JP2007233284A - Voice processing device and voice processing method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To protect privacy of a speaker without displeasing a person by effectively preventing voice superimposed on the speaker's voice to be output from becoming high-pitched voice. <P>SOLUTION: By storing data concerning two or more different spectral envelopes in a spectral envelope database 15, extracting a spectral fine structure 14 from voice signals of the speaker, and selecting a data concerning the spectral envelope from among the data concerning the spectral envelope stored in the spectral envelope database 15, a spectrum 17 of the anti-hearing sound output by synthesizing the selected spectral envelope and the spectral fine structure and superimposing the speaker's voice on the synthetic sound is created. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a speech processing apparatus and speech processing method for generating a speech spectrum of speech output over a speaker's voice in order to protect the privacy of the speaker (including confidentiality of conversation). TECHNICAL FIELD The present invention relates to a speech processing apparatus and a speech processing method that can effectively prevent a sound output from being covered with a voice of a voice and protect a speaker's privacy without causing discomfort to a person. .

  Conventionally, in open spaces such as banks, hospitals, and securities companies, conversations related to privacy have been frequently performed. For this reason, for the purpose of protecting the privacy of a speaker, a masking device that outputs a masking sound with respect to a speaker's voice (voice by conversation) has been developed (see, for example, Patent Document 1).

  Specifically, white noise, BGM, and other disturbing sounds that obscure the speaker's speech are output over the speaker's speech as the “masking sound”, and the speaker's speech is erased. Protect the speaker's privacy by making the content difficult to hear.

JP-A-6-175666

  However, the above-described prior art has a problem that the privacy of the speaker cannot be protected with high accuracy. That is, in the above-described prior art, since the voice having little relation to the voice of the speaker is output as the masking sound, the voice of the speaker and the masking sound are recognized as separate voices and recognized by the listener. As a result, the privacy of the speaker could not be protected with high accuracy.

  In order to solve such a problem, it is also conceivable to generate a hearing-proof sound (described later) using the voice of the speaker and output it as a masking sound. Specifically, the hearing spectrum is generated by detecting the speech spectrum of the speech voice of the speaker and transforming the speech spectrum by inverting and shifting the positions of peaks and valleys in the speech spectrum.

  However, when the peaks and valleys in the voice spectrum are simply inverted / shifted, the hearing loss sound becomes high-pitched sound, and there is a problem in that the hearing person is uncomfortable. Therefore, it is an important problem how to effectively prevent the hearing loss from becoming high-pitched sound and not to give unpleasant feeling to the person who hears the hearing loss.

  The present invention has been made to solve the above-described problems caused by the prior art, and effectively prevents the hearing loss from becoming a high-pitched sound. An object of the present invention is to provide an audio processing apparatus and an audio processing method capable of protecting privacy.

  In order to solve the above-described problems and achieve the object, a speech processing device according to claim 1 is a speech processing device that generates a speech spectrum of speech output over a speaker's speech, and is different. A spectral envelope database for storing data related to a plurality of spectral envelopes, a spectral fine structure extracting means for extracting a spectral fine structure from a speech signal of a speaker, and data related to a spectral envelope stored in the spectral envelope database Spectral envelope selection means for selecting data related to the spectral envelope, and the voice spectrum of the voice output over the voice of the speaker by synthesizing the spectral envelope selected by the spectral envelope selection means and the spectral fine structure. And a voice spectrum generating means for generating.

  According to a second aspect of the present invention, there is provided the speech processing apparatus according to the first aspect of the present invention, wherein the spectrum envelope selection means is configured to generate a spectrum from the spectrum envelope database when a temporal change amount of a speaker's voice is a predetermined value or more. The data related to the envelope is newly selected, and the speech spectrum generation means outputs the voice over the speaker by synthesizing the spectrum envelope newly selected by the spectrum envelope selection means and the spectrum fine structure. A voice spectrum of voice is newly generated.

  According to a third aspect of the present invention, there is provided the speech processing apparatus according to the first or second aspect, wherein the spectrum envelope selecting means relates to a spectrum envelope from data related to a spectrum envelope stored in the spectrum envelope database. Data is selected at random.

  According to a fourth aspect of the present invention, there is provided a speech processing apparatus according to the first or second aspect of the present invention, further comprising spectrum envelope extraction means for extracting a spectrum envelope from a speech signal of a speaker, wherein the spectrum envelope selection means is Based on the degree of similarity between the spectrum envelope extracted by the spectrum envelope extraction means and the spectrum envelope stored in the spectrum envelope database, the data related to the spectrum envelope stored in the spectrum envelope database It is characterized by selecting data relating to the spectral envelope.

  According to a fifth aspect of the present invention, there is provided a voice processing apparatus for generating a voice spectrum of a voice outputted over a speaker's voice, and extracting a spectral fine structure from the speaker's voice signal. A speech spectrum for generating a speech spectrum of speech output over a speaker's voice by synthesizing a spectral fine structure extraction means and the spectral fine structure extracted by the spectral fine structure extraction means and a predetermined spectral envelope It is characterized by comprising: generating means; and frequency intensity correcting means for correcting the sound spectrum by suppressing the spectrum intensity in a predetermined frequency region of the sound spectrum generated by the sound spectrum generating means.

  According to a sixth aspect of the present invention, in the fifth aspect of the present invention, the frequency intensity correcting means includes a voice spectrum obtained from a voice signal of a speaker and a voice generated by the voice spectrum generating means. The correction amount of the spectrum intensity is set based on the difference from the spectrum.

  According to a seventh aspect of the present invention, there is provided a voice processing method for generating a voice spectrum of a voice outputted over a speaker's voice, and extracting a spectral fine structure from the speaker's voice signal. A spectrum for selecting data related to a spectrum envelope from a plurality of different spectrum envelope data stored in advance in a spectrum envelope database when a spectrum fine structure is extracted by the spectrum extraction step and the spectrum fine structure extraction step An envelope selection step, and a speech spectrum generation step of generating a speech spectrum of speech output over a speaker's speech by synthesizing the spectrum envelope selected by the spectrum envelope selection step and the spectrum fine structure. It is characterized by that.

  The voice processing method according to the invention of claim 8 is a voice processing method for generating a voice spectrum of a voice outputted over a speaker's voice, and extracts a spectral fine structure from the voice signal of the speaker. A speech spectrum for generating a speech spectrum of speech output over a speaker's speech by synthesizing a spectral fine structure extracted step and a spectral fine structure extracted by the spectral fine structure extracting means and a predetermined spectral envelope It includes a generation step and a frequency intensity correction step of correcting the voice spectrum by suppressing the spectrum intensity in a predetermined frequency region of the voice spectrum generated by the voice spectrum generation step.

  According to the invention of claim 1 or 7, the spectrum envelope database stores data related to a plurality of different spectrum envelopes, extracts the spectral fine structure from the speech signal of the speaker, and stores the spectrum envelope stored in the spectrum envelope database. Since the data related to the spectral envelope is selected from the data, and the selected spectral envelope and the spectral fine structure are synthesized, the voice spectrum of the voice output over the speaker's voice is generated. Since the spectrum of the hearing loss is generated using the spectral fine structure and the spectrum envelope that hold the speaker's sound source information, the hearing loss holds the speaker's sound source information. This makes it difficult to hear the speaker's speech and transforms the speaker's voice spectrum to reduce hearing loss. Rather than creating a hearing loss using a spectrum envelope pre-registered in the spectrum envelope database, it effectively prevents the hearing loss from becoming a high pitched sound and causing discomfort to the person. There is an effect that can be.

  According to the second aspect of the present invention, when the amount of time change of the speaker's voice is equal to or greater than a predetermined value, data related to the spectrum envelope is newly selected from the spectrum envelope database, and the newly selected spectrum envelope is selected. Since the speech spectrum of the speech that is output over the speaker's voice is newly generated by synthesizing with the spectral fine structure, the spectrum envelope suitable for making it difficult to hear the speech content of the speaker is spoken. The effect is that it is possible to select following the change of the person's voice.

  According to the invention of claim 3, since the data related to the spectrum envelope is randomly selected from the data related to the spectrum envelope stored in the spectrum envelope database, there is no discomfort to the person. There is an effect that it is possible to efficiently select a spectrum envelope used for generation of the hearing protection sound.

  According to the invention of claim 4, the spectrum envelope is further extracted from the speech signal of the speaker, and based on the similarity between the extracted spectrum envelope and the spectrum envelope whose data is stored in the spectrum envelope database. Thus, since the data related to the spectral envelope is selected from the data related to the spectral envelope stored in the spectral envelope database, it is possible to effectively select the spectral envelope representing the phoneme far from the speaker's phoneme. It is possible to produce a hearing-proof sound that makes it difficult to hear the content of the speaker's speech.

  According to the invention of claim 5 or 8, the spectral fine structure is extracted from the voice signal of the speaker, and the extracted spectral fine structure and a predetermined spectral envelope are synthesized to cover the voice of the speaker. Since the sound spectrum of the output sound is generated and the sound spectrum is corrected by suppressing the spectrum intensity in a predetermined frequency region of the generated sound spectrum, the hearing loss becomes a high-pitched sound. By suppressing the spectral intensity in the frequency domain, it is possible to effectively prevent a person from feeling uncomfortable.

  According to the invention of claim 6, a speech spectrum generated by synthesizing the spectral fine structure extracted from the speech signal of the speaker and a predetermined spectrum envelope, and a speech spectrum obtained from the speech signal of the speaker Since the correction amount of the spectral intensity is set based on the difference between the two, the effect of being able to appropriately set the correction amount of the spectral intensity in the frequency domain that causes the hearing-proof sound to become a high-pitched sound is obtained.

  Exemplary embodiments of an audio processing device and an audio processing method according to the present invention will be explained below in detail with reference to the accompanying drawings.

  First, the concept of audio processing according to the first embodiment will be described. FIG. 1 is a diagram illustrating the concept of sound processing according to the first embodiment.

  As shown in FIG. 1, this voice processing is intended to prevent the voice of the speaker from obscuring the conversation voice (this voice is intended to prevent the voice of the conversation from being heard by a third party) In the following, this voice is called hearing-proof sound), and the speaker's voice signal is acquired by a microphone, etc., and the spectrum of the voice signal is analyzed at a predetermined time interval, and the conversation such as sound pressure and frequency distribution is performed. Extract voice features. FIG. 1 shows an example of a spectrogram 11 obtained as a result of applying spectrum analysis to a speech waveform 10 of “A”, “I”, “U”, “E”, and “O”.

  Then, from the short-time spectrum 12 obtained from such a spectrogram 11, a spectrum envelope 13 representing phonological information and a spectral fine structure 14 representing sound source information are extracted.

  On the other hand, in this speech processing, representative speech signals of a person are extracted in advance using a statistical method such as clustering, and a plurality of spectrum envelopes of the extracted speech signals are registered in the spectrum envelope database 15.

  Then, the spectrum envelope that is not most similar to the spectrum envelope 13 obtained from the spectrum analysis of the speaker's speech (the spectrum distance is maximum) is selected from the spectrum envelopes registered in the spectrum envelope database 15; The selected spectrum envelope 16 replaces the spectrum envelope 13 obtained from the speech of the speaker.

  Here, the reason why the spectrum envelope most similar to the spectrum envelope 13 obtained from the spectrum analysis of the speaker's speech is selected is based on the spectrum envelope representing the phoneme far from the phoneme of the speaker's speech. This is to make it difficult to hear the content of the speaker's speech by generating a hearing-proof sound.

  Subsequently, the spectrum envelope 16 obtained by replacing the spectrum envelope 13 obtained from the conversational voice of the speaker and the spectrum fine structure 14 are synthesized, and a spectrum 17 of the hearing loss sound is generated. And the hearing-aid sound is produced | generated from the spectrum 17 of this hearing-aid sound, and the hearing-aid sound is output from a speaker.

  In this way, in this audio processing, the spectrum of the hearing loss sound is generated using the spectrum fine structure 14 holding the sound source information of the speaker and the spectrum envelope of the representative voice signal of the person. Since the sound source information of the speaker is held, the hearing-proof sound can be fused with the conversation voice of the speaker, and the content of the speaker's speech can be made difficult to hear.

  In addition, hearing loss is not generated by inverting and shifting the positions of peaks and valleys in the speaker's speech spectrum, but rather by using the spectral envelope of a typical human speech signal as it is. It is possible to prevent the hearing sound from becoming unnaturally high-pitched sound and causing discomfort to the person who hears the hearing-proof sound.

  FIG. 2 is a diagram showing a comparison between the spectrogram of the hearing protection sound in the first embodiment and the spectrogram of the conventional hearing protection sound. FIG. 2 shows a spectrogram 20 of the original voice uttered by the speaker, a spectrogram 21 of the conventional hearing aid sound, and a spectrogram 22 of the hearing aid sound of the present embodiment. In each spectrogram of FIG. 2, dark regions correspond to regions with high intensity.

  Here, the spectrogram 21 of the conventional hearing protection sound is obtained by inverting and shifting the positions of peaks and valleys in the spectrum envelope 13 of the speaker's conversational speech and synthesizing with the spectrum fine structure 14. Spectrogram. Further, the hearing-aid spectrogram 22 of this embodiment is a spectrogram of the hearing-aid generated as described with reference to FIG.

  As shown in FIG. 2, in the conventional deafening sound spectrogram 21, the intensity in the frequency range of the middle range (1 kHz to 4 kHz) is increased compared to the spectrogram 20 of the original speech. And the increase in the intensity in the mid frequency range causes a high-pitched sound.

  On the other hand, in the spectrogram 22 of the hearing-aid sound of the present embodiment, it can be seen that an increase in intensity in the mid-frequency range is suppressed. In this way, by suppressing an increase in intensity in the mid-frequency range, it is possible to effectively prevent the hearing loss from becoming a high-pitched sound.

  Note that when the spectrum of the hearing loss sound is generated using the spectrum envelope of a typical human voice signal as in the present embodiment, the time change of the spectrogram 20 of the original voice as shown in FIG. In response, the spectral envelope used to generate the hearing loss is reselected.

  Specifically, as shown in FIG. 2, when the time change of the original speech exceeds a predetermined value, the spectrum envelope most similar to the spectrum envelope extracted from the speaker's speech signal is the spectrum. A new spectrum is selected from the spectrum envelopes registered in the envelope database 14, and the spectrum of the hearing loss sound is generated using the selected spectrum envelope. As a result, it is possible to appropriately select a spectrum envelope suitable for making it difficult to hear the content of the speaker's speech following the change in the original speech.

  Next, a functional configuration of the speech processing apparatus according to the first embodiment will be described. FIG. 3 is a diagram illustrating a functional configuration of the voice processing device 30 according to the first embodiment. As illustrated in FIG. 3, the speech processing device 30 includes an input unit 31, a display unit 32, a speech input reception unit 33, a spectrum envelope database 34, a speech generation unit 35, a speech output unit 36, and a control unit 37.

  The input unit 31 is an input device such as a keyboard and a mouse used for inputting various types of information. The display unit 32 is a display device such as a display that outputs various types of information. The voice input receiving unit 33 is a receiving unit that receives a speaker's voice signal from a microphone or the like, performs A / D conversion and amplification processing, and outputs the result to the control unit 37.

  The spectrum envelope database 34 is a database that stores spectrum envelope data that are candidates for replacing the spectrum envelope extracted from the speech signal of the speaker when the spectrum of the hearing loss sound is generated as described in FIG. .

  The sound generation unit 35 is a generation unit that generates a sound signal of the hearing-aid sound from the spectrum of the hearing-aid sound generated by the control unit 37 described later. The audio output unit 36 is an output unit that performs D / A conversion and amplification processing on the audio signal generated by the audio generation unit 35 and outputs the result to a speaker.

  The control unit 37 includes a control program such as an OS (Operating System), a program that defines processing procedures for various processes, and a memory for storing various data, and is a control unit that executes various processes.

  The control unit 37 includes a spectrum analysis unit 37a, a spectrum fine structure extraction unit 37b, a spectrum envelope extraction unit 37c, a spectrum envelope selection unit 37d, and a spectrum generation unit 37e.

  The spectrum analysis unit 37a receives a digitized audio signal from the audio input reception unit 33 and performs cepstrum analysis. Among the cepstrum coefficients obtained as a result, the spectrum analysis unit 37a converts the high and low quefrency portions into a spectrum fine structure extraction unit 37b. And an analysis unit that outputs the spectrum envelope extraction unit 37c.

  Specifically, the spectrum analysis unit 37a applies a predetermined window function such as a Hanning window or a Hamming window to the audio signal, and performs short-time spectrum analysis using Fast Fourier Transform (FFT). To do.

  Subsequently, the spectrum analysis unit 37a calculates an absolute value of a value obtained as a result of the fast Fourier transform, and further calculates a logarithm of the absolute value. Then, the spectrum analyzing unit 37a applies an inverse fast Fourier transform (IFFT) to the calculated logarithmic value to calculate a cepstrum coefficient.

  Thereafter, the spectrum analyzing unit 37a extracts a high quefrency portion and a low quefrency portion by performing liftering on the calculated cepstrum coefficient using a cepstrum window.

  The spectrum analyzing unit 37a stores the spectrum of the audio signal received from the audio input receiving unit 33 in the past. Then, the spectrum analysis unit 37a calculates a spectrum distance between the spectrum of the newly received sound signal and the spectrum of the sound signal received in the past, and the spectrum envelope when the spectrum distance becomes a predetermined value or more. Processing for instructing the selection unit 37d to select a new spectrum envelope is performed.

  The spectrum fine structure extraction unit 37b is an extraction unit that receives a high quefrency part from the spectrum analysis unit 37a and extracts a spectral fine structure by applying a fast Fourier transform. The spectrum envelope extraction unit 37c is an extraction unit that receives a low quefrency unit from the spectrum analysis unit 37a and extracts a spectrum envelope by applying a fast Fourier transform.

  The spectrum envelope selection unit 37d calculates a spectral distance between the spectrum envelope extracted by the spectrum envelope extraction unit 37c and the spectrum envelope registered in the spectrum envelope database 34, and the spectrum envelope registered in the spectrum envelope database 34 Is a selection unit that selects the spectrum envelope having the maximum spectrum distance as a replacement for the spectrum envelope extracted by the spectrum envelope extraction unit 37c.

  Here, as the spectral distance, a Euclidean distance of a vector composed of the components of the low kerfrenality part is used. Note that the spectral distance used here is not limited to this, and various conventionally proposed spectral distances such as a spectral distance based on FFT and a spectral distance based on a spectral envelope obtained by linear predictive (LPC) analysis. Spectral distance may be used.

  The spectrum generation unit 37e is a generation unit that synthesizes the spectrum fine structure extracted by the spectrum fine structure extraction unit 37b and the spectrum envelope selected by the spectrum envelope selection unit 37d to generate a spectrum of hearing loss.

  Next, a processing procedure of audio processing according to the first embodiment will be described. FIG. 4 is a flowchart of the sound processing procedure according to the first embodiment. As shown in FIG. 4, first, the voice input receiving unit 33 of the voice processing device 30 receives an input of a voice signal from the microphone (step S101).

  Then, the spectrum analysis unit 37a performs spectrum analysis of the speech waveform of the input speech signal, and calculates a high quefrency portion and a low quefrency portion in the cepstrum coefficient (step S102).

  Subsequently, the spectral fine structure extraction unit 37b acquires a high quefrency part from the spectral analysis unit 37a and extracts the spectral fine structure (step S103). And the spectrum envelope extraction part 37c acquires a low quefrency part from the spectrum analysis part 37a, and extracts a spectrum envelope (step S104).

  After that, the spectrum analysis unit 37a compares the spectrum of the input voice signal with the spectrum of the voice signal input in the past, and checks whether or not the time variation of the spectrum has become a predetermined value or more (step S105). ).

  When the time variation of the spectrum is not equal to or greater than the predetermined value (No at Step S105), the spectrum envelope selection unit 37d checks whether or not the spectrum envelope has been selected (Step S106).

  If the spectrum envelope has not been selected (No at Step S106), the spectrum envelope selection unit 37d reads the spectrum envelope data registered in the spectrum envelope database 34 (Step S107).

  Subsequently, the spectrum envelope selecting unit 37d selects the spectrum envelope having the largest spectrum distance and the spectrum envelope extracted by the spectrum envelope extracting unit 37c from the spectrum envelopes registered in the spectrum envelope database 34 (step S108).

  Thereafter, the spectrum generation unit 37e generates a spectrum of the hearing loss sound obtained by synthesizing the selected spectrum envelope and the spectrum fine structure extracted by the spectrum fine structure extraction unit 37b (step S109).

  In step S105, when the time variation of the spectrum is equal to or greater than the predetermined value (step S105, Yes), the process proceeds to step S107, and the subsequent processing is continued. In step S106, when the spectrum envelope has been selected (step S106, Yes), the process proceeds to step S109, and the subsequent processing is continued.

  After step S109, the sound generation unit 35 generates a hearing-aid sound signal from the hearing-aid spectrum generated by the spectrum generation unit 37e (step S110). Then, the audio output unit 36 outputs the hearing-aid audio signal generated by the audio generation unit 35 to the speaker (step S111).

  Thereafter, the control unit 37 checks whether or not an instruction to end the hearing-proof sound output process has been issued (step S112). If an instruction to end is given (Yes in step S112), the process ends. If the end instruction has not been given (No at Step S112), the process proceeds to Step S101, and the subsequent processing is repeated.

  In step S105, when the time variation of the spectrum becomes equal to or greater than the predetermined value (step S105, Yes), the spectrum generation unit 37e rejects the set spectrum envelope, proceeds to step S107, and changes the spectrum envelope. Perform a new setting process.

  In the above embodiment, the spectrum envelope is selected from the spectrum envelopes registered in the spectrum envelope database 34 based on the spectrum distance. However, the spectrum envelope is selected from the spectrum envelopes registered in the spectrum envelope database 34. May be selected randomly, or may be selected by other methods.

  As described above, according to the first embodiment, the spectral envelope database 34 stores data related to a plurality of different spectral envelopes, and the spectral fine structure extraction unit 37b extracts the spectral fine structure from the speech signal of the speaker. The spectrum envelope selection unit 37d selects the data related to the spectrum envelope from the data related to the spectrum envelope stored in the spectrum envelope database 34, and the spectrum generation unit 37e displays the selected spectrum envelope and the spectrum fine structure. Since it was decided to generate a spectrum of the hearing loss sound that is output over the speaker's voice by combining it, the spectrum of the hearing loss sound is obtained by using the spectral fine structure and the spectrum envelope that hold the speaker's sound source information. Because the sound-proofing sound holds the sound source information of the speaker, In addition, it is possible to make it difficult to hear the content of the speaker's speech, and to prevent the use of a spectrum envelope registered in the spectrum envelope database 34 in advance, instead of generating a hearing-proof sound by modifying the speaker's speech spectrum. Since the hearing sound is generated, it is possible to effectively prevent the hearing-proof sound from becoming a high-pitched sound and causing discomfort to the person.

  Further, according to the first embodiment, the spectrum envelope selection unit 37d newly selects data related to the spectrum envelope from the spectrum envelope database 34 when the temporal change amount of the speaker's voice is equal to or greater than a predetermined value, and generates a spectrum. Since the unit 37e newly generates the speech spectrum of the speech that is output over the speech of the speaker by synthesizing the newly selected spectral envelope and the spectral fine structure, the content of the speech of the speaker Thus, it is possible to select a spectral envelope suitable for making it difficult to hear the voice following the change of the speaker's voice.

  Further, according to the first embodiment, the spectrum envelope selection unit 37d randomly selects the data related to the spectrum envelope from the data related to the spectrum envelope stored in the spectrum envelope database 34. It is possible to efficiently select a spectrum envelope used for generating a hearing-proof sound that does not give a pleasant feeling.

  Further, according to the first embodiment, the spectrum envelope extraction unit 37c extracts the spectrum envelope from the speech signal of the speaker, and the spectrum envelope selection unit 37d stores the extracted spectrum envelope and the data in the spectrum envelope database 34. Since the spectrum envelope data is selected from the spectrum envelope data stored in the spectrum envelope database 34 based on the spectrum distance between the measured spectrum envelope and the spectrum envelope database 34, it is far from the phoneme of the speaker. A spectrum envelope representing a phoneme can be effectively selected, and a hearing-proof sound that makes it difficult to hear the content of a speaker's speech can be generated.

  By the way, in Example 1, the spectrum envelope of the representative hearing sound signal is used to generate the spectrum of the hearing-aid sound in which the instep height is suppressed. By suppressing the spectral intensity in the frequency region that causes the height, the hearing-proof sound may be prevented from becoming a high-pitched sound. Therefore, in the second embodiment, a case will be described in which the spectrum intensity in the frequency domain that causes the upper height in the spectrum of the hearing-proof sound is suppressed.

  First, the concept of audio processing according to the second embodiment will be described. FIG. 5 is a diagram illustrating audio processing according to the second embodiment.

  In this sound processing, a spectrum intensity correction amount 40 in the frequency domain that causes the height of the hearing-proof sound spectrum is calculated in advance. In the example of FIG. 5, the frequency range of 1 kHz to 2 kHz is a major cause of the instep height, and the correction amount is particularly large.

  The spectral intensity correction amount 40 as shown in FIG. 5 is obtained by comparing the characteristics of the spectrum of the voice signals of various speakers with the spectrum of the hearing loss generated based on the voice signals of the speakers. Calculated.

  FIG. 6 is a diagram for explaining a method for calculating the spectrum intensity correction amount 40. As shown in FIG. 6, when calculating the spectrum intensity correction amount 40, the spectrums of the speech signals of various speakers are collected, and the average value of the collected spectra (the spectrum average of the original sound) is calculated. On the other hand, the spectrum of the hearing signal sound signal generated from the speech signals of various speakers is collected, and the average value of the collected spectrum (spectrum average of the hearing sound) is calculated.

  Then, an increase in the spectrum obtained by subtracting the spectrum average of the original sound from the spectrum average of the hearing loss (a spectrum increase of the hearing loss) is calculated. Then, a frequency band in which the spectrum increase of the hearing loss sound is a positive value is detected, and the spectrum increase amount of the hearing loss sound in that frequency band is set as a spectrum intensity correction amount 40 that is subtracted from the spectrum intensity of the spectrum of the hearing protection sound.

  In this way, by suppressing the spectral intensity in the frequency region that causes the height of the instep, it is possible to effectively prevent the hearing-proof sound from becoming a high-intensity sound.

  Next, a functional configuration of the sound processing apparatus according to the second embodiment will be described. FIG. 7 is a diagram illustrating a functional configuration of the speech processing apparatus 50 according to the second embodiment. Here, it is assumed that the speech processing device 50 generates the spectrum of the hearing-proof sound by changing the spectrum envelope extracted from the speech signal of the speaker. However, the generation method of the hearing-proof sound spectrum is not limited to this, and may be generated by other methods such as the method described in the first embodiment.

  As illustrated in FIG. 7, the voice processing device 50 includes an input unit 51, a display unit 52, a voice input reception unit 53, a voice generation unit 54, a voice output unit 55, and a control unit 56.

  The input unit 51 is an input device such as a keyboard and a mouse used for inputting various information. The display unit 52 is a display device such as a display that outputs various types of information. The voice input reception unit 53 is a reception unit that receives a speaker's voice signal from a microphone or the like, performs A / D conversion and amplification processing, and outputs the result to the control unit 56.

  The sound generation unit 54 is a generation unit that generates a hearing-aid sound signal from the hearing-aid spectrum generated by the control unit 56 described later. The audio output unit 55 is an output unit that performs D / A conversion and amplification processing on the audio signal generated by the audio generation unit 54 and outputs the result to a speaker.

  The control unit 56 includes a control program such as an OS, a program that defines processing procedures for various processes, and a memory for storing various data, and is a control unit that executes various processes.

  The control unit 56 includes a spectrum analysis unit 56a, a spectrum fine structure extraction unit 56b, a spectrum envelope extraction unit 56c, a spectrum envelope deformation unit 56d, a spectrum generation unit 56e, a frequency intensity correction amount calculation unit 56f, and a frequency intensity correction unit 56g. .

  Similarly to the spectrum analysis unit 56a described in the first embodiment, the spectrum analysis unit 56a receives a digitized voice signal from the voice input reception unit 53, performs cepstrum analysis, and among the cepstrum coefficients obtained as a result, It is an analysis part which outputs a high quefrency part and a low quefrency part to the spectrum fine structure extraction part 56b and the spectrum envelope extraction part 56c, respectively.

  The spectral fine structure extraction unit 56b is an extraction unit that receives a high quefrency part from the spectral analysis unit 56a and extracts a spectral fine structure by applying a fast Fourier transform. The spectrum envelope extraction unit 56c is an extraction unit that receives a low quefrency unit from the spectrum analysis unit 56a and extracts a spectrum envelope by applying a fast Fourier transform.

  The spectrum envelope deformation unit 56d is a deformation unit that deforms the shape of the spectrum envelope by changing the positions of the peaks and valleys of the extracted spectrum envelope. Specifically, the spectrum envelope deforming unit 56d sets a predetermined inversion axis with respect to the spectrum envelope, and inverts the positions of peaks and valleys around the inversion axis.

  The spectrum generation unit 56e is a generation unit that generates a spectrum of hearing loss by synthesizing the spectrum fine structure extracted by the spectrum fine structure extraction unit 56b and the spectrum envelope deformed by the spectrum envelope deformation unit 56d.

  The frequency intensity correction amount calculation unit 56f is a calculation unit that calculates the correction amount of the spectrum intensity in the spectrum of the hearing aid sound generated by the spectrum generation unit 56e. Specifically, the frequency intensity correction amount calculation unit 56f receives the spectrum information of the speech signals of various speakers from the spectrum analysis unit 56a, and calculates the average value of the received spectrum (the spectrum average of the original sound). .

  Further, the frequency intensity correction amount calculation unit 56f receives information on the spectrum of the hearing loss generated based on the voice signals of various speakers from the spectrum generation unit 56e, and receives the average value of the received spectrum (the spectrum of the hearing loss) Average).

  Thereafter, the frequency intensity correction amount calculation unit 56f calculates an increase in spectrum (a spectrum increase in the hearing-aid sound) by subtracting the spectrum average in the original sound from the spectrum average in the hearing-aid sound. Then, the frequency intensity correction amount calculation unit 56f detects a frequency band in which the increase in the spectrum of the hearing loss is a positive value, and sets the increase in the spectrum of the hearing loss in that frequency band as the correction amount of the spectrum intensity.

  The frequency intensity correction unit 56g corrects the spectrum intensity in a predetermined frequency region of the spectrum of the hearing loss generated by the spectrum generation unit 56e, and outputs the spectrum of the hearing loss whose spectrum intensity is corrected to the sound generation unit 54. Part. Specifically, the frequency intensity correction unit 56g corrects the spectrum intensity of the hearing-aid sound spectrum based on the correction amount information calculated by the frequency intensity correction amount calculation unit 56f.

  Next, a processing procedure of audio processing according to the second embodiment will be described. FIG. 8 is a flowchart of the sound processing procedure according to the second embodiment. As shown in FIG. 8, first, the voice input receiving unit 53 of the voice processing device 50 receives an input of a voice signal from the microphone (step S201).

  Then, the spectrum analysis unit 56a performs spectrum analysis of the speech waveform of the input speech signal, and calculates a high quefrency portion and a low quefrency portion in the cepstrum coefficient (step S202).

  Subsequently, the spectral fine structure extraction unit 56b acquires a high quefrency part from the spectral analysis unit 56a and extracts the spectral fine structure (step S203). And the spectrum envelope extraction part 56c acquires a low quefrency part from the spectrum analysis part 56a, and extracts a spectrum envelope (step S204).

  Thereafter, the spectrum envelope deforming unit 56d deforms the spectrum envelope by changing the positions of the peaks and valleys of the spectrum envelope extracted by the spectrum envelope extracting unit 56c (step S205).

  Then, the spectrum generation unit 56e generates a spectrum of hearing loss that combines the spectrum envelope deformed by the spectrum envelope deformation unit 56d and the spectrum fine structure extracted by the spectrum fine structure extraction unit 56b (step S206).

  Subsequently, the frequency intensity correction unit 56g corrects the spectrum intensity in the preset frequency region of the spectrum of the hearing aid sound generated by the spectrum generation unit 56e (step S207). The procedure for setting the correction amount of the spectral intensity will be described in detail later.

  Then, the sound generation unit 54 generates a hearing-aid sound signal from the hearing-aid spectrum whose spectrum intensity is corrected by the frequency intensity correction unit 56g (step S208). Then, the audio output unit 55 outputs the audio signal of the hearing-proof sound generated by the audio generation unit 54 to the speaker (step S209).

  Thereafter, the control unit 56 checks whether or not an instruction to end the hearing-proof sound output process has been issued (step S210), and if an instruction to end is given (step S210, Yes), the process ends. If no termination instruction has been given (No at step S210), the process proceeds to step S201, and the subsequent processing is repeated.

  Next, a processing procedure for setting the spectral intensity correction amount will be described. FIG. 9 is a flowchart showing a processing procedure for setting the spectral intensity correction amount. As shown in FIG. 9, first, the frequency intensity correction amount calculation unit 56f of the speech processing device 50 receives the spectrum information of the input speech from the spectrum analysis unit 56a and accumulates the information (step S301).

  Further, the frequency intensity correction amount calculation unit 56f acquires the information of the spectrum of the hearing-aid sound generated based on the input sound from the spectrum generation unit 56e, and accumulates the information ( Step S302).

  Thereafter, the frequency intensity correction amount calculating unit 56f calculates the average of the spectrum of the input sound (the average of the spectrum of the original sound) (step S303). Further, the frequency intensity correction amount calculation unit 56f calculates the average of the spectrum of the hearing loss (the spectrum average of the hearing loss) (step S304).

  Then, the frequency intensity correction amount calculation unit 56f calculates an increase in spectrum (a decrease in the spectrum of the hearing loss) obtained by subtracting the spectrum average of the input sound signal from the spectrum average of the hearing loss (step S305).

  Thereafter, the frequency intensity correction amount calculation unit 56f detects a frequency band in which the increase in the spectrum of the hearing loss is a positive value (step S306), and uses the increase in the spectrum of the hearing loss in that frequency band as the correction amount of the spectrum intensity. Setting is performed (step S307).

  In the second embodiment, the frequency intensity correction unit 56g automatically corrects the spectrum intensity of the spectrum of the hearing-aid sound by a preset correction amount. However, the frequency intensity correction unit 56g has the input unit 51. It is also possible to accept the correction amount input by the user via the, and to correct the spectrum intensity of the hearing-aid sound spectrum by the correction amount.

  In this case, the frequency intensity correction unit 56g outputs the information on the increase in the spectrum of the hearing loss sound as shown in FIG. 6 to the display unit 52, and also accepts the specification of the correction amount of the spectrum intensity for each frequency region from the user. The intensity correction acceptance screen is output to the display unit 52.

  FIG. 10 is a diagram illustrating an example of the spectrum intensity correction reception screen 60 output to the display unit 52. In order to prevent the hearing loss from becoming a high pitched sound, the user refers to the information on the increase in the spectrum of the hearing loss, operates the mouse for each frequency band to move the slider that adjusts the increase / decrease of the spectrum intensity, Determine the amount of intensity correction.

  Then, the frequency intensity correction unit 56g receives the information on the correction amount of the spectrum intensity, and corrects the spectrum intensity of the spectrum of the hearing loss based on the received information on the correction amount.

  As described above, according to the second embodiment, the spectral fine structure extracting unit 56b extracts the spectral fine structure from the speech signal of the speaker, and the spectrum generating unit 56e uses the extracted spectral fine structure and the predetermined spectrum. By synthesizing the envelope, the spectrum of the hearing loss sound that is output over the speaker's voice is generated, and the frequency intensity correction unit 56g suppresses the spectrum intensity in a predetermined frequency region of the generated spectrum. Since the spectrum is corrected, it is possible to effectively prevent the person from feeling uncomfortable by suppressing the spectrum intensity in the frequency region that causes the hearing-proof sound to become a high-pitched sound.

  In addition, according to the second embodiment, the frequency intensity correction amount calculation unit 56f combines the spectrum fine structure extracted from the speech signal of the speaker and the predetermined spectrum envelope, and the spectrum of the hearing loss sound, Since the spectrum intensity correction amount is set based on the difference from the spectrum obtained from the person's voice signal, the spectrum intensity correction amount in the frequency domain that causes the hearing-proof sound to become a high-pitched sound is set appropriately. be able to.

  Although the embodiments of the present invention have been described so far, the present invention can be implemented in various different embodiments within the scope of the technical idea described in the claims other than the embodiments described above. Is also good.

  In addition, among the processes described in the above embodiment, all or part of the processes described as being automatically performed can be manually performed, or the processes described as being manually performed can be performed. All or a part can be automatically performed by a known method. In addition, the processing procedure, control procedure, specific name, and information including various data and parameters shown in the above-described document and drawings can be arbitrarily changed unless otherwise specified.

  Each component of each illustrated device is functionally conceptual and does not necessarily need to be physically configured as illustrated. In other words, the specific form of distribution / integration of each device is not limited to that shown in the figure, and all or a part thereof may be functionally or physically distributed or arbitrarily distributed in arbitrary units according to various loads or usage conditions. Can be integrated and configured. Further, all or any part of each processing function performed in each device may be realized by a CPU and a program analyzed and executed by the CPU, or may be realized as hardware by wired logic.

  The voice processing method described in the above embodiment can be realized by executing a program prepared in advance on a computer such as a personal computer or a workstation. This program can be distributed via a network such as the Internet. The program can also be executed by being recorded on a computer-readable recording medium such as a hard disk, a flexible disk (FD), a CD-ROM, an MO, and a DVD, and being read from the recording medium by the computer.

  As described above, the sound processing device and the sound processing method according to the present invention effectively prevent the sound output over the speaker's voice from becoming a high-pitched sound, and make the person uncomfortable. This is useful for speech processing systems that need to protect the privacy of the speaker without the need.

It is a figure explaining the concept of the audio | voice process which concerns on Example 1. FIG. It is a figure which shows the comparison with the spectrogram of the hearing-aid sound in Example 1, and the spectrogram of the conventional hearing-aid sound. 1 is a diagram illustrating a functional configuration of a voice processing device 30 according to a first embodiment. 3 is a flowchart illustrating a processing procedure of audio processing according to the first embodiment. It is a figure explaining the audio | voice process which concerns on Example 2. FIG. It is a figure explaining the calculation method of the spectrum intensity correction amount. It is a figure which shows the function structure of the audio | voice processing apparatus 50 which concerns on Example 2. FIG. 10 is a flowchart illustrating a processing procedure of audio processing according to the second embodiment. It is a flowchart which shows the process sequence of the setting process of a spectrum intensity correction amount. It is a figure which shows an example of the spectrum intensity change reception screen 60 output to the display part 52. FIG.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 Speech waveform 11 Spectrogram 12 Short-time spectrum 13 Spectrum envelope 14 Spectrum fine structure 15 Spectrum envelope database 16 Replaced spectrum envelope 17 Spectrum of hearing loss 20 Spectrogram of original speech 21 Spectrogram of conventional hearing loss 21 Spectrogram of conventional hearing loss 22 Spectrogram 30 speech processing device 31 input unit 32 display unit 33 speech input reception unit 34 spectrum envelope database 35 speech generation unit 36 speech output unit 37 control unit 37a spectrum analysis unit 37b spectrum fine structure extraction unit 37c spectrum envelope extraction unit 37d spectrum envelope selection Unit 37e spectrum generation unit 40 spectrum intensity correction amount 50 audio processing device 51 input unit 52 display unit 53 audio input reception unit 54 audio generation unit 55 audio output unit 6 control unit 56a spectral analyzer 56b spectrum fine structure extracting unit 56c spectrum envelope extracting unit 56d spectrum envelope deforming unit 56e spectrum generating unit 56f frequency intensity correction amount calculation unit 56g frequency intensity corrector 60 spectral intensity correction acceptance screen

Claims (8)

A speech processing device that generates a speech spectrum of speech output over a speaker's speech,
A spectral envelope database for storing data relating to a plurality of different spectral envelopes;
Spectral fine structure extraction means for extracting the spectral fine structure from the speech signal of the speaker;
Spectrum envelope selection means for selecting data related to a spectrum envelope from data related to a spectrum envelope stored in the spectrum envelope database;
Voice spectrum generation means for generating a voice spectrum of a voice to be output over a speaker's voice by synthesizing the spectrum envelope selected by the spectrum envelope selection means and the spectrum fine structure. Voice processing device.   The spectrum envelope selection means newly selects data related to a spectrum envelope from the spectrum envelope database when a temporal change amount of a speaker's voice is a predetermined value or more, and the voice spectrum generation means selects the spectrum envelope selection. The speech spectrum of claim 1, wherein a speech spectrum of the speech output over the speech of the speaker is newly generated by synthesizing the spectrum envelope newly selected by the means and the spectrum fine structure. Processing equipment.   The speech processing apparatus according to claim 1 or 2, wherein the spectrum envelope selection unit randomly selects data related to a spectrum envelope from data related to a spectrum envelope stored in the spectrum envelope database.   The apparatus further comprises spectrum envelope extraction means for extracting a spectrum envelope from the speech signal of the speaker, wherein the spectrum envelope selection means includes the spectrum envelope extracted by the spectrum envelope extraction means and a spectrum in which data is stored in the spectrum envelope database. The speech processing according to claim 1 or 2, wherein data related to a spectrum envelope is selected from data related to a spectrum envelope stored in the spectrum envelope database based on a similarity between the envelope and the envelope. apparatus. A speech processing device that generates a speech spectrum of speech output over a speaker's speech,
Spectral fine structure extraction means for extracting the spectral fine structure from the speech signal of the speaker;
A voice spectrum generating means for generating a voice spectrum of a voice to be output over a speaker's voice by synthesizing the spectral fine structure extracted by the spectral fine structure extracting means and a predetermined spectral envelope;
An audio processing apparatus comprising: frequency intensity correcting means for correcting the audio spectrum by suppressing the spectrum intensity in a predetermined frequency region of the audio spectrum generated by the audio spectrum generating means.   The frequency intensity correction means sets a spectrum intensity correction amount based on a difference between a voice spectrum obtained from a voice signal of a speaker and a voice spectrum generated by the voice spectrum generation means. Item 6. The voice processing device according to Item 5. A speech processing method for generating a speech spectrum of speech output over a speaker's speech,
A spectral extraction process for extracting the spectral fine structure from the speech signal of the speaker;
When a spectral fine structure is extracted by the spectral fine structure extraction step, a spectral envelope selection step of selecting data related to a spectral envelope from among a plurality of different spectral envelope data stored in advance in a spectral envelope database;
A speech spectrum generation step of generating a speech spectrum of speech output over a speaker's speech by synthesizing the spectrum envelope selected by the spectrum envelope selection step and the spectrum fine structure. Voice processing method. A speech processing method for generating a speech spectrum of speech output over a speaker's speech,
A spectral fine structure extraction step for extracting the spectral fine structure from the speech signal of the speaker;
A speech spectrum generation step of generating a speech spectrum of a speech output over a speaker's speech by synthesizing the spectrum microstructure extracted by the spectrum microstructure extraction means and a predetermined spectrum envelope;
And a frequency intensity correction step of correcting the audio spectrum by suppressing a spectrum intensity in a predetermined frequency region of the audio spectrum generated by the audio spectrum generation step.

Images