JP3579640B2 - Acoustic characteristic control device - Google Patents

Description

この伝達関数Ｈ（ｚ）をコンボルバ１に設定する事により、マイクロホン３の出力信号ｙ（ω）は次の様に制御される。
【０００４】

ただし、ｆ［・］：離散フーリエ変換即ち出力信号ｙ（ω）は、入力信号ｓ（ω）を、参照信号生成フィルタ４に与えた所望の振幅、及び位相・周波数特性へ補正した信号になる。
【０００５】
【発明が解決しようとする課題】
説明を単純・明瞭にするため、周波数ω_０に対応する入力信号ｓ（ω_０）のみを考え、この入力信号は、時刻ｋとともに次の様に変化するものと仮定する。
ｋ≦ｋ_１ ： ｓ（ω_０）≠０
ｋ_１＜ｋ≦ｋ_２ ： ｓ（ω_０）＝０
ｋ_２＜ｋ ： ｓ（ω_０）≠０
（ａ）ｋ≦ｋ_１の時
前述の通り、周波数ω_０に対応する適応フィルタ６の伝達関数Ｈ（ω_０）（＝Ｃ（ω_０），Ｃ（ω_０）：周波数ω_０に対応するコンボルバ１の伝達関数）は（１）式に示される最適値へ収束する。この値をＨ_ＯＰＴ≠０とする。
（ｂ）ｋ_１＜ｋ≦ｋ_２の時
参照信号生成フィルタ４の出力信号ｄ（ω_０）、マイクロホン３の出力信号ｙ（ω_０）、及び第２遅延器１２の出力信号ｒ（ω_０）は全て零となる。｜ｅ（ω_０）｜^２＝｜ｄ（ω_０）−Ｃ（ω_０）｜^２つまり｜ｅ（ω_０）｜^２＝｜０−（０＋ｕ（ω_０）Ｈ（ω_０））｜^２を最小化することになり雑音ｕ（ω_０）は非零であるから、適応フィルタ６の伝達関数Ｈ（ω_０）は零へ収束する。
（ｃ）ｋ_２＜ｋの時
適応フィルタ６の伝達関数Ｈ（ω_０）は、（ａ）の場合と同様に再び最適値Ｈ_ＯＰＴへ収束する。
【０００６】
即ち、従来技術では、入力信号ｓ（ω_０）が変動し、繰返し零になると、この変動に応じて入力信号ｓ（ω_０）を所望特性へ補正する適応フィルタ６の伝達関数Ｈ（ω_０）（＝Ｃ（ω_０））は最適値Ｈ_ＯＰＴと零との間を揺れ動き、ｓ（ω_０）を所望特性に補正し続ける事は出来ない。
この発明の目的は、上述の様なｓ（ω_０）の変動に関わり無く、ｓ（ω_０）を所望の振幅、及び位相・周波数特性へ補正し続ける事の出来る音響特性補正装置を提供する事にある。
【０００７】
【課題を解決するための手段】
上記目的を達成するため、請求項１記載のこの発明では、従来の第２遅延器１２を取り除き、コンボルバ１と同じ伝達関数を有する第２のコンボルバを設け、出力側のみを適応フィルタ６と並列に接続し、且つこの第２のコンボルバに雑音発生部７で発生される疑似雑音の逆相信号を入力した。また、雑音発生部７では参照信号ｄ（ω）を分析し、この信号が零となる周波数ω_０（ω_０は一つとは限らない）を同定し、この周波数の疑似雑音ｕ（ω_０）を発生する。この変更により、入力信号が零となり、従って参照信号も零となる周波数ω_０については、「誤差パワー｜ｅ（ω_０）｜^２＝｜Ｈ（ω_０）ｕ（ω_０）−Ｃ（ω_０）ｕ（ω_０）｜^２の最小化」を指導原理とするアルゴリズムにより、適応フィルタ６の伝達関数Ｈ（ω_０）は第２のコンボルバに既に設定されていた最適値Ｃ（ω_０）に拘束される。また、従来の第１遅延器１１は不要となる。
【０００８】
【発明の実施の形態】
図１にこの発明の請求項１で提案する音響特性補正装置の実施例を示す。図２と対応する部分には同一の符号を付している。入力信号ｓ（ω）はコンボルバ１のみを通じてスピーカ２と、参照信号生成フィルタ４とへ供給される。また、コンボルバ１と同一伝達関数を有するコンボルバ５が設けられ、コンボルバ５の出力は適応フィルタ６の出力と加算されてレプリカｃ（ω）とされる。また雑音発生部７からの疑似雑音ｕ（ω）が位相反転器１０を通じてコンボルバ５に入力される。雑音発生部７においては疑似雑音発生器９からの雑音ｎ（ω）が、乗算器８で重みα（ω）を与えられて疑似雑音ｕ（ω）＝α（ω）・ｎ（ω）として出力される。
【０００９】
コンボルバ１及び５の伝達関数Ｃ（ｚ）の初期値はｍサンプル遅延ｚ^−ｍである。入力信号ｓ（ω）は、コンボルバ１による処理を受け、信号ｘ（ω）としてスピーカ２と参照信号生成フィルタ４へ入力される。参照信号生成フィルタ４の伝達特性としてｔ＋ｍサンプル分の遅延を受けた所望特性ｚ^{−（ｔ＋ｍ）}Ｒ（ｚ）を設定する。この遅延は適応フィルタ６の伝達関数Ｈ（ｚ）を安定に収束させるためのものである。マイクロホン３の出力信号ｙ（ω）は、雑音発生部７で発生する疑似雑音ｕ（ω）と加算されて適応フィルタ６へ入力される。疑似雑音ｕ（ω）は、位相反転器１０において逆相信号となり、第２コンボルバ５へ入力される。雑音発生部７は、参照信号生成フィルタ４の出力ｄ（ω）を周波数重み計算部１３で分析し、その結果により適当な周波数重みα（ω）を計算し、疑似雑音発生器９で発生する雑音ｎ（ω）に周波数重みα（ω）を乗算器８で掛け、次の疑似雑音ｕ（ω）＝α（ω）ｎ（ω）を得る。雑音ｎ（ω）は全周波数にわたってほぼ一定の強度をもつものとする。周波数重みα（ω）は、例えば次の様に与えられる。
【００１０】
｜ｄ（ω_０）｜^２／Σω｜ｄ（ω）｜^２≦βの時、
α（ω_０）＝１ …（３ａ）
｜ｄ（ω_０）｜^２／Σω｜ｄ（ω）｜^２＞βの時、
０＜α（ω_０）＜＜１ …（３ｂ）
ただし、｜ｎ（ω）｜^２≒｜ｙ（ω）｜^２ （ω≠ω_０）
β：正の定数（例えば、１０^−４）
適応フィルタ６の伝達関数Ｈ（ｚ）は、「誤差パワーΣω｜ｅ（ω）｜^２＝Σω｜ｄ（ω）−ｃ（ω）｜^２の最小化」を指導原理とするアルゴリズムより求められる。
このアルゴリズムが収束すると、コンボルバ１及び５の伝達関数Ｃ（ｚ）はＨ（ｚ）に置き換えられる。以下、「発明が解決しようとする課題」の項の説明と同様に入力信号を変化させ、つまり周波数ω_０に対応する入力信号ｓ（ω_０）を考え、前記従来技術の問題点が解消されていることを確認する。入力信号ｓ（ω_０）は、時刻ｋとともに次の様に変化する。
【００１１】
ｋ≦ｋ_１ ： ｓ（ω_０）≠０
ｋ_１＜ｋ≦ｋ_２ ： ｓ（ω_０）＝０
ｋ_２＜ｋ ： ｓ（ω_０）≠０
（ａ）ｋ≦ｋ_１の時
式（３ｂ）を満す状態であり、｜ｕ（ω_０）｜^２＜＜｜ｙ（ω_０）｜^２となり、周波数ω_０に対応する適応フィルタ６の伝達関数Ｈ（ω_０）（＝Ｃ（ω_０），Ｃ（ω_０）：コンボルバ１及び５の伝達関数）は最適値ｆ［Ｚ^−ｍＲ（ｚ）／Ｇ（ｚ）］（ω＝ω_０）へ収束する。この値をＨ_ＯＰＴ≠０とする。
（ｂ）ｋ_１＜ｋ≦ｋ_２の時
参照信号生成フィルタ４の出力信号ｄ（ω_０）及びマイクロホン３の出力信号ｙ（ω_０）は零となる。また式（３ａ）を満す状態であり、十分な大きさの疑似雑音ｕ（ω_０）が雑音発生部７で発生され、「課題を解決するための手段」の項で述べた通り｜ｅ（ω_０）｜^２＝｜Ｈ（ω_０）ｕ（ω_０）−Ｃ（ω_０）ｕ（ω_０）｜^２の最小化により伝達関数Ｈ（ω_０）はＣ（ω_０）＝Ｈ_ＯＰＴに拘束される。
（ｃ）ｋ_２＜ｋの時
式（３ｂ）を満す状態であり、伝達関数Ｈ（ω_０）は（ａ）ｋ≦ｋ_１の時のようにＨ_ＯＰＴ≠０に収束することになるが、Ｈ（ω_０）＝Ｃ（ω_０）は既にＨ_ＯＰＴになっているからスピーカ２とマイクロホン３との間の伝達関数ｚ^−ｔＧ（ｚ）が変化しない限り、Ｈ（ω_０）は最適値Ｈ_ＯＰＴのままである。
【００１２】
即ち、この発明では、上述の様なｓ（ω_０）の変動に関わり無く、ｓ（ω_０）を所望の振幅、及び位相・周波数特性へ補正し続ける事が出来る。
【００１３】
【発明の効果】
以上説明したように、この発明によれば「適当な伝達関数を有する第１コンボルバを用いてスピーカ入力信号を生成し、且つ第１コンボルバと同じ伝達関数を有する第２コンボルバに出力側が並列接続された適応フィルタにより、スピーカ再生される音信号特性を所望特性に補正する伝達関数を求めて、この伝達関数を前記コンボルバへ設定する」と云う一連の処理を繰り返すことにより、音声や楽音の様に、供給される周波数成分が時間変動する信号をスピーカ再生する場合においても、音場内の１受聴点で観測されるスピーカ再生音の特性（音圧、及び位相・周波数特性）を所望特性へ補正し続けることが出来る。
【図面の簡単な説明】
【図１】この発明の一実施例に関わる音響特性補正装置の構成を示すブロック図。
【図２】従来技術を説明するためのブロック図。[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention corrects the sound pressure from a speaker observed at one listening point in an arbitrary sound field, and the phase and frequency characteristics to desired characteristics, thereby achieving characteristics unique to the speaker and characteristics unique to the sound field. The present invention relates to an acoustic characteristic correction device which prevents deterioration of sound quality due to sound and enables high-quality reproduction of sound signals such as voice and musical sound at the listening point.
[0002]
[Prior art]
The prior art (Japanese Patent Application No. 10-059703) will be described with reference to FIG. The initial value of the transfer function C (z) of the convolver 1 is zero. The input signal s (ω) (ω: discrete frequency) is processed by the delay unit 11 and the convolver 1 connected in parallel to the delay unit 11, and sent to the speaker 2 and the reference signal generation filter 4 as a signal x (ω). Is entered. The output signal y (ω) of the microphone 3 is input to the second delay unit 12 having the same delay time as the first delay unit 11 and the adaptive filter 6 connected in parallel to the second delay unit 12. The noise u (ω) (| u (ω) | ² << | y (ω) | ² ) supplied from the noise generator 9 is also input to the adaptive filter 6. The adaptive filter 6 generates a replica c (ω) of the output signal d (ω) of the reference signal generation filter 4 from the output r (ω) of the second delay unit 12.
[0003]
The transfer function of the delay units 11 and 12 is z ^−m (m: discrete time), the transfer function between the speaker 2 and the microphone 3 is z ^−t G (z), and the transfer function of the reference signal generation filter 4 is z ⁻ if ^{(m + t) R (z} ), the transfer function H of the adaptive filter 6 (z) is "error power ^{Σω | e (ω) | 2} = Σω | d (ω) -c (ω) | 2 minimum It converges to the following value by the algorithm based on the guiding principle.

By setting the transfer function H (z) in the convolver 1, the output signal y (ω) of the microphone 3 is controlled as follows.
[0004]

Here, f [·]: discrete Fourier transform, that is, the output signal y (ω) is a signal obtained by correcting the input signal s (ω) to the desired amplitude and phase / frequency characteristics given to the reference signal generation filter 4. .
[0005]
[Problems to be solved by the invention]
For simplicity and clarity, consider only the input signal s (ω ₀ ) corresponding to frequency ω ₀ , and assume that this input signal changes as follows with time k.
k ≦ k ₁ : s (ω ₀ ) ≠ 0
k ₁ <k ≦ k ₂ : s (ω ₀ ) = 0
k ₂ <k: s (ω ₀ ) ≠ 0
(A) k as described above when ≦ _{k 1,} the transfer function H of the adaptive filter 6 which corresponds to the frequency _{ω 0 (ω 0) (=} C (ω 0), C (ω 0): corresponding to the frequency omega ₀ The transfer function of the convolver 1 converges to the optimum value shown in the equation (1). This value is set as H _OPT ≠ 0.
(B) When k ₁ <k ≦ k ₂ The output signal d (ω ₀ ) of the reference signal generation filter 4, the output signal y (ω ₀ ) of the microphone 3, and the output signal r (ω ₀ ) of the second delay unit 12 ) Are all zero. | E (ω ₀ ) | ² = | d (ω ₀ ) −C (ω ₀ ) | ^2, that is, | e (ω ₀ ) | ² = | 0- (0 + u (ω ₀ ) H (ω ₀ )) | ² And the noise u (ω ₀ ) is non-zero, so that the transfer function H (ω ₀ ) of the adaptive filter 6 converges to zero.
(C) When k ₂ <k The transfer function H (ω ₀ ) of the adaptive filter 6 converges again to the optimum value H _OPT as in the case of (a).
[0006]
That is, in the prior art, when the input signal s (ω ₀ ) fluctuates and repeatedly becomes zero, the transfer function H (ω ₀ ) of the adaptive filter 6 that corrects the input signal s (ω ₀ ) to desired characteristics according to this fluctuation. ) (= C (ω ₀ )) fluctuates between the optimum value H _OPT and zero, and s (ω ₀ ) cannot be continuously corrected to a desired characteristic.
An object of the present invention is to provide an acoustic characteristic correction apparatus capable of continuously correcting s (ω ₀ ) to a desired amplitude and phase / frequency characteristics irrespective of the fluctuation of s (ω ₀ ) as described above. Is in the thing.
[0007]
[Means for Solving the Problems]
In order to achieve the above object, according to the first aspect of the present invention, the conventional second delay unit 12 is removed, a second convolver having the same transfer function as the convolver 1 is provided, and only the output side is connected in parallel with the adaptive filter 6. , And a negative-phase signal of pseudo noise generated by the noise generator 7 was input to the second convolver. Further, the noise generator 7 analyzes the reference signal d (ω), identifies a frequency ω _{0 at} which the signal becomes zero (ω ₀ is not necessarily one), and generates a pseudo noise u (ω ₀ ) of this frequency. Occurs. This change, whereby the input signal is zero, therefore the frequency omega ₀ of the reference signal also becomes zero, "error power ^{_{| e (ω 0) | 2}} = | H (ω 0) u (ω 0) -C (ω ₀ ), the transfer function H (ω ₀ ) of the adaptive filter 6 is set to the optimum value C (ω ₀ ) which has already been set in the second convolver by an algorithm based on the guiding principle of “ ₀ ) u (ω ₀ ) | ² ”. Be bound by. Further, the conventional first delay unit 11 becomes unnecessary.
[0008]
BEST MODE FOR CARRYING OUT THE INVENTION
FIG. 1 shows an embodiment of an acoustic characteristic correction device proposed in claim 1 of the present invention. Parts corresponding to those in FIG. 2 are denoted by the same reference numerals. The input signal s (ω) is supplied to the speaker 2 and the reference signal generation filter 4 only through the convolver 1. Further, a convolver 5 having the same transfer function as the convolver 1 is provided, and the output of the convolver 5 is added to the output of the adaptive filter 6 to be a replica c (ω). The pseudo noise u (ω) from the noise generator 7 is input to the convolver 5 through the phase inverter 10. In the noise generator 7, the noise n (ω) from the pseudo noise generator 9 is given a weight α (ω) by the multiplier 8 and becomes pseudo noise u (ω) = α (ω) · n (ω). Is output.
[0009]
The initial value of the transfer function C (z) of the convolvers 1 and 5 is m sample delay z− ^m . The input signal s (ω) is processed by the convolver 1 and input to the speaker 2 and the reference signal generation filter 4 as a signal x (ω). The desired characteristic z− ^{(t + m)} R (z) delayed by t + m samples is set as the transfer characteristic of the reference signal generation filter 4. This delay is for stably converging the transfer function H (z) of the adaptive filter 6. The output signal y (ω) of the microphone 3 is added to the pseudo noise u (ω) generated in the noise generator 7 and input to the adaptive filter 6. The pseudo noise u (ω) becomes an inverted phase signal in the phase inverter 10 and is input to the second convolver 5. The noise generator 7 analyzes the output d (ω) of the reference signal generation filter 4 by the frequency weight calculator 13, calculates an appropriate frequency weight α (ω) based on the analysis result, and generates the pseudo weight by the pseudo noise generator 9. The noise n (ω) is multiplied by the frequency weight α (ω) by the multiplier 8 to obtain the following pseudo noise u (ω) = α (ω) n (ω). The noise n (ω) has a substantially constant intensity over all frequencies. The frequency weight α (ω) is given, for example, as follows.
[0010]
| D (ω ₀ ) | ² / Σω | d (ω) | ² ≦ β,
α (ω ₀ ) = 1 (3a)
When | d (ω ₀ ) | ² / Σω | d (ω) | ² > β,
0 <α (ω ₀ ) <<<< 1 (3b)
Where | n (ω) | ² ≒ | y (ω) | ² (ω ≠ ω ₀ )
β: positive constant (for example, 10 ⁻⁴ )
The transfer function H (z) of the adaptive filter 6 is obtained from an algorithm based on the guiding principle of “minimizing the error power Σω | e (ω) | ² = Σω | d (ω) −c (ω) | ² ”. .
When this algorithm converges, the transfer functions C (z) of the convolvers 1 and 5 are replaced by H (z). Hereinafter, the input signal is changed in the same manner as described in the section of “Problems to be Solved by the Invention”, that is, the input signal s (ω ₀ ) corresponding to the frequency ω ₀ is considered, and the above-described problem of the related art is solved. Make sure that The input signal s (ω ₀ ) changes with time k as follows.
[0011]
k ≦ k ₁ : s (ω ₀ ) ≠ 0
k ₁ <k ≦ k ₂ : s (ω ₀ ) = 0
k ₂ <k: s (ω ₀ ) ≠ 0
(A) When k ≦ k ₁ , Equation (3b) is satisfied, | u (ω ₀ ) | ² << | y (ω ₀ ) | ² , and the adaptive filter 6 corresponding to the frequency ω ₀ The transfer function H (ω ₀ ) (= C (ω ₀ ), C (ω ₀ ): the transfer function of the convolvers 1 and 5) has an optimal value f [Z− ^mR (z) / G (z)] (ω = ω ₀ ). This value is set as H _OPT ≠ 0.
(B) When k ₁ <k ≦ k ₂ The output signal d (ω ₀ ) of the reference signal generation filter 4 and the output signal y (ω ₀ ) of the microphone 3 become zero. In addition, the state satisfies Expression (3a), and a sufficiently large pseudo noise u (ω ₀ ) is generated by the noise generation unit 7, and as described in the section of “Means for Solving the Problem”, | e (Ω ₀ ) | ² = | H (ω ₀ ) u (ω ₀ ) −C (ω ₀ ) u (ω ₀ ) | ² , the transfer function H (ω ₀ ) becomes C (ω ₀ ) = H _Be bound by the _OPT .
(C) When k ₂ <k The condition (3b) is satisfied, and the transfer function H (ω ₀ ) converges to H _OPT ≠ 0 as in (a) k ≦ k _{1. There, H (ω 0) = C} (ω 0) is already long as the transfer function ^z -t G between the speaker 2 and the microphone 3 because they become _{H OPT} (z) does not change, H (ω _{0) Remains at} the optimal value H _OPT .
[0012]
That is, according to the present invention, s (ω ₀ ) can be continuously corrected to a desired amplitude and phase / frequency characteristics irrespective of the fluctuation of s (ω ₀ ) as described above.
[0013]
【The invention's effect】
As described above, according to the present invention, “a speaker input signal is generated using a first convolver having an appropriate transfer function, and the output side is connected in parallel to a second convolver having the same transfer function as the first convolver. The adaptive filter obtains a transfer function for correcting the sound signal characteristic reproduced by the speaker to a desired characteristic, and sets this transfer function to the convolver. Even when a signal whose supplied frequency component fluctuates with time is reproduced by a speaker, the characteristics (sound pressure, phase and frequency characteristics) of the speaker reproduction sound observed at one listening point in the sound field are corrected to desired characteristics. You can continue.
[Brief description of the drawings]
FIG. 1 is a block diagram showing a configuration of an acoustic characteristic correction device according to an embodiment of the present invention.
FIG. 2 is a block diagram for explaining a conventional technique.

Claims (2)

The original sound signal is supplied to the first convolver,
An output of the first convolver is supplied to a reference signal generation filter,
The signal observed at the listening point and the pseudo noise from the noise generator are input to the adaptive filter,
Pseudo noise from the noise generator is supplied to a second convolver,
A replica is generated by subtracting the output signal of the second convolver from the output signal of the adaptive filter,
The filter coefficient of the adaptive filter is calculated such that an error between the replica and the reference signal output from the reference signal generation filter is minimized,
The acoustic characteristic control device, wherein the filter coefficient is set in the first convolver and the second convolver. The frequency weight of the reference signal is less than or equal to a predetermined value for each frequency component, and a weight close to 0 is calculated by the frequency weight calculator, and the noise of substantially the same level in the required frequency band is multiplied by the frequency weight by the multiplier. 2. The acoustic characteristic control device according to claim 1, wherein the pseudo noise is used.

Images