法。
The echo cancellation method according to claim 7,
A procedure for determining the amount of acoustic coupling as a ratio of the short-time average level of the collected signal to the short-time average level of the reproduced signal;
Updating the conversion area echo simulation filter coefficient, which determines the magnitude of the update amount of the conversion area echo simulation filter coefficient in accordance with the magnitude of the acoustic coupling amount.
請求項7に記載の反響消去方法において、
前記模擬反響の生成は、過去の前記再生信号変換列を記憶し、前記記憶された再生信号変換列に対応した変換領域反響模擬フィルタを有し、
前記再生信号変換列の記憶時期に従い記憶時期の古い前記再生信号変換列に対応する前記変換領域反響模擬フィルタ係数ほどの更新量を小さくすることにより前記変換領域反響模擬フィルタ係数の更新を行うことを特徴とする反響消去方法。
The echo cancellation method according to claim 7,
The generation of the simulated reverberation stores the past reproduced signal conversion sequence, and includes a conversion region reverberation simulation filter corresponding to the stored reproduced signal conversion sequence,
Updating the conversion area reverberation simulation filter coefficient by reducing the update amount as much as the conversion area reverberation simulation filter coefficient corresponding to the reproduction signal conversion row having an earlier storage time according to the storage time of the reproduction signal conversion row. Characteristic echo cancellation method.
再生される音響信号を入力し、一定時間蓄積し再生信号列を得る処理と、
前記音響信号が再生される空間と同一空間から収音された収音信号を入力し一定時間蓄積し収音信号列を得る処理と、
前記再生信号列を線形変換し再生信号変換列を得る処理と、
前記再生信号変換列を入力し再生された音響信号が収音信号へと回り込む反響信号の線形変換列を模擬する模擬反響信号変換列を生成する処理と、
前記模擬反響信号変換列と前記収音信号列とを入力し模擬反響信号変換列の模擬誤差を出力する処理と、
前記再生信号変換列と前記模擬誤差を入力し前記模擬反響信号変換列を生成する変換領域反響模擬フィルタ係数の更新列を生成する処理と、をコンピュータに実行させる反響消去プログラムにおいて、
前記変換領域反響模擬フィルタ係数の更新に用いる、前記再生信号変換列に非線形処理を与える再生信号非線形処理と前記模擬誤差に非線形処理を与える模擬誤差非線形処理のうち何れか、または両方の処理を備えた反響消去プログラム。
A process of inputting an acoustic signal to be reproduced, accumulating for a certain period of time, and obtaining a reproduced signal sequence;
A process of inputting a collected sound signal collected from the same space as the space in which the sound signal is reproduced and accumulating the collected sound signal for a certain period of time to obtain a collected sound signal sequence;
A process of linearly converting the reproduction signal sequence to obtain a reproduction signal conversion sequence;
A process of generating a simulated reverberation signal conversion sequence that simulates a linear conversion sequence of a reverberation signal in which the reproduced signal is input and the reproduced acoustic signal is wrapped around to the collected signal,
A process of inputting the simulated reverberation signal conversion sequence and the collected sound signal sequence and outputting a simulated error of the simulated reverberation signal conversion sequence;
A reverberation canceling program for causing a computer to execute the process of generating an updated sequence of transform domain reverberation simulating filter coefficients by inputting the reproduction signal conversion sequence and the simulated error and generating the simulated reverberation signal conversion sequence,
Used for updating the transform region echo simulation filter coefficient, and includes one or both of a reproduction signal non-linear process for applying a non-linear process to the reproduction signal conversion sequence and a simulation error non-linear process for applying a non-linear process to the simulation error. Echo cancellation program.
請求項13に記載の反響消去プログラムにおいて、
再生信号非線形処理あるいは模擬誤差非線形処理の少なくとも一方の非線形処理対象を非線形処理対象の大きさで除する、あるいは、非線形処理対象に双曲線正接関数を作用することにより非線形処理を実現する反響消去プログラム。
The echo cancellation program according to claim 13,
A reverberation canceling program that divides at least one of a reproduced signal nonlinear process and a simulated error nonlinear process by a size of the nonlinear process target, or realizes a nonlinear process by applying a hyperbolic tangent function to the nonlinear process target.
請求項13に記載の反響消去プログラムにおいて、
再生信号非線形処理あるいは模擬誤差非線形処理の少なくとも一方に入力される前記再生信号変換列あるいは前記模擬誤差の列の各要素毎に異なる非線形関数により非線形処理を実現する反響消去プログラム。
The echo cancellation program according to claim 13,
A reverberation canceling program for realizing a nonlinear process by a different nonlinear function for each element of the reproduced signal conversion sequence or the model error sequence input to at least one of the reproduced signal nonlinear process and the simulation error nonlinear process.
請求項13に記載の反響消去プログラムにおいて、
再生信号非線形処理あるいは模擬誤差非線形処理の少なくとも一方に入力される前記再生信号変換列あるいは前記模擬誤差の列の各要素毎の一部には非線形処理を施さないことを特徴とする反響消去プログラム。
The echo cancellation program according to claim 13,
A reverberation canceling program, wherein non-linear processing is not performed on a part of each element of the reproduction signal conversion sequence or the simulation error sequence input to at least one of the reproduction signal nonlinear processing and the simulation error nonlinear processing.
請求項13に記載の反響消去プログラムにおいて、
上記収音信号の短時間平均レベルの上記再生信号の短時間平均レベルに対する比として音響結合量を求める処理と、
前記音響結合量の大小に対応して前記変換領域反響模擬フィルタ係数の更新量の大小を決定する処理と、を有する反響消去プログラム。
The echo cancellation program according to claim 13,
A process of obtaining an acoustic coupling amount as a ratio of the short-time average level of the collected signal to the short-time average level of the reproduction signal;
Determining the magnitude of the update amount of the transform area reverberation simulation filter coefficient in accordance with the magnitude of the acoustic coupling amount.
請求項13に記載の反響消去プログラムにおいて、
前記模擬反響の生成は、過去の前記再生信号変換列を記憶し、前記記憶された再生信号変換列に対応した変換領域反響模擬フィルタ係数を有し、
前記再生信号変換列の記憶時期に従い記憶時期の古い前記再生信号変換列に対応する前記変換領域反響模擬フィルタ係数ほど更新量を小さくする処理を有することを特徴とする反響消去プログラム。
The echo cancellation program according to claim 13,
The generation of the simulated reverberation stores the past reproduced signal conversion sequence, and includes a conversion region reverberation simulated filter coefficient corresponding to the stored reproduced signal conversion sequence,
A reverberation erasing program, characterized in that the reverberation erasing program has a process of reducing the update amount as the conversion area reverberation simulating filter coefficient corresponding to the reproduction signal conversion sequence having an earlier storage time in accordance with the storage time of the reproduced signal conversion sequence.
JP2003062656A 2003-03-10 2003-03-10 Echo canceling apparatus, method, and echo canceling program Expired - Fee Related JP3920795B2 (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
JP2003062656A JP3920795B2 (en) 2003-03-10 2003-03-10 Echo canceling apparatus, method, and echo canceling program

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
JP2003062656A JP3920795B2 (en) 2003-03-10 2003-03-10 Echo canceling apparatus, method, and echo canceling program

Publications (2)

Publication Number Publication Date
JP2004274412A true JP2004274412A (en) 2004-09-30
JP3920795B2 JP3920795B2 (en) 2007-05-30

Family

ID=33124462

Family Applications (1)

Application Number Title Priority Date Filing Date
JP2003062656A Expired - Fee Related JP3920795B2 (en) 2003-03-10 2003-03-10 Echo canceling apparatus, method, and echo canceling program

Country Status (1)

Country Link
JP (1) JP3920795B2 (en)

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2006121589A (en) * 2004-10-25 2006-05-11 Nippon Telegr & Teleph Corp <Ntt> Echo cancellation method, apparatus for executing same, program, and recording medium thereof
JP2007060644A (en) * 2005-07-28 2007-03-08 Toshiba Corp Signal processor
JP2008124914A (en) * 2006-11-14 2008-05-29 Nippon Telegr & Teleph Corp <Ntt> Echo cancelling apparatus, method and program, and recording medium therefor
JP2011103512A (en) * 2009-11-10 2011-05-26 Nippon Telegr & Teleph Corp <Ntt> Device, method and program for canceling acoustic echo
US8335311B2 (en) 2005-07-28 2012-12-18 Kabushiki Kaisha Toshiba Communication apparatus capable of echo cancellation
CN115118282A (en) * 2022-07-25 2022-09-27 重庆邮电大学 Assembly line ADC variable step length LMS calibration system based on hyperbolic tangent function
CN115967398A (en) * 2022-12-30 2023-04-14 成都电科星拓科技有限公司 Method, system and equipment for updating conversion coefficient of time-to-digital converter

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2006121589A (en) * 2004-10-25 2006-05-11 Nippon Telegr & Teleph Corp <Ntt> Echo cancellation method, apparatus for executing same, program, and recording medium thereof
JP4504782B2 (en) * 2004-10-25 2010-07-14 日本電信電話株式会社 Echo cancellation method, apparatus for implementing this method, program, and recording medium therefor
JP2007060644A (en) * 2005-07-28 2007-03-08 Toshiba Corp Signal processor
US8335311B2 (en) 2005-07-28 2012-12-18 Kabushiki Kaisha Toshiba Communication apparatus capable of echo cancellation
JP2008124914A (en) * 2006-11-14 2008-05-29 Nippon Telegr & Teleph Corp <Ntt> Echo cancelling apparatus, method and program, and recording medium therefor
JP4705554B2 (en) * 2006-11-14 2011-06-22 日本電信電話株式会社 Echo canceling apparatus, method thereof, program thereof, and recording medium thereof
JP2011103512A (en) * 2009-11-10 2011-05-26 Nippon Telegr & Teleph Corp <Ntt> Device, method and program for canceling acoustic echo
CN115118282A (en) * 2022-07-25 2022-09-27 重庆邮电大学 Assembly line ADC variable step length LMS calibration system based on hyperbolic tangent function
CN115118282B (en) * 2022-07-25 2024-05-24 南京模数智芯微电子科技有限公司 Pipeline ADC variable-step LMS calibration system based on hyperbolic tangent function
CN115967398A (en) * 2022-12-30 2023-04-14 成都电科星拓科技有限公司 Method, system and equipment for updating conversion coefficient of time-to-digital converter

Also Published As

Publication number Publication date
JP3920795B2 (en) 2007-05-30

Similar Documents

Publication Publication Date Title
CN109727604B (en) Frequency domain echo cancellation method for speech recognition front end and computer storage medium
JP5362894B2 (en) Neural network filtering technique to compensate for linear and nonlinear distortion of speech converters
JP4161628B2 (en) Echo suppression method and apparatus
US8078433B2 (en) Optimal estimation of transducer parameters
JP2008197284A (en) Filter coefficient calculation device, filter coefficient calculation method, control program, computer-readable recording medium, and audio signal processing apparatus
CN108010536A (en) Echo cancel method, device, system and storage medium
JP2004274412A (en) Echo canceler, canceling method, and canceling program
JP2003250193A (en) Echo elimination method, device for executing the method, program and recording medium therefor
JP5524316B2 (en) Parameter estimation apparatus, echo cancellation apparatus, parameter estimation method, and program
KR20230057287A (en) Method for generating active noise reduction filter, storage medium and earphone
Wesselink et al. Fast affine projections and the regularized modified filtered-error algorithm in multichannel active noise control
JP5438629B2 (en) Stereo echo canceling method, stereo echo canceling device, stereo echo canceling program
JP4034853B2 (en) Distortion removing device, multiprocessor and amplifier
JPWO2012157783A1 (en) Audio processing apparatus, audio processing method, and recording medium recording audio processing program
JP6304643B2 (en) Nonlinear distortion reduction apparatus, method, and program for speaker
JP2013113866A (en) Reverberation removal method, reverberation removal device and program
JP5826712B2 (en) Multi-channel echo canceling apparatus, multi-channel echo canceling method, and program
JP3616341B2 (en) Multi-channel echo cancellation method, apparatus thereof, program thereof, and recording medium
JP5325134B2 (en) Echo canceling method, echo canceling apparatus, program thereof, and recording medium
JP3917116B2 (en) Echo canceling apparatus, method, echo canceling program, and recording medium recording the program
JP2006148453A (en) Method, apparatus, and program for signal estimation, and recording medium for the program
JP6075783B2 (en) Echo canceling apparatus, echo canceling method and program
WO2023276424A1 (en) Howling suppression device, howling suppression method, and howling suppression program
JP2004320204A (en) Method and apparatus for echo cancellation and echo cancellation program
JP5086969B2 (en) Echo canceling apparatus, method thereof, program thereof, and recording medium thereof

Legal Events

Date Code Title Description
A621 Written request for application examination

Free format text: JAPANESE INTERMEDIATE CODE: A621

Effective date: 20050126

A977 Report on retrieval

Free format text: JAPANESE INTERMEDIATE CODE: A971007

Effective date: 20061011

RD03 Notification of appointment of power of attorney

Free format text: JAPANESE INTERMEDIATE CODE: A7423

Effective date: 20061018

A131 Notification of reasons for refusal

Free format text: JAPANESE INTERMEDIATE CODE: A131

Effective date: 20061107

A521 Written amendment

Free format text: JAPANESE INTERMEDIATE CODE: A523

Effective date: 20070104

TRDD Decision of grant or rejection written
A01 Written decision to grant a patent or to grant a registration (utility model)

Free format text: JAPANESE INTERMEDIATE CODE: A01

Effective date: 20070206

A61 First payment of annual fees (during grant procedure)

Free format text: JAPANESE INTERMEDIATE CODE: A61

Effective date: 20070215

R150 Certificate of patent or registration of utility model

Free format text: JAPANESE INTERMEDIATE CODE: R150

FPAY Renewal fee payment (event date is renewal date of database)

Free format text: PAYMENT UNTIL: 20110223

Year of fee payment: 4

FPAY Renewal fee payment (event date is renewal date of database)

Free format text: PAYMENT UNTIL: 20110223

Year of fee payment: 4

FPAY Renewal fee payment (event date is renewal date of database)

Free format text: PAYMENT UNTIL: 20120223

Year of fee payment: 5

FPAY Renewal fee payment (event date is renewal date of database)

Free format text: PAYMENT UNTIL: 20130223

Year of fee payment: 6

S531 Written request for registration of change of domicile

Free format text: JAPANESE INTERMEDIATE CODE: R313531

R350 Written notification of registration of transfer

Free format text: JAPANESE INTERMEDIATE CODE: R350

LAPS Cancellation because of no payment of annual fees