JP2015210423A - Specific voice suppressor, specific voice suppression method and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a specific voice suppressor which suppresses voice of a specific speaker in the mixture voice signal, a specific voice suppression method and a program.SOLUTION: A specific voice suppressor comprises: a sound source separation unit which generates, when i is an element of {1,...,M}, a voice estimated value Sof a speaker i and a power parameter Pcorresponding to a voice power of the speaker i from a voice signal including voice of M speakers; a loud voice determination unit which compensates, when j is an element of {1,...,M} other than i, the voice power Pof the voice of the speaker i specified by the power parameter Pusing a distance d(U, L) between a microphone that is used when recording the voice signal and the speaker i, generates the compensated voice power P, and calculates a loud voice degree Ethat represents a loud voice degree of the speaker i to the other speakers j using the power P; and a mixture voice signal generation unit which combines the voice estimated value Sother than the estimated value Scorresponding to the loud voice degree Ethat is a threshold A or more to generate a mixture voice signal.

Description

  The present invention relates to a technique for suppressing the voice of a specific speaker from a voice signal including the voices of M speakers.

  Patent Document 1 is known as a conventional technique for detecting a voice section corresponding to an angry emotion from voice data. In Patent Literature 1, the learning data is used to learn the relationship between the voice feature quantity and the emotional expression level, and the code book in which each voice feature quantity is associated with the emotional expression probability is learned. Then, by searching the code book based on the voice feature amount extracted from the input voice data, the emotion expression probability of the extracted voice feature amount is obtained, and whether or not it is a section corresponding to the anger emotion. judge.

JP 2005-345496 A

  However, the conventional technology does not support an audio signal including the voices of a plurality of speakers. In the prior art, when the audio signal is an audio signal including a plurality of speaker's audio (hereinafter also referred to as a mixed audio signal), emotion classification is performed based on an audio feature amount corresponding to the mixed audio signal. Therefore, it is not possible to extract only the speech segment of the anger feeling of a specific person included in the mixed speech signal.

  An object of the present invention is to provide a specific speech suppression apparatus, a specific speech suppression method, and a program for suppressing a specific speaker's speech from a mixed speech signal.

In order to solve the above-described problem, according to one aspect of the present invention, the specific speech suppressing apparatus sets iε {1,..., M}, and from the speech signal including the speech of M speakers, a sound source separation unit for generating an estimated value S i of the voice of _i and a power parameter P _{1, i} corresponding to the power of the voice of the speaker i, j∈ {1,..., M} \ i, Using the distance d (U, L _{2, i} ) between the microphone and the speaker i used to pick up the sound, the power P _{2, i} of the voice of the speaker i specified by the power parameter P _{1 , i} To generate a corrected voice power P _{3, i} , and use the power P _{3, i} to calculate loudness E _i representing loudness of the speaker i with respect to another speaker j except a determining unit, the estimated value S _{i_2} corresponding to the threshold a or more loud degree E _{i_2,} mixed signal which synthesizes an estimate S _{i - 3} audio, to produce a mixed audio signal Including a generating unit, a.

In order to solve the above-described problem, according to another aspect of the present invention, the specific speech suppression apparatus sets iε {1,..., M}, and from the speech signal including speech of M speakers, A sound source separation unit that generates an estimated value S _i of the voice of the person i and a power parameter P _{1, i} corresponding to the power of the voice of the speaker i, j∈ {1,..., M} \ i, Using the distance d (U, L _{2, i} ) between the microphone and the speaker i used when collecting the signal, the power P ₂ of the voice of the speaker i specified by the power parameter P _{1, i i} is corrected, corrected power P _{3, i} is generated, and loudness E _i representing the loudness of speaker i with respect to other speaker j is calculated using power P _{3, i.} a loud determination unit generates a vocal tract spectrum v _{i_2} estimate S _{i_2} corresponding to the threshold a or more loud degree E _{i_2,} delta features from vocal tract spectrum v _{i_2} v and delta feature quantity calculation unit for calculating a _{i_2,} except an estimate S _{i - 3} which is substantially 0. The interval corresponding to the delta feature quantity Delta] v _{i - 3} exceeding the threshold value B, and combining the estimated value S _{i - 4} voice, mixed A mixed signal generation unit for generating a signal.

In order to solve the above-described problem, according to another aspect of the present invention, a specific speech suppression method uses iε {1,..., M} and a speech signal including speech of M speakers. A sound source separation step for generating an estimated value S _i of the voice of the person i and a power parameter P _{1, i} corresponding to the power of the voice of the speaker i, j∈ {1,..., M} \ i, Using the distance d (U, L _{2, i} ) between the microphone and the speaker i used when collecting the signal, the power P ₂ of the voice of the speaker i specified by the power parameter P _{1, i i} is corrected, corrected power P _{3, i} is generated, and loudness E _i representing the loudness of speaker i with respect to other speaker j is calculated using power P _{3, i.} a loud determining step, except for the estimated value S _{i_2} corresponding to the threshold a or more loud degree E _{i_2,} synthesizes the estimated values S _{i - 3} audio, the mixed audio signal Comprising a mixed signal generation step of forming, the.

In order to solve the above-described problem, according to another aspect of the present invention, a specific speech suppression method uses iε {1,..., M} and a speech signal including speech of M speakers. A sound source separation step for generating an estimated value S _i of the voice of the person i and a power parameter P _{1, i} corresponding to the power of the voice of the speaker i, j∈ {1,..., M} \ i, Using the distance d (U, L _{2, i} ) between the microphone and the speaker i used when collecting the signal, the power P ₂ of the voice of the speaker i specified by the power parameter P _{1, i i} is corrected, corrected power P _{3, i} is generated, and loudness E _i representing the loudness of speaker i with respect to other speaker j is calculated using power P _{3, i.} a loud determining step, and generates a vocal tract spectrum v _{i_2} estimate S _{i_2} corresponding to the threshold a or more loud degree E _{i_2,} from the vocal tract spectrum v _{i_2} Except delta feature quantity calculation step of calculating the filter characteristic quantity Delta] v _{i_2,} the estimated value S _{i - 3} which is substantially 0. The interval corresponding to the delta feature quantity Delta] v _{i - 3} exceeding the threshold value B, by combining the estimated values S _{i - 4} voice And a mixed signal generating step for generating a mixed signal.

  According to the present invention, it is possible to suppress the voice of a specific speaker from the mixed voice signal.

The functional block diagram of the specific audio | voice suppression apparatus which concerns on 1st embodiment. The figure which shows the example of the processing flow of the specific audio | voice suppression apparatus which concerns on 1st embodiment. The functional block diagram of the specific audio | voice suppression apparatus which concerns on 2nd embodiment. The figure which shows the example of the processing flow of the specific audio | voice suppression apparatus which concerns on 2nd embodiment.

  Hereinafter, embodiments of the present invention will be described. In the drawings used for the following description, constituent parts having the same function and steps for performing the same process are denoted by the same reference numerals, and redundant description is omitted.

<Specific Voice Suppressing Device According to First Embodiment>
FIG. 1 is a functional block diagram of the specific speech suppressing apparatus 100 according to the first embodiment, and FIG. 2 shows an example of its processing flow.

  The specific speech suppressing apparatus 100 includes a sound source separation unit 110, a loud voice determination unit 120, and a mixed signal generation unit 140.

  The specific speech suppressing apparatus 100 receives the mixed speech signal X (t) including the speech of M speakers, generates and outputs a mixed speech signal ^ X (t) in which the speech of the specific speaker is suppressed. Note that t is an index representing time.

  The input mixed audio signal X (t) may be an audio signal collected in real time, or may be an audio signal recorded in advance such as a television program or a sports video.

<Sound source separation unit 110>
The sound source separation unit 110 receives the mixed speech signal X (t), and uses the conventional sound source separation technology to estimate the speech signal (sound source signal) S _i (t) of each speaker i (sound source), A power parameter P _{1, i} (t) corresponding to the voice power of the speaker i and a sound source position parameter L _{1, i} (t) corresponding to the position of the speaker i are calculated (s110) and output. Note that i is an index representing a speaker, and iε {1,..., M}. For example, Reference 1 can be used as a conventional technique for sound source separation.
(Reference 1) JP 2012-173592 A

<Large voice determination unit 120>
The loudness determination unit 120 receives M power parameters P _{1, i} (t) and M sound source position parameters L _{1, i} (t), and determines whether or not the voice of the speaker i is loud. (S120), a set of indices i ₂ (t) of loud speakers is output. Note that determination processing is performed on the voices of all the speakers i.

  For example, the loudness determination unit 120 includes a distance calculation unit 121, a power correction unit 122, a loudness calculation unit 123, and a first determination unit 124.

(Distance calculation unit 121)
The distance calculation unit 121 receives M sound source position parameters L _{1, i} (t). The distance calculation unit 121 identifies the position L _{2, i} of the speaker i using the sound source position parameter L _{1, i} (t). Distance calculation 121, using the position U of the microphones used in picking up mixed sound signal X (t) and position L _{2, i,} the distance d (U a microphone and speaker i, L _{2 , i} , t) is calculated (s121) and output. The sound source position parameter L _{1, i} (t) may be a parameter for specifying the position L _{2, i} of the speaker i, and may be the position L _{2, i} itself. For example, the Euclidean distance may be used as the distance d (U, L _{2, i} , t). The microphone position U is given in advance by a user or the like.

The positional relationship between the microphone and the speaker i may be premised on not depending on time. In that case, the distance may be given in advance, and the distance calculation unit 121 may not be provided. In this case, the sound source separation unit 110 may be configured not to obtain M sound source position parameters L _{1, i} (t). Alternatively, the distance may be calculated only once, and the following processing may be repeated using the distance.

(Power correction unit 122)
The power correction unit 122 receives M power parameters P _{1, i} (t) and M distances d (U, L _{2, i} , t). The power correction unit 122 specifies the power P _{2, i} (t) of the voice of the speaker i using the power parameter P _{1, i} (t). The power correction unit 122 corrects the power P _{2, i} (t) of the voice of the speaker i using the distance d (U, L _{2, i} , t), and the corrected power P _{3, i} of the voice. (t) is generated (s122) and output.

If the speaker i _A, which are in the distance A from the microphone, and a speaker i _B that are in the distance B (> A) from the microphone is speaking in the voice of the same size, distance is less power _{P 2,} i_A (t) _Becomes larger than the power P _{2, i_B} (t) having a larger distance. This point is corrected using the distance d (U, L _{2, i} , t). Note that “_ (under bar)” in the subscript indicates that the character immediately after the under bar is the subscript of the immediately preceding character. That is, X_Y in the subscript indicates _XY .

For example, the distance _{d (U, L 2, i} , t) as the audio power increases as increases, in other words, the distance _{d (U, L 2, i} , t) the speech power with increasing the The power of the sound is corrected so as to increase monotonously. For example, the correction is performed by normalizing the power P _{2, i} (t) by the distance d (U, L _{2, i} , t) according to the following equation.
P _{3, i} (t) = P _{2, i} (t) / d '(U, L _{2, i} , t)
However, d ′ (U, L _{2, i} , t) is a function that monotonously decreases as the distance d (U, L _{2, i} , t) increases. It is known that the sound power decreases by about 6 dB when the distance is doubled. Based on this characteristic, the power P _{3, i} (t) may be obtained.

(Loudness calculator 123)
Loud calculation unit 123 receives the M power P _{3, i} (t), using the M power P _{3, i} (t), the degree of loud to other speaker j speaker i The loudness E _i (t) to be expressed is calculated (s123) and output. For example, loudness E _i (t) is calculated by the following equation.

(First determination unit 124)
The first determination unit 124 receives M loudnesses E _i (t), compares the _loudness E _i (t) with a threshold A, and an index i ₂ (corresponding to _loudness E _{i_2} greater than or equal to the threshold A) ( Output a set of t).

The difference between the power P _{3, i} (t) of one corrected sound and the power P _{3, j} (t) of the remaining corrected sound is calculated, and the remaining power P _{3, j} (t) is calculated. Is a process for identifying a speaker i having a power P _{3, i} (t) greater than or equal to a predetermined threshold A. There is a high possibility that an audio signal that is 30 dB or more higher than the power of the audio signal of another speaker (Reference Document 2) is annoying voice such as a screaming voice. Utilizing this fact, the first determination unit 124 identifies the speaker index of the voice signal whose volume is particularly large compared to other voice signals. For example, the threshold A is set to 30 dB.
(Reference 2) Nanjo, Kunimatsu, Kawano, Nakayama, Nishiura, “Fundamental study of cry for acoustic security system”, 2008 Acoustical Society Spring Meeting, 1-Q-17, 2008.

<Mixed signal generator 140>
The mixed signal generation unit 140 receives a set of indexes i ₂ (t) and M estimated values S _i (t), and corresponds to the index i ₂ (t) from the M estimated values S _i (t). except for estimate S _{i_2} (t) of the audio signal, generating a synthesized and mixed sound signals ^ X (t) the estimated value S _{i - 3} (t) of the remaining audio signal (s140), and outputs.

<Effect>
With such a configuration, the voice of a specific speaker can be suppressed from the mixed voice signal, and the mixed voice signal suppresses a voice that gives an unpleasant impression to the listener (for example, loud voices including field and screams). An audio signal can be generated.

<Modification>
In the present embodiment, the sound source separation unit 110 calculates a sound source position parameter L _{1, i} (t) corresponding to the position of the speaker i from the mixed speech signal X (t), and the distance calculation unit 121 uses the sound source position parameter. The distance d (U, L _{2, i} , t) between the microphone and the speaker i is calculated using L _{1, i} (t) and the microphone position U given in advance. However, in the known sound source separation method, the distance d (U, L2 _{, i} , t) between the microphone microphone and the speaker i is determined from the mixed sound signal X (t) without being given the microphone position U in advance. There is also a method that can be calculated. Using such a known sound source separation method, the sound source separation unit receives the mixed speech signal X (t), and estimates S _i (t) of the speech signal (sound source signal) of each speaker i (sound source). And the power parameter P _{1, i} (t) corresponding to the voice power of the speaker i and the distance d (U, L _{2, i} , t) between the microphone and the speaker i are calculated and output. Also good. In that case, the distance calculation unit 121 may not be provided.

<Specific Voice Suppression Device 200 According to Second Embodiment>
A description will be given centering on differences from the first embodiment.

  FIG. 3 is a functional block diagram of the specific speech suppressing apparatus 200, and FIG. 4 shows an example of the processing flow.

  The specific speech suppressing apparatus 100 includes a sound source separation unit 110, a loud voice determination unit 120, a screaming determination unit 230, and a mixed signal generation unit 240.

<Scream Determination Unit 230>
The scream determination unit 230 receives the set of indexes i ₂ (t) and the M estimated values S _i (t), determines whether or not the voice of the speaker i ₂ (t) is a scream (s230), Outputs the index i ₃ (t) of the screaming speaker. Note that the determination process is performed only on the voice of the speaker corresponding to the index i ₂ (t), not all the speakers i.

  For example, the screaming determination unit 230 includes a vocal tract spectrum generation unit 231, a delta feature amount calculation unit 232, and a second determination unit 233.

(Vocal tract spectrum generation unit 231)
The vocal tract spectrum generation unit 231 receives a set of indexes i ₂ (t) and M estimated values S _i (t), and estimates S _{i_2} (t) of speech signals corresponding to the index i ₂ (t). of the vocal tract spectrum v generates _{i_2} a (t) (s231), and outputs a set of vocal tract spectrum v _{i_2} (t).

(Delta feature amount calculation unit 232)
Delta feature quantity calculation unit 232 receives a set of vocal tract spectrum v _{i_2} (t), using these values, to calculate a delta feature quantity _{Δv i_2 (t) (s232)} , the delta feature amount Delta] v _{i_2} (t ) Is output. For example, (see reference 3) using the following equation to calculate a delta feature quantity Delta] v _{i_2} a (t).

Here, w ₁ is set so that 2w ₁ +1 is, for example, 50 milliseconds.
(Reference 3) FURUI S., "Speaker-independent isolated word recognition using dynamic features of speech spectrum", IEEE Trans. Acoust., Speech and Signal Processing ASSP-34 (1), 1986, pp. 52-59.

(Second determination unit 233)
Second determination unit 233 receives a set of delta feature quantity Δv _{i_2} (t), delta feature quantity Delta] v _{i_2} almost becomes zero interval to determine whether more than a threshold value B, delta characteristic amount exceeding the threshold value B and it outputs a set of index i ₃ (t) corresponding to the _Δv i_3.

Incidentally, it may be determined whether the absolute value is less than or equal to a positive value ε small enough interval of the delta feature quantity Δv _{i_2} (t) exceeds the threshold value B, the following values are sufficiently small positive You may determine whether it is below the value (epsilon).

Here, w ₂ is set so that 2w ₂ +1 becomes, for example, a threshold B (for example, 300 milliseconds).

  The delta feature amount is a feature amount that represents a change in sound for each predetermined time interval, and the greater the value, the greater the change in sound. The state in which the delta feature value is almost 0 is a state in which there is no change in sound, and it is either a state where no voice is produced (silence) or a state where the sound is extended. It is assumed. However, in the present embodiment, since only the sound signal having a large sound power is the analysis target of the screaming determination unit 230 in the first determination unit 124, there is no possibility of silence. In other words, the screaming determination unit 230 determines whether or not the sound is being extended, and extracts the speaker index of the voice signal that is likely to increase the sound.

  Shouting generally tends to stretch vowels. Since it is almost impossible to extend the sound longer than that in the calm state with reference to the length of the long vowel in the speech database, it is possible to extract a speech signal that is likely to be a scream by this process. .

<Mixed signal generator 240>
The mixed signal generation unit 240 receives a set of indexes i ₃ (t) and M estimated values S _i (t), and _calculates an estimated value S i — ₃ (t) of the speech signal corresponding to the index i ₃ (t). Except for this, the estimated value S _{i — 4} (t) of the remaining audio signal is synthesized to generate a mixed audio signal ^ X (t) (s240) and output.

<Effect>
With such a configuration, it is possible to accurately suppress loud voices, particularly voices that are annoying, such as field and screams, and do not contain important information for the viewer.

<Other variations>
The present invention is not limited to the above-described embodiments and modifications. For example, the various processes described above are not only executed in time series according to the description, but may also be executed in parallel or individually as required by the processing capability of the apparatus that executes the processes. In addition, it can change suitably in the range which does not deviate from the meaning of this invention.

<Program and recording medium>
In addition, various processing functions in each device described in the above embodiments and modifications may be realized by a computer. In that case, the processing contents of the functions that each device should have are described by a program. Then, by executing this program on a computer, various processing functions in each of the above devices are realized on the computer.

  The program describing the processing contents can be recorded on a computer-readable recording medium. As the computer-readable recording medium, for example, any recording medium such as a magnetic recording device, an optical disk, a magneto-optical recording medium, and a semiconductor memory may be used.

  The program is distributed by selling, transferring, or lending a portable recording medium such as a DVD or CD-ROM in which the program is recorded. Further, the program may be distributed by storing the program in a storage device of the server computer and transferring the program from the server computer to another computer via a network.

  A computer that executes such a program first stores, for example, a program recorded on a portable recording medium or a program transferred from a server computer in its storage unit. When executing the process, this computer reads the program stored in its own storage unit and executes the process according to the read program. As another embodiment of this program, a computer may read a program directly from a portable recording medium and execute processing according to the program. Further, each time a program is transferred from the server computer to the computer, processing according to the received program may be executed sequentially. Also, the program is not transferred from the server computer to the computer, and the above-described processing is executed by a so-called ASP (Application Service Provider) type service that realizes the processing function only by the execution instruction and result acquisition. It is good. Note that the program includes information provided for processing by the electronic computer and equivalent to the program (data that is not a direct command to the computer but has a property that defines the processing of the computer).

  In addition, although each device is configured by executing a predetermined program on a computer, at least a part of these processing contents may be realized by hardware.

  The present invention can be applied to, for example, audio output of a television. When watching a program such as a live broadcast, it is possible to suppress unpleasant sounds such as the field and screaming and comfortably view it.

For example, it can be implemented on the receiving side of a television. For example, a specific voice suppression device is incorporated in the television, and the reception side suppresses field screams and screams. On the receiving side, the microphone position U or the distance between the microphone and the speaker can be acquired in advance as data. For example, when the microphone position U can be acquired as data on the receiving side, the processing described in the first embodiment may be performed. When the distance d (U, L _{2, i} , t) between the microphone and the speaker can be acquired as data on the receiving side, the server on the broadcasting station side includes a sound source separation unit and a distance calculation unit, and the receiving side has a sound source A configuration other than the separation unit and the distance calculation unit may be included. The server on the broadcast station side executes a sound source separation unit and a distance calculation unit on the collected sound signal (mixed sound signal), and the estimated value S _i (t) of the sound signal of the speaker i and the sound source The power parameter P _{1, i} (t) and the distance d (U, L _{2, i} , t) between the microphone and the speaker i are obtained. Then, the estimated value S _i (t) of the voice signal of the speaker i, the sound source power parameter P _{1, i} (t), and the distance d (U, L _{2, i} , t) between the microphone and the speaker i. Are distributed to the receiving side. On the receiving side, the estimated value S _i (t) of the voice signal of the speaker i, the sound source power parameter P _{1, i} (t), and the distance d (U, L _{2, i} , Using t), the same processing as the first embodiment can be obtained by performing the processing after the loudness determination unit. As described in the modification of the first embodiment, the sound source separation unit calculates the distance d (U, L _{2, i} , t) between the microphone and the speaker i from the mixed sound signal X (t). If possible, the server on the broadcasting station side only needs to distribute at least the mixed audio signal X (t), and the receiving side may perform the processing described in the first embodiment and its modifications.

  It can also be implemented on the TV broadcasting station side. The mixed sound signal output from the microphone is used as the input of the specific sound suppressing device, and the broadcasting station may transmit the output signal of the specific sound suppressing device. In this case, it is possible to suppress the field and screams at the time of distribution. The position U of the microphone may be input by the user, or may be acquired using a camera image or the like. The speaker position is not necessarily acquired by the sound source separation unit 110, and a camera image or the like may be used.

Also, for example, it can be applied to a hearing aid. For example, a specific voice suppression device may be incorporated in the hearing aid. The position U of the microphone is the position of the hearing aid, and the position of the speaker i with respect to the hearing aid is the distance d (U, L _{2, i} , t).

Claims (7)

Assume that i ∈ {1,..., M}, and from a speech signal including the speech of M speakers, an estimated value S i of the speech of speaker _i and a power parameter P ₁ corresponding to the speech power of speaker i _{, i} and a sound source separation unit,
j∈ {1,..., M} \ i, and using the distance d (U, L _{2, i} ) between the microphone and the speaker i used for collecting the voice signal, the power parameter P _{1 , i} is used to correct the power P _2, i of the voice of the speaker _i specified by i _{, i} to generate a corrected power P _{3, i} of the voice _, and using the power P _{3, i} , A loudness determination unit for calculating loudness E _i representing the loudness of the speaker j
Except for the estimated value S _{i_2} corresponding to the threshold A or more loud degree E _{i_2,} including by combining the estimated value S _{i - 3} audio, the mixing signal generator for generating a mixed sound signal, and
Specific voice suppression device. The characteristic speech suppressing apparatus of claim 1,
The threshold A is 30 dB.
Characteristic voice suppression device. Assume that i ∈ {1,..., M}, and from a speech signal including the speech of M speakers, an estimated value S i of the speech of speaker _i and a power parameter P ₁ corresponding to the speech power of speaker i _{, i} and a sound source separation unit,
j∈ {1,..., M} \ i, and using the distance d (U, L _{2, i} ) between the microphone and the speaker i used for collecting the voice signal, the power parameter P _{1 , i} is used to correct the power P _2, i of the voice of the speaker _i specified by i _{, i} to generate a corrected power P _{3, i} of the voice _, and using the power P _{3, i} , A loudness determination unit for calculating loudness E _i representing the loudness of the speaker j
It generates a vocal tract spectrum v _{i_2} estimate S _{i_2} corresponding to the threshold A or more loud degree E _{i_2,} and delta feature quantity calculation unit for calculating a delta feature quantity Delta] v _{i_2} from the vocal tract spectrum v _{i_2,}
A mixed signal generating unit that generates a mixed signal by synthesizing the estimated value S _{i_4 of the} speech by excluding the estimated value S _{i_3} corresponding to the delta feature quantity Δv _{i_3} in which the section that is substantially zero exceeds the threshold B ,
Specific voice suppression device. The characteristic speech suppressing apparatus according to claim 3,
The threshold A is 30 dB, and the threshold B is 300 milliseconds.
Characteristic voice suppression device. Assume that i ∈ {1,..., M}, and from a speech signal including the speech of M speakers, an estimated value S i of the speech of speaker _i and a power parameter P ₁ corresponding to the speech power of speaker i _{, i} and a sound source separation step;
j∈ {1,..., M} \ i, and using the distance d (U, L _{2, i} ) between the microphone and the speaker i used for collecting the voice signal, the power parameter P _{1 , i} is used to correct the power P _2, i of the voice of the speaker _i specified by i _{, i} to generate a corrected power P _{3, i} of the voice _, and using the power P _{3, i} , Loudness determination step of calculating loudness E _i representing the loudness of the speaker j
Except for the estimated value S _{i_2} corresponding to the threshold A or more loud degree E _{i_2,} including by combining the estimated value S _{i - 3} audio, the mixed signal generation step of generating a mixed sound signal, and
Specific voice suppression method. Assume that i ∈ {1,..., M}, and from a speech signal including the speech of M speakers, an estimated value S i of the speech of speaker _i and a power parameter P ₁ corresponding to the speech power of speaker i _{, i} and a sound source separation step;
j∈ {1,..., M} \ i, and using the distance d (U, L _{2, i} ) between the microphone and the speaker i used for collecting the voice signal, the power parameter P _{1 , i} is used to correct the power P _2, i of the voice of the speaker _i specified by i _{, i} to generate a corrected power P _{3, i} of the voice _, and using the power P _{3, i} , Loudness determination step of calculating loudness E _i representing the loudness of the speaker j
It generates a vocal tract spectrum v _{i_2} estimate S _{i_2} corresponding to the threshold A or more loud degree E _{i_2,} and delta feature quantity calculation step of calculating a delta feature quantity Delta] v _{i_2} from the vocal tract spectrum v _{i_2,}
And a mixed signal generating step of generating a mixed signal by synthesizing the estimated value S _{i_4 of the} speech by excluding the estimated value S _{i_3} corresponding to the delta feature quantity Δv _{i_3} in which the interval that is substantially zero exceeds the threshold B. ,
Specific voice suppression method.   The program for functioning a computer as a specific audio | voice suppression apparatus in any one of Claims 1-4.

Images