JP2012215715A - Noise estimation device and noise estimation program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To estimate noise with high estimation accuracy even for the noise which unexpectedly varies.SOLUTION: First noise estimation means (4) outputs a first estimated noise frequency characteristic indicating a noise frequency characteristic of noise included in input sound (6). Second noise estimation means (8) outputs a second estimated noise frequency characteristic of the noise specified by using prior information (12) on the noise relevant to the input sound as well as current information (10) to specify the noise. Adding means (14) outputs an estimated noise frequency characteristic in accordance with the first estimated noise frequency characteristic as well as the second estimated noise frequency characteristic.

Description

The present invention relates to a noise estimation device and a noise estimation program for estimating frequency characteristics of noise.

  As a technique for estimating the frequency characteristics of noise, it is known to discriminate between speech and non-speech sections, learn a noise spectrum in the non-speech section, and estimate the noise frequency characteristics.

  Regarding this frequency characteristic estimation technique, Patent Document 1 describes that in an audio period, updating of an estimated noise signal is stopped by background noise estimation, or updating is performed at a tracking speed for an input signal that is slower than the noise period. .

  Further, Patent Document 2 discloses determination of speech and non-speech by using speech feature values such as power of the speech signal, zero-crossing rate, peak frequency of power spectrum, and pitch period from the speech signal on which environmental noise is superimposed. It is described to do.

JP 09-31698 A JP 2002-258881 A

  By the way, when the frequency characteristics are estimated by capturing the steady nature of the noise, if the noise suddenly changes, the actual noise and the learned noise are different, and the error between them increases. In particular, when the noise changes in a section including speech or sudden noise, it is difficult to learn the change of the sound, and the error becomes remarkable.

Therefore, an object of the noise estimation device and the noise estimation method of the present disclosure is to perform noise estimation with high estimation accuracy even for noise that changes suddenly.

In order to achieve the above object, in the noise estimation device and the noise estimation program of the present disclosure, the first noise estimation unit outputs a first estimated noise frequency characteristic indicating the noise frequency characteristic of the noise included in the input sound. The second noise estimation means outputs a second estimated noise frequency characteristic of the noise specified using the prior information of the noise related to the input sound and the current information specifying the noise. Then, the adding means outputs an estimated noise frequency characteristic corresponding to the first estimated noise frequency characteristic and the second estimated noise frequency characteristic.

  According to the noise estimation device and the noise estimation program of the present disclosure, any of the following effects can be obtained.

  (1) Estimated noise frequency characteristic according to the first noise frequency characteristic obtained from the input sound and the second noise frequency characteristic obtained from the prior information of the noise related to the input sound and the current information specifying the noise Thus, the estimation accuracy can be improved.

(2) Since current information and prior information are reflected in the estimation of frequency characteristics, the accuracy of noise estimation is reduced even if the input sound contains speech or sudden noise changes such as in tunnels. It is possible to perform noise estimation without making it.

Other objects, features, and advantages of the present invention will become clearer with reference to the accompanying drawings and each embodiment.

It is a figure which shows an example of the noise estimation apparatus and noise estimation program which concern on 1st Embodiment. It is a figure which shows the structural example of the noise estimation apparatus which concerns on 2nd Embodiment. It is a figure which shows the structural example of a specific noise information table. It is a figure which shows an example of another specific noise information table. It is a flowchart which shows the process sequence of noise estimation. It is a figure which shows the structural example of the noise estimation apparatus which concerns on 3rd Embodiment. It is a figure which shows an example of a specific noise information table. It is a flowchart which shows the process sequence of noise estimation. It is a figure which shows the structural example of the noise estimation apparatus which concerns on 4th Embodiment. It is a flowchart which shows an example of the update process procedure of the frequency characteristic of specific sound source noise. It is a flowchart which shows the process sequence of noise estimation. It is a figure which shows the structural example of the noise estimation apparatus which concerns on 5th Embodiment. It is a flowchart which shows an example of the update process procedure of the frequency characteristic of specific sound source noise. It is a flowchart which shows the process sequence of noise estimation. It is a figure which shows the structural example of the noise estimation apparatus which concerns on 6th Embodiment. It is a flowchart which shows the process sequence of noise estimation. It is a figure which shows the structural example of a computer system.

[First Embodiment]

  FIG. 1 shows an example of a noise estimation device and a noise estimation program according to the first embodiment. The configuration shown in FIG. 1 is an example, and the present invention is not limited to such a configuration.

  The noise estimation apparatus 2 shown in FIG. 1 includes first noise estimation means 4, which outputs first estimated noise frequency characteristics indicating noise frequency characteristics of noise included in the input sound 6. . As an example, it is assumed that the input sound 6 includes a user's voice.

  The second noise estimation means 8 calculates the noise frequency characteristic corresponding to the input sound 6 by using the current information 10 and the prior information 12, and outputs the second estimated noise frequency characteristic. The current information 10 is current specific noise information such as date and time, position, and engine speed. The prior information 12 is information that can be acquired in advance such as a route of a transportation facility such as a bullet train or a time table.

  Then, the adding means 14 adds, for example, the first estimated noise frequency characteristic and the second estimated noise frequency characteristic, thereby estimating the estimated noise frequency according to the first estimated noise frequency characteristic and the second estimated noise frequency characteristic. Output characteristics. The adding means 14 may add these estimated noise frequency characteristics at the same ratio, or may add the weighted values with a predetermined contribution coefficient.

  In such a configuration, since the noise frequency characteristic corresponding to the first estimated noise frequency characteristic and the second estimated noise frequency characteristic is output and noise is estimated, high estimation accuracy can be obtained. The input sound 6 can be used to estimate the frequency characteristics of a voice signal (superimposed noise) on which ambient noise is superimposed. In addition, since the frequency characteristics of noise such as engine noise and tunnel running noise are stored in advance as an advance information in an environment where the route and operating time of trains are fixed, it is possible to improve the estimation performance of the estimated frequency characteristics of noise. it can.

[Second Embodiment]

  FIG. 2 shows an example of a noise estimation apparatus according to the second embodiment. The configuration shown in FIG. 2 is an example, and the present invention is not limited to such a configuration. 2, the same parts as those in FIG. 1 are denoted by the same reference numerals.

  The noise estimation apparatus 2-1 shown in FIG. 2 includes a first noise estimation unit 4-1, a prior information holding unit 16, an existing specific noise information calculation unit 18, a second noise estimation unit 8, and a third Noise estimation means 20 is provided.

  The noise estimation means 4-1 calculates the frequency characteristic of noise from the input sound and outputs this as the first estimated noise frequency characteristic. As an example, the input sound includes an input sound including sound.

  For the processing of the noise estimation means 4-1, a noise estimation method or a voice determination method can be used. The noise estimation method converts an input signal on the time axis into a signal on the frequency axis, and estimates background noise in the input signal while following the input signal based on the converted signal on the frequency axis. A method of forming an estimated noise signal (for example, Japanese Patent No. 0326969) may be used. That is, using the signal spectrum signal Xf, k of noise (background noise) that does not include speech and the estimated noise signal N ′ f, k−1 one analysis frame before, the estimated noise signal N ′ f, k of the current analysis frame is used. A method for calculating the value may be used. In this case, Xf, k and N′f, k−1 are multiplied by a coefficient to give a · N′f, k−1 and (1-a) · Xf, k, respectively, and a · N′f, k− 1 and (1-a) · Xf, k are added to calculate the estimated noise signal N′f, k of the current analysis frame. In addition, a speech determination method is a method for determining whether or not speech is included in each frame in which a speech signal on which environmental noise is superimposed as an input signal is given in chronological order (for example, Japanese Patent No. 03849116). ). In other words, the feature quantity that emphasizes the high frequency component of the audio signal feature quantity and the audio frequency feature quantity is calculated, and the determination is made based on the feature quantity that emphasizes the high frequency component in addition to the feature quantity of the audio signal. You can use the method of doing. Further, a method may be used in which the feature amount of the audio signal and the regularity of the resonance period of the audio are calculated and the determination is made based on the regularity of the resonance cycle of the audio in addition to the feature amount of the audio signal.

  The prior information holding means 16 acquires information about noise in advance and holds this information as the prior information 12 when a route or a time table is fixed, such as a bullet train.

  An example of prior information and its acquisition are as follows. Here, the frequency characteristic is a numerical value [dB] indicating the magnitude of the power spectrum for each frequency.

1) Recording of standard noise

In recording the reference noise, the frequency characteristic of the noise at the reference engine speed is measured. In this case, the frequency characteristics of noise under the conditions of the reference engine speed, the frequency f, and the reference engine speed are as follows.
r _ref : reference engine speed n _ref (f): frequency characteristic of noise under the condition of frequency f and reference engine speed f: frequency (Hz)

2) Calculation of relationship between engine speed and noise spectrum change

  In calculating the relationship between the engine speed and the change in the noise spectrum, the noise frequency characteristic when the engine speed is changed is measured, and the coefficient β (f) of equation (1) is calculated.

但し、ｒ：エンジン回転数、
α（ｆ，ｒ）：エンジン回転数ｒ，周波数ｆの場合の、エンジン騒音の周波数成分を算出するための係数、
β（ｆ）：周波数ｆのエンジン回転数と基準のエンジン回転数の差によって変化する周波数成分の大きさの係数。

Where r: engine speed,
α (f, r): coefficient for calculating the frequency component of engine noise in the case of engine speed r and frequency f,
β (f): A coefficient of the magnitude of the frequency component that changes depending on the difference between the engine speed at the frequency f and the reference engine speed.

The frequency characteristic n _e (f, r) of the engine sound when the engine speed is r is calculated from the equation (2).

但し、ｎ_e （ｆ，ｒ）：エンジン回転数がｒの場合のエンジンによる騒音周波数特性〔ｄＢ〕。

Where _ne (f, r): noise frequency characteristic [dB] by the engine when the engine speed is r.

In this case, β (f) is calculated in advance by measuring a deviation (r−r _ref ) from the reference of the engine speed and a deviation from the reference of the power component of the engine noise for each frequency component.

3) Noise change factors and recording of noise when changed

  In recording the noise change factors and the noises at the time of the change, noise is recorded in advance for each noise change factor such as a tunnel, a bridge, and an oncoming vehicle. Then, the amount of change in noise due to the change factor is calculated.

  The amount of noise spectrum change t (i, f) [dB] at the frequency f of the tunnel is calculated by equation (3).

但し、ｔ（ｉ，ｆ）：ｉ番目のトンネルの周波数ｆの騒音スペクトル変化量〔ｄＢ〕、
ｔ_obs （ｉ，ｆ，ｔ）：ｉ番目のトンネルの周波数ｆの時刻ｔにおける騒音スペ クトル〔ｄＢ〕、
Ｔ：スペクトルを算出した時間フレームのインデックス、
Ｎ：スペクトルを平均する時間フレームの数。

Where t (i, f): noise spectrum change amount [dB] of the frequency f of the i-th tunnel,
t _obs (i, f, t): noise spectrum [dB] at time t at frequency f of the i-th tunnel,
T: Index of the time frame in which the spectrum was calculated,
N: The number of time frames that average the spectrum.

Similarly, the bridge noise b (i, f) and the oncoming vehicle noise o (i, f) are calculated in the same manner from the equations (4) and (5). b (i, f) is a noise spectrum change amount [dB] of the frequency f of the i-th bridge. Further, o (i, f) is a noise spectrum change amount [dB] of the frequency f of the i-th oncoming vehicle. b _obs (i, f, t) is the noise spectrum [dB] at the time t at the frequency f of the i-th bridge, and o _obs (i, f, t) is the frequency f of the i-th oncoming vehicle. It is a noise spectrum [dB] at time t.

  The existing specific noise information calculation means 18 determines, for example, a specific noise list in advance when current information such as date and time, position, and engine speed is given, and determines whether or not it exists from this specific noise list. The existing specific noise information is calculated with reference.

  The existing specific noise information calculating means 18 checks the presence / absence of an oncoming vehicle or a bridge / tunnel by inputting the current information (date and time, position, engine speed) and associating with the predetermined specific noise. For example, the existing specific noise source information is generated by registering the date and position of a bridge / tunnel running and a passing vehicle, and comparing the current date and position.

  The existing specific noise source information is shown in FIGS. In the specific noise information table 21 shown in FIG. 3, “reference engine speed”, “number of tunnels”, “number of bridges”, and “number of oncoming vehicles” are set in the item 2102. It is set in advance according to operation conditions and routes. The “current engine speed” in the specific noise information table 21 is updated by inputting information from the outside by the existing specific noise information calculation means 18.

  In the specific noise information table 22 shown in FIG. 4, items 2202, 2204, and 2206 are set, “existence” is set in the item 2202, specific noise type is set in the item 2204, and an index is set in the item 2206. In the “presence / absence” of the item 2202, information is input and updated by existing specific noise source information obtained by comparing the current date / time and position. “1” of “presence / absence” indicates that the specific noise exists, and “0” indicates that the specific noise does not exist.

    For example, if you are currently driving through the 5th tunnel and are passing the oncoming vehicle at the 7th point, set the 5th index tunnel presence / absence column to "1" and the 7th index oncoming vehicle. The column of presence / absence of “1” is “1”, and all others are “0”.

  The noise estimation means 8 calculates the frequency characteristic of the noise using the prior information 12 and the existing specific noise information, and outputs it as the second estimated noise frequency characteristic. The noise estimation means 8 performs the following processing.

  1) Input existing specific noise information.

2) Obtain standard ambient noise information. That is, the noise spectrum n _ref (f) with respect to the reference engine speed is obtained from the prior information.

  3) The noise is corrected by the engine speed. That is, the engine speed (r) is acquired from the existing specific noise information, and the noise frequency characteristic is corrected by the change in the engine speed. This is calculated by the equation (2) described above.

  4) Perform noise correction based on position and time. That is, the existing noise information is obtained from the existing specific noise information. Correction of noise frequency characteristics due to noise change factors such as tunnels, bridges, and oncoming vehicles. This is calculated by equation (6).

但し、Ｎｔ：トンネルのインデックスの数、
Ｎｂ：橋のインデックスの数、
Ｎｏ：対向車両のインデックスの数、
Ｅｔ（ｉ）：現在インデックスｉのトンネルを走行中の際は“１”、その他は“０”である関数。
Ｅｂ（ｉ）：現在インデックスｉの橋を走行中の際は" １" 、その他は“０”である関数。
Ｅｏ（ｉ）：現在インデックスｉの対向車両とすれ違い中の際は“１”、その他は“０”である関数。

Where Nt: number of tunnel indexes,
Nb: number of bridge indexes,
No: Number of oncoming vehicle index,
Et (i): A function that is “1” when currently traveling through a tunnel of index i, and “0” otherwise.
Eb (i): A function that is “1” when the vehicle is currently traveling on a bridge of index i, and “0” in other cases.
Eo (i): A function that is “1” when passing the oncoming vehicle of the current index i, and “0” otherwise.

  The noise estimation means 20 is an example of the above-described addition means 14 (first embodiment). The noise estimation means 20 calculates a third estimated noise frequency characteristic by weighting and adding with a predetermined contribution coefficient using the first estimated noise frequency characteristic and the second estimated noise frequency characteristic. Is output as an estimated noise frequency characteristic. That is, an estimated noise frequency characteristic corresponding to the first estimated noise frequency characteristic and the second estimated noise frequency characteristic is output.

  The noise estimation unit 20 calculates the noise frequency characteristic calculated by the noise estimation unit 4-1 and the noise frequency characteristic calculated by the noise estimation unit 10 according to Expression (7).

但し、ｃｏｅｆｆ２：寄与係数（０〜１の数）、
ｎ１＿ｐｒｅｄ（ｆ）：第１の推定騒音周波数特性、
ｎ２＿ｐｒｅｄ（ｆ）：第２の推定騒音周波数特性、
ｎ３＿ｐｒｅｄ（ｆ）：第３の推定騒音周波数特性。
寄与係数は、予め与えた値または後述の寄与係数算出手段２２−１（図６）から得ればよい。

Where coeff2: contribution coefficient (number of 0 to 1),
n1_pred (f): first estimated noise frequency characteristic,
n2_pred (f): second estimated noise frequency characteristic,
n3_pred (f): third estimated noise frequency characteristic.
The contribution coefficient may be obtained from a value given in advance or contribution coefficient calculation means 22-1 (FIG. 6) described later.

  Next, FIG. 5 is referred to regarding the processing procedure of the noise estimation. FIG. 5 shows an example of a noise estimation processing procedure.

  The processing procedure illustrated in FIG. 5 is an example of the noise estimation program of the present disclosure. In this processing procedure, the first noise estimation means 4-1 described above calculates a first estimated noise frequency characteristic from the input sound (step S11). The existing specific noise information calculation means 18 calculates the existing specific noise information from the current information indicating the date and time, the position, and the current engine speed (step S12). The second noise estimation means 8 calculates the second estimated noise frequency characteristic using the existing specific noise information and the prior information (step S13). Then, the third noise estimation means 20 calculates the third estimated noise frequency characteristic using the first estimated noise frequency characteristic and the second estimated noise frequency characteristic described above, and outputs it as the estimated noise frequency characteristic. (Step S14).

  In the noise estimation apparatus 2-1 of this embodiment, the noise estimation means 4 calculates the noise frequency characteristic (first estimated noise frequency characteristic), and the noise estimation means 8 uses the prior information and the current information to calculate the noise level. A frequency characteristic (second estimated noise frequency characteristic) is calculated, and the noise estimation means 20 uses the first and second estimated noise frequency characteristics.

  According to this embodiment, when the user's voice is included in the input sound and the noise cannot be estimated correctly, or when the noise suddenly changes and it is difficult to follow, such as when entering or exiting the tunnel, the second Noise can be estimated by using the estimated noise frequency characteristic. Compared with the prior art, it is possible to calculate a high-performance estimated noise frequency characteristic.

  [Third Embodiment]

  FIG. 6 shows an example of a noise estimation apparatus according to the third embodiment. The configuration shown in FIG. 6 is an example, and the present invention is not limited to such a configuration. In FIG. 6, the same parts as those in FIG.

  The noise estimation apparatus 2-2 shown in FIG. 6 includes a first noise estimation unit 4-1, a prior information holding unit 16, an existing specific noise information calculation unit 18, a second noise estimation unit 8, and a third noise estimation unit 2-2. Noise estimation means 20 and contribution coefficient calculation means 22-1. The noise estimation means 4-1, the prior information holding means 16, the existing specific noise information calculation means 18 and the noise estimation means 8 are the same as those in the second embodiment.

  The contribution coefficient calculation means 22-1 uses the existing specific noise information to calculate a contribution coefficient indicating the magnitude of the contribution of the second estimated noise frequency characteristic to the first estimated noise frequency characteristic depending on the type of existing noise. To do.

  The contribution coefficient calculation means 22-1 calculates the coefficient of the existing specific noise type. In this case, the coefficient of the specific noise type is held in advance as the specific noise information table 23 as shown in FIG. The coefficient of the second noise frequency characteristic is calculated by the equation (8) using the coefficient of the specific noise type.

但し、ｃｏｅｆｆ２＿ｏｒｇ：予め設定された第２の騒音周波数特性の係数、
ｃｏｅｆｆ２：第２の騒音の周波数特性の寄与係数（０〜１の数）、
ｃｏｅｆｆ＿ｎ（ｋ）：特定音源種の係数（インデックスｋ）、
特定音源種の係数は、現存する場合は特定騒音情報テーブル２１の値、現存しない場合は“１．０" とする。
Ｎｋ：特定騒音の数

Where coeff2_org: coefficient of the second noise frequency characteristic set in advance,
coeff2: contribution coefficient (number of 0 to 1) of the frequency characteristic of the second noise,
coeff_n (k): coefficient of specific sound source type (index k),
The coefficient of the specific sound source type is set to the value of the specific noise information table 21 when existing, or “1.0” when not existing.
Nk: Number of specific noise

  Then, as in the second embodiment, the noise estimation means 20 calculates the estimated noise frequency characteristic by weighting and adding the first estimated noise frequency characteristic and the second estimated noise frequency characteristic with the contribution coefficient. That is, also in this embodiment, an estimated noise frequency characteristic corresponding to the first estimated noise frequency characteristic and the second estimated noise frequency characteristic is output.

  Next, the noise estimation processing procedure will be described with reference to FIG. FIG. 8 shows an example of a noise estimation processing procedure.

  The processing procedure shown in FIG. 8 is an example of the noise estimation program of the present disclosure. In this processing procedure, the first noise estimation unit 4-1 calculates a first estimated noise frequency characteristic from the input sound (step S21). The existing specific noise information calculation means 18 calculates existing specific noise information from the current information (step S22). The second noise estimation means 8 calculates a second estimated noise frequency characteristic from the existing specific noise information and the prior information (step S23). The contribution coefficient calculation means 22-1 inputs the existing specific noise information, calculates a contribution coefficient, and outputs it (step S24). Then, the third noise estimation means 20 calculates and outputs a third estimated noise frequency characteristic from the first estimated noise frequency characteristic, the second estimated noise frequency characteristic, and the contribution coefficient (step S25).

  According to this noise estimation apparatus 2-2, the contribution coefficient calculation means 22-1 calculates the contribution coefficient using the existing specific noise. In addition to the effects of the first embodiment, when the engine speed changes depending on the type of sound source, or when noise suddenly changes, such as when passing the oncoming vehicle, the contribution of the second estimated noise frequency characteristic Can be increased. As a result, the follow-up performance of the noise change can be accelerated, and the noise estimation performance can be further improved.

  [Fourth Embodiment]

  FIG. 9 shows an example of a noise estimation apparatus according to the fourth embodiment. The configuration shown in FIG. 9 is an example, and the present invention is not limited to such a configuration. 9, the same parts as those in FIG. 2 are denoted by the same reference numerals.

  The noise estimation apparatus 2-3 shown in FIG. 9 includes a first noise estimation unit 4-2, a prior information holding unit 16, an existing specific noise information calculation unit 18, a second noise estimation unit 8, and a third noise estimation unit. Noise estimating means 20 and prior information updating means 24 are provided. The existing specific noise information calculation means 18, the noise estimation means 8, and the third noise estimation means 20 are the same as those in the second embodiment.

  The first noise estimation means 4-2 calculates the frequency characteristic of the noise (first estimated noise frequency characteristic), the current frequency characteristic, and the voice determination result. The first estimated noise frequency characteristic can be obtained by calculating the frequency characteristic of noise from the input sound. The input sound includes an input sound including sound. The calculation of the sound determination result is a determination of whether or not sound is included in the input sound. For the processing of the noise estimation unit 4-2, a noise estimation method or a voice determination method can be used. The noise estimation method converts the input signal on the time axis described above into a signal on the frequency axis, and follows the background noise in the input signal based on the converted signal on the frequency axis while following the input signal. A method of estimating and forming an estimated noise signal (for example, Japanese Patent No. 3269969) may be used. That is, using the signal spectrum signal Xf, k of noise (background noise) that does not include speech and the estimated noise signal N ′ f, k−1 one analysis frame before, the estimated noise signal N ′ f, k of the current analysis frame is used. A method for calculating the value may be used. In this case, Xf, k and N′f, k−1 are multiplied by a coefficient to give a · N′f, k−1 and (1-a) · Xf, k, respectively, and a · N′f, k− 1 and (1-a) · Xf, k are added to calculate the estimated noise signal N′f, k of the current analysis frame. The voice determination method is a method for determining whether or not a voice is included in the frame for each frame in which a voice signal on which environmental noise is superimposed as an input signal described above is given in chronological order (for example, Japanese Patent No. 3849116). No.) can be used. In other words, the feature quantity that emphasizes the high frequency component of the audio signal feature quantity and the audio frequency feature quantity is calculated, and the determination is made based on the feature quantity that emphasizes the high frequency component in addition to the feature quantity of the audio signal. You can use the method of doing. Further, a method may be used in which the feature amount of the audio signal and the regularity of the resonance period of the audio are calculated and the determination is made based on the regularity of the resonance cycle of the audio in addition to the feature amount of the audio signal.

  The prior information updating unit 24 updates the prior information 12 of the prior information holding unit 16 using the existing specific noise information, the voice determination result, and the current frequency characteristic. The second noise estimation means 8 calculates noise frequency characteristics (second estimated noise frequency characteristics) using the prior information 12 updated by the prior information update means 24 and the existing specific noise information. Then, the third noise estimation unit 20 calculates the estimated noise frequency characteristic by weighting and adding the first estimated noise frequency characteristic and the second estimated noise frequency characteristic with a predetermined contribution coefficient. That is, the third noise estimation means 20 can obtain an estimated noise frequency characteristic output corresponding to the first estimated noise frequency characteristic and the second estimated noise frequency characteristic.

  Next, FIG. 10 will be referred to regarding the advance information update process. FIG. 10 shows a processing procedure for updating prior information.

  This processing procedure is executed by the prior information update unit 24. In the processing procedure shown in FIG. 10, it is determined whether it is a non-voice section (step S31). If it is a non-voice section (YES in step S31), the number of specific sound source noise types is calculated (step S32).

  For the calculated number of specific sound source noise types, it is determined whether the number of specific sound source noise types = 0 (step S33). When the calculated number of specific noise types = 0, it indicates that there are no specific noise types other than noise caused by engine rotation. If the number of specific sound source noise types = 0 (YES in step S33), whether the absolute value of the value obtained by subtracting the reference engine speed from the engine speed is smaller than the threshold (TH), that is, | engine speed-reference It is determined whether or not the engine rotational speed | <TH (step S34).

  If | engine speed−reference engine speed | <TH (YES in step S34), the frequency characteristic of the reference engine speed is updated (step S35), and this process ends. In step S34, if | engine speed−reference engine speed | <TH is not satisfied (NO in step S34), the process is terminated.

  Update the frequency characteristics of the standard engine speed using Equation (9).

但し、ｃｏｅｆｆ：０〜１の間の数（たとえば、０．９）、
ｎｒｅｆ（ｆ）：更新後の基準のエンジン回転数の周波数特性〔ｄＢ〕、
ｎｒｅｆ＿ｐｒｅｖ（ｆ）：更新前の基準のエンジン回転数の周波数特性〔ｄＢ〕、
ｎ＿ｃｕｒ（ｆ）：現在の周波数特性〔ｄＢ〕。

However, coeff: a number between 0 and 1 (for example, 0.9),
nref (f): frequency characteristics [dB] of the reference engine speed after the update,
nref_prev (f): frequency characteristic [dB] of the reference engine speed before update,
n_cur (f): Current frequency characteristic [dB].

  In step S33, if the number of specific sound source noise types is not 0 (NO in step S33), it is determined whether the number of specific sound source noise types is 1 (step S36). If the number of specific sound source noise types = 1 (YES in step S36), the frequency characteristics of the specific sound source noise are updated (step S37), and this process ends. If the number of specific sound source noise types is not 1 (NO in step S36), the processing is terminated without updating the frequency characteristics of the specific sound source noise.

  For the frequency characteristics of the existing specific noise type, the existing specific noise type (one) is updated using the following formula (10), (11) or (12).

但し、ｃｏｅｆｆ：０〜１の間の数（たとえば、０．９）
ｔ（ｉ，ｆ）：ｉ番目のトンネルの周波数ｆの騒音周波数特性〔ｄＢ〕
ｔ＿ｐｒｅｖ（ｉ，ｆ）：更新前のｉ番目のトンネルの周波数ｆの騒音周波数特性〔ｄＢ〕
ｂ（ｉ，ｆ）：ｉ番目の橋の周波数ｆの騒音周波数特性〔ｄＢ〕
ｂ＿ｐｒｅｖ（ｉ，ｆ）：更新前のｉ番目の橋の周波数ｆの騒音周波数特性〔ｄＢ〕
ｏ（ｉ，ｆ）：ｉ番目の対向車両の周波数ｆの騒音周波数特性〔ｄＢ〕
ｏ＿ｐｒｅｖ（ｉ，ｆ）：更新前のｉ番目の対向車両の周波数ｆの騒音周波数特性〔ｄＢ〕
ｎ＿ｃｕｒ２（ｆ）：現在の周波数特性〔ｄＢ〕から、エンジン音の成分を除去した周波数ｆの騒音周波数特性〔ｄＢ〕
ｎ＿ｃｕｒ２（ｆ）は、式(13)に示す通りである。

However, coeff: a number between 0 and 1 (for example, 0.9)
t (i, f): noise frequency characteristic of the frequency f of the i-th tunnel [dB]
t_prev (i, f): noise frequency characteristic [dB] of the frequency f of the i-th tunnel before update
b (i, f): noise frequency characteristic of the frequency f of the i-th bridge [dB]
b_prev (i, f): noise frequency characteristic [dB] of the frequency f of the i-th bridge before update
o (i, f): noise frequency characteristic [dB] of the frequency f of the i-th oncoming vehicle
o_prev (i, f): noise frequency characteristic [dB] of the frequency f of the i-th oncoming vehicle before update
n_cur2 (f): Noise frequency characteristic [dB] of the frequency f obtained by removing the engine sound component from the current frequency characteristic [dB].
n_cur2 (f) is as shown in Expression (13).

  Next, FIG. 11 is referred to regarding the processing procedure of the noise estimation. FIG. 11 shows an example of a processing procedure for noise estimation.

  The processing procedure shown in FIG. 11 is an example of the noise estimation program of the present disclosure. In this processing procedure, the first noise estimation unit 4-2 calculates the current frequency characteristic from the input sound (step S41). The first noise estimation means 4-2 calculates a first estimated noise frequency characteristic from the input sound (step S42). The first noise estimation unit 4-2 performs voice determination of the input sound (step S43).

  The existing specific noise information calculation means 18 calculates existing specific noise information from the current information (step S44). The prior information update unit 24 inputs the current frequency characteristics, the voice determination result of the input sound, and the existing specific noise information, and updates the prior information 12 in the prior information holding unit 16 (step S45).

  The second noise estimation means 8 calculates a second estimated noise frequency characteristic from the existing specific noise information and the prior information 12 (step S46).

  Similarly to the second embodiment, the third noise estimation means 20 inputs the first estimated noise frequency characteristic and the second estimated noise frequency characteristic, calculates the third estimated noise frequency characteristic, and outputs it. (Step S47).

  According to the noise estimation apparatus 2-3, the prior information update unit 24 updates the noise information held in the prior information 12 so as to approach the actual noise when the previously held noise differs from the actual noise. Therefore, the noise estimation performance can be further improved.

  [Fifth Embodiment]

  FIG. 12 shows an example of a noise estimation apparatus according to the fifth embodiment. The configuration shown in FIG. 12 is an example, and the present invention is not limited to such a configuration.

  The noise estimation device 2-4 shown in FIG. 12 includes a first noise estimation unit 4-2, a prior information holding unit 16, an existing specific noise information calculation unit 18, a second noise estimation unit 8, and a third noise estimation unit 4-2. Noise estimation means 20, contribution coefficient calculation means 22-2, and reliability calculation means 26.

  The first noise estimation means 4-2 calculates the frequency characteristics (first estimated noise frequency characteristics) of the noise, the voice determination, and the current frequency characteristics from the input sound by the method described above. The sound determination is a determination as to whether or not sound is included in the input sound. Since this first noise estimation means 4-2 is the same as that of the fourth embodiment, the description of the specific processing is omitted.

  The prior information holding means 16 holds the prior information 12 related to noise acquired in advance when a route or a time table is fixed, such as a bullet train.

  The existing specific noise information calculating means 18 calculates the existing specific noise information when the date and time, the position, and the engine speed are given as the current information. The existing specific noise information is determination information as to whether or not a predetermined specific noise list exists.

  The reliability calculation means 26 calculates the reliability information of the prior information using the existing specific noise information, the voice determination result, and the current frequency characteristic. The reliability information is information in which the reliability of each specific noise is expressed by a numerical value for a predetermined specific noise list. The numerical value representing the reliability is, for example, a value from 0: minimum reliability to 1: maximum reliability.

  The second noise estimation means 8 calculates noise frequency characteristics (second estimated noise frequency characteristics) using the prior information and the existing specific noise information.

  The contribution coefficient calculation means 22-2 calculates a contribution coefficient using the existing specific noise information and the reliability information. This contribution coefficient is information representing the magnitude of the contribution of the second estimated noise frequency characteristic to the first estimated noise frequency characteristic.

  Similarly to the third embodiment, the third noise estimation means 20 calculates the estimated noise frequency characteristic by weighting and adding the first estimated noise frequency characteristic and the second estimated noise frequency characteristic with the contribution coefficient. . That is, the noise estimation means 20 can obtain an estimated noise frequency characteristic output according to the first estimated noise frequency characteristic and the second estimated noise frequency characteristic.

  Next, FIG. 13 is referred to for the reliability calculation means 26. FIG. 13 shows an example of a processing procedure of reliability calculation by the reliability calculation means 26.

  For the reliability calculation, the existing specific noise information, voice determination, and current frequency characteristics are used. The determination result of the first noise estimation means 4-2 is used for the sound determination.

  In the processing procedure shown in FIG. 13, it is determined whether or not it is a non-voice section (step S51). If the voice determination result of the first noise estimation unit 4-2 is a non-voice section (YES in step S51), the number of specific noise types in the existing specific noise information is calculated (step S52). If the speech determination result is not a non-speech interval (NO in step S51), this process ends.

  It is determined whether the calculated number of specific noise types = 0 (step S53). When the calculated number of specific noise types = 0, it indicates that there are no specific noise types other than noise caused by engine rotation. If the number of specific noise types = 0 (YES in step S53), whether the absolute value of the value obtained by subtracting the reference engine speed from the engine speed is smaller than the threshold (TH), that is, | engine speed-reference It is determined whether the engine speed | <TH (step S54).

  If | engine speed−reference engine speed | <TH (YES in step S54), the reliability of the frequency characteristic of the reference engine speed is calculated (step S55), and this process ends. If | engine speed−reference engine speed | <TH (NO in step S54), the process ends without calculating the reliability of the frequency characteristics of the reference engine speed.

[Calculation of reliability of frequency characteristics of standard engine speed]

  The reliability of the frequency characteristic of the reference engine speed in step S55 is calculated from equations (14) and (15).

但し、ｎｒｅｆ＿ｒ（ｆ）：エンジン回転数が基準の場合のエンジンによる騒音周波数 特性の信頼度（０〜１）、
Δ（ｆ）：基準のエンジン回転数の騒音周波数特性の推定値と事前情 報の誤差〔ｄＢ〕、
ＴＨ３：小さな値（５〔ｄＢ〕など）。

However, nref_r (f): reliability of the noise frequency characteristic by the engine when the engine speed is a reference (0 to 1),
Δ (f): Estimated value of noise frequency characteristic of standard engine speed and error [dB] of prior information,
TH3: A small value (5 [dB] or the like).

  If the number of specific noise types calculated in step S52 is not zero (NO in step S53), it is determined whether the number of specific noise types = 1 (step S56). If the number of specific noise types = 1 (YES in step S56), the reliability of specific sound source noise is calculated (step S57), and this process is terminated. If the number of specific noise types is not 1 (NO in step S56), the process ends without calculating the reliability of the specific sound source noise.

[Calculation of reliability of existing specific noise types]

  The calculation of the reliability of the existing specific noise type in step S57 is performed using Equation (16) and Equation (17).

但し、ｎｒｅｆ＿ｒ（ｆ）：エンジン回転数が基準の場合のエンジンによる騒音周波数 特性の信頼度（０〜１）、
Δｔ（ｉ，ｆ）：インデックスｉのトンネルの騒音周波数特性の推定値と事 前情報の誤差〔ｄＢ〕。

However, nref_r (f): reliability of the noise frequency characteristic by the engine when the engine speed is a reference (0 to 1),
Δt (i, f): an error [dB] between the estimated value of the noise frequency characteristic of the tunnel of index i and the prior information.

    If the existing specific noise type is a bridge or an oncoming vehicle, the reliability may be calculated using Equations (16) and (17) as in the case of the tunnel described above.

  Thus, the number of existing specific noise types held in the prior information and the reliability of the existing specific noise types are obtained.

  In the contribution coefficient calculation means 22-2, the coefficient of the second noise frequency characteristic is calculated by the equation (18) using the reliability of the specific noise type.

但し、ｃｏｅｆｆ２＝ｃｏｅｆｆ２＿ｏｒｇ ＊ ｒｅｌ ＝１．０
ｃｏｅｆｆ２＿ｏｒｇ ＊ｒｅｌ ≦１
ｃｏｅｆｆ２＿ｏｒｇ ＊ｒｅｌ ＞１
ｃｏｅｆｆ２＿ｏｒｇ ：予め設定された寄与係数
ｒｅｌ：現在の特定騒音種の信頼度
ｃｏｅｆｆ２：第２の騒音の周波数特性の寄与係数（０〜１の数）
Ｎｔ：トンネルのインデックスの数
Ｎｂ：橋のインデックスの数
Ｎｏ：対向車両のインデックスの数
Ｅｔ（ｉ）：現在インデックスｉのトンネルを走行中の際は“１”、その他は“０”である関数。
Ｅｂ（ｉ）：現在インデックスｉの橋を走行中の際は“１”、その他は“０”である関数。
Ｅｏ（ｉ）：現在インデックスｉの対向車両とすれ違い中の際は“１”、その他は“０”である関数。
Ｎｅ：現存する特定騒音種の数
Ｆ：周波数の帯域数
ｎｒｅｆ＿ｒ（ｆ）：エンジン回転数が基準の場合のエンジンによる騒音スペクトルの信頼度（０〜１）
ｔ＿ｒ（ｆ）：インデックスｉのトンネル走行の場合の騒音スペクトルの信頼度（０〜１）
ｂ＿ｒ（ｆ）：インデックスｉの橋走行の場合の騒音スペクトルの信頼度（０〜１）
ｏ＿ｒ（ｆ）：インデックスｉの対向車両の場合の騒音スペクトルの信頼度（０〜１）

However, coeff2 = coeff2_org * rel = 1.0
coeff2_org * rel ≦ 1
coeff2_org * rel> 1
coeff2_org: contribution coefficient set in advance
rel: The reliability of the current specific noise type
coeff2: contribution factor (number of 0 to 1) of the frequency characteristic of the second noise
Nt: Number of tunnel indexes
Nb: Number of bridge indexes
No: Number of oncoming vehicle indexes
Et (i): A function that is “1” when currently traveling through a tunnel of index i, and “0” otherwise.
Eb (i): A function that is “1” when the vehicle is currently running on the bridge of index i, and “0” in other cases.
Eo (i): A function that is “1” when passing the oncoming vehicle of the current index i, and “0” otherwise.
Ne: Number of existing specific noise types
F: Number of frequency bands
nref_r (f): reliability of the noise spectrum by the engine when the engine speed is a reference (0 to 1)
t_r (f): reliability of noise spectrum in the case of tunnel traveling at index i (0 to 1)
b_r (f): reliability of noise spectrum in the case of traveling on bridge with index i (0 to 1)
o_r (f): reliability of noise spectrum in the case of oncoming vehicle with index i (0 to 1)

  Next, FIG. 14 is referred to regarding the processing procedure of this noise estimation. FIG. 14 shows an example of a processing procedure for noise estimation.

  The processing procedure illustrated in FIG. 14 is an example of the noise estimation program of the present disclosure. In this processing procedure, the first noise estimation means 4-2 calculates the current frequency characteristic from the input sound (step S61). The first noise estimation means 4-2 calculates a first estimated noise frequency characteristic from the input sound (step S62). The first noise estimation means 4-2 performs voice determination of the input sound (step S63).

  The existing specific noise information calculation means 18 calculates existing specific noise information from the current information (step S64). The reliability calculation means 26 calculates the reliability (reliability information) of the prior information from the first estimated noise frequency characteristic, the voice determination result of the input sound, and the existing specific noise information (step S65).

  The contribution coefficient calculation means 22-2 calculates a contribution coefficient from the existing specific noise information and the reliability information (step S66). The second noise estimation means 8 calculates a second estimated noise frequency characteristic from the existing specific noise information and the prior information 12 (step S67). The third noise estimation means 20 receives the first estimated noise frequency characteristic, the second estimated noise frequency characteristic, and the contribution coefficient, calculates the third estimated noise frequency characteristic, and outputs it (step S68).

  According to this noise estimation apparatus 2-4, the reliability of the prior information 12 is calculated based on the difference between the noise held in advance by the reliability calculation means 26 and the actual noise, and the contribution coefficient calculation means 22-2 is trusted by the sound source. Since the contribution coefficient of the second estimated noise frequency characteristic is calculated according to the degree, the noise estimation performance can be further improved.

  [Sixth Embodiment]

  FIG. 15 shows an example of a noise estimation apparatus according to the sixth embodiment. The configuration shown in FIG. 15 is an example, and the present invention is not limited to such a configuration.

  A noise estimation device 2-5 shown in FIG. 15 includes a first noise estimation unit 4-3, a prior information holding unit 16, an existing specific noise information calculation unit 18, a second noise estimation unit 8, and a third noise estimation unit 8. Noise estimation means 20 and contribution coefficient calculation means 22-3.

  The first noise estimation means 4-3 calculates noise frequency characteristics (first estimated noise frequency characteristics) from the input sound according to the method described above, and performs voice determination.

  For the processing of the noise estimation unit 4-3, a noise estimation method or a voice determination method can be used. The noise estimation method converts the input signal on the time axis described above into a signal on the frequency axis, and follows the background noise in the input signal based on the converted signal on the frequency axis while following the input signal. A method of estimating and forming an estimated noise signal (for example, Japanese Patent No. 3269969) may be used. That is, using the signal spectrum signal Xf, k of noise (background noise) that does not include speech and the estimated noise signal N ′ f, k−1 one analysis frame before, the estimated noise signal N ′ f, k of the current analysis frame is used. A method for calculating the value may be used. In this case, Xf, k and N′f, k−1 are multiplied by a coefficient to give a · N′f, k−1 and (1-a) · Xf, k, respectively, and a · N′f, k− 1 and (1-a) · Xf, k are added to calculate the estimated noise signal N′f, k of the current analysis frame. The voice determination method is a method for determining whether or not a voice is included in the frame for each frame in which a voice signal on which environmental noise is superimposed as an input signal described above is given in chronological order (for example, Japanese Patent No. 3849116). No.) can be used. In other words, the feature quantity that emphasizes the high frequency component of the audio signal feature quantity and the audio frequency feature quantity is calculated, and the determination is made based on the feature quantity that emphasizes the high frequency component in addition to the feature quantity of the audio signal. You can use the method of doing. Further, a method may be used in which the feature amount of the audio signal and the regularity of the resonance period of the audio are calculated and the determination is made based on the regularity of the resonance cycle of the audio in addition to the feature amount of the audio signal.

  The prior information holding means 16 acquires and holds prior information related to noise in advance when a route or a time table is determined for a bullet train or the like.

  The existing specific noise information calculation means 18 calculates the existing specific noise information which is determination information as to whether or not the predetermined specific noise list exists when the current information of the date, position, and engine speed is given. .

  The contribution coefficient calculation means 22-3 calculates a contribution coefficient representing the magnitude of the contribution of the second estimated noise frequency characteristic to the first estimated noise frequency characteristic using the existing specific noise information and the voice determination result.

  The second noise estimation means 8 calculates noise frequency characteristics (second estimated noise frequency characteristics) using the prior information 12 and the existing specific noise information.

  The third noise estimation means 20 calculates the estimated noise frequency characteristic by weighting and adding the first estimated noise frequency characteristic and the second estimated noise frequency characteristic with the contribution coefficient, as in the third embodiment. To do. That is, the noise estimation means 20 can obtain an estimated noise frequency characteristic output corresponding to the first estimated noise frequency characteristic and the second estimated noise frequency characteristic.

In the contribution coefficient of the contribution coefficient calculation unit 22-3, the contribution coefficient is calculated using the speech determination result calculated by the first noise estimation unit 4-3. In this case, the result of the voice determination is as follows.
coeff2 = 0: audio presence determination
coeff2 = 1: Determination of no voice According to the result of the voice determination, the contribution coefficient is as follows.
coeff2: contribution factor (number of 0 to 1) of the frequency characteristic of the second noise

  Next, FIG. 16 is referred to regarding the processing procedure of this noise estimation. FIG. 16 shows an example of a processing procedure for noise estimation.

  The processing procedure illustrated in FIG. 16 is an example of the noise estimation program of the present disclosure. In this processing procedure, the first noise estimation unit 4-3 calculates a first estimated noise frequency characteristic from the input sound (step S71). The first noise estimation unit 4-3 performs voice determination of the input sound (step S72).

  The existing specific noise information calculation means 18 calculates existing specific noise information from the current information (step S73). The contribution coefficient calculation unit 22-3 calculates a contribution coefficient from the existing specific noise information and the voice determination result (step S74).

  The second noise estimation means 8 calculates a second estimated noise frequency characteristic from the existing specific noise information and the prior information 12 (step S75).

  The third noise estimation means 20 calculates and outputs a third estimated noise frequency characteristic from the first estimated noise frequency characteristic, the second estimated noise frequency characteristic, and the contribution coefficient (step S76).

  According to this noise estimation apparatus 2-5, the contribution coefficient calculation means 22-3 uses the speech determination result calculated by the first noise estimation means 4-1, and calculates the contribution degree of the second estimated noise frequency characteristic. Manipulate. For this reason, when there is a high possibility that speech is included, the noise estimation performance can be further enhanced by increasing the contribution of the second estimated noise frequency characteristic.

[Hardware of the above embodiment]

  FIG. 17 is referred to for hardware used in the above embodiment. The noise estimation apparatuses 2-1, 2-2, 2-3, 2-4, and 2-5 described above can be configured by the computer system 200. FIG. 17 shows a configuration example of the computer system 200.

  A computer system 200 shown in FIG. 17 includes a CPU (Central Processing Unit) 202, a ROM (Read-Only Memory) 204, a RAM (Random-Access Memory) 206, a communication interface 208, a storage device 210, and an input / output device. 212, a reader 214, a voice input unit 215, and a bus 216.

  The ROM 204 or the storage device 210 is an example of a program storage unit. A storage device such as a hard disk or a magnetic disk can be used as the storage device 210. The storage device 210 or the ROM 204 stores the program shown in the above-described flowchart and the data of the tables 21, 22, and 23 described above. The RAM 206 constitutes a work area.

  The CPU 202 executes the above-described program, and executes the processes described in the above-described flowchart, such as noise estimation, calculation of specific noise information, and acquisition of prior information.

  The voice input unit 215 is used to input the input sound 6, the current information 10, or the prior information 12 (FIG. 1) described above, and specifically, a microphone 217 that converts voice input or noise input into an electrical signal, or the like. It has. In addition, digitized voice may be input for voice input.

  The communication interface 208 is connected to the information providing unit 220 via the network 218, and can receive various information described above. As this information, information stored in the portable recording medium 222 may be read by the reading device 214. It is also possible to store the above-described program in the portable recording medium 222 for distribution.

  The portable storage medium 222 includes various types of storage media such as a CD-ROM, a flexible disk, an optical disk, and a magneto-optical disk. Therefore, the program stored in the storage medium 222 can be read by the reading device 214 to perform the noise estimation described above.

[Other Embodiments]

  (1) In the noise estimation apparatus 2-3 according to the fourth embodiment, the contribution coefficient calculation unit 22-1 described in the third embodiment may be applied.

  (2) In the noise estimation device 2-4 of the fifth embodiment or the noise estimation device 2-5 of the sixth embodiment, the prior information updating means 24 described in the fourth embodiment is applied. Also good.

  (3) In the addition process of the first or second frequency characteristic of the third noise estimation unit 20 of the above embodiment, the weighting may be a process given to either one or both.

As described above, the most preferred embodiment of the noise estimation device and the noise estimation program has been described. However, the present invention is not limited to the above description, and is described in the scope of claims. It goes without saying that various modifications and changes can be made by those skilled in the art based on the gist of the invention disclosed in the embodiment, and such modifications and changes are included in the scope of the present invention. Yes.

DESCRIPTION OF SYMBOLS 2 Noise estimation apparatus 2-1 Noise estimation apparatus 2-2 Noise estimation apparatus 2-3 Noise estimation apparatus 2-4 Noise estimation apparatus 2-5 Noise estimation apparatus 4 1st noise estimation means 4-1 1st noise estimation means 4-2 First noise estimation means 6 Input sound 8 Second noise estimation means 10 Current information 12 Prior information 14 Addition means 16 Prior information holding means 18 Existing specific noise information calculation means 20 Third noise estimation means 21 Specific noise Information table 22 Specific noise information table 22-1 Contribution coefficient calculation means 22-2 Contribution coefficient calculation means 22-3 Contribution coefficient calculation means 23 Specific noise type table 24 Prior information update means 26 Reliability calculation means

Claims (7)

First noise estimation means for outputting a first estimated noise frequency characteristic indicating a noise frequency characteristic of noise included in the input sound;
Second noise estimation means for outputting a second estimated noise frequency characteristic of the noise identified using the prior information of the noise related to the input sound and the current information identifying the noise;
Adding means for outputting an estimated noise frequency characteristic according to the first estimated noise frequency characteristic and the second estimated noise frequency characteristic;
A noise estimation device comprising: Further, noise information is calculated from the current information, and the existing specific noise information calculating means for outputting the noise information as existing specific noise information,
Prior information holding means for holding frequency characteristics of a specific noise type as the prior information;
The noise estimation apparatus according to claim 1, further comprising: Furthermore, contribution coefficient calculation means for calculating a contribution coefficient representing the magnitude of the contribution of the second estimated noise frequency characteristic to the first estimated noise frequency characteristic using the existing specific noise information;
The noise estimation unit according to claim 2, wherein the adding means adds the second estimated noise frequency characteristic reflecting the contribution coefficient and the first estimated noise frequency characteristic. apparatus. Further, prior information update means for updating the prior information with update information calculated using the existing specific noise information, the current frequency characteristics, and the voice determination result;
The first noise estimation unit calculates the voice determination result from the input sound based on the first estimated noise frequency characteristic, the current frequency characteristic, and the voice determination. The noise estimation apparatus according to 2 or 3. Further, reliability calculation means for calculating reliability information of the prior information from the existing specific noise information, current frequency characteristics, and voice determination results;
Contribution coefficient calculation means for calculating a contribution coefficient representing the magnitude of the contribution of the second estimated noise frequency characteristic to the first estimated noise frequency characteristic using the existing specific noise information and the reliability information;
The first noise estimation means calculates the first estimated noise frequency characteristic, the current frequency characteristic, and the voice determination result from the input sound,
The adding means adds to the second estimated noise frequency characteristic reflecting the contribution coefficient and the first estimated noise frequency characteristic.
The noise estimation apparatus according to any one of claims 2 to 4, wherein And a contribution coefficient calculation means for outputting a contribution coefficient representing the magnitude of the contribution of the second estimated noise frequency characteristic to the first estimated noise frequency characteristic from the existing specific noise information and the voice determination result;
The first noise estimation means outputs the first estimated noise frequency characteristic and the voice determination result from the input sound,
The adding means adds the second estimated noise frequency characteristic reflecting the contribution coefficient calculated by the contribution coefficient calculating means and the first estimated noise frequency characteristic;
The noise estimation apparatus according to any one of claims 2 to 5, wherein A noise estimation program to be executed by a computer,
Outputting a first estimated noise frequency characteristic indicating a noise frequency characteristic of the noise included in the input sound of the first noise estimating means;
A second noise estimation means for outputting a second estimated noise frequency characteristic of the noise identified using the prior information of the noise related to the input sound and the current information identifying the noise;
A noise estimation program executed by the computer, wherein the adding means outputs an estimated noise frequency characteristic corresponding to the first estimated noise frequency characteristic and the second estimated noise frequency characteristic.

Images