JP2009015209A - Speech articulation improving system and speech articulation improving method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a speech articulation improving system and a speech articulation improving method, which can appropriately perform a loudness compensation by accurately detecting change of a level around mid/highrange by using a uni-directional microphone. <P>SOLUTION: The speech articulation improving system, which controls a gain of a speech signal on the basis of the speech power of the speech signal generated by a speech signal generation section, and power of surrounding sound, includes: the uni-directional microphone for detecting the speech outputted by a loud speaker and the surrounding sound; an estimating section for estimating each of the speech power and the power of the surrounding sound, from a microphone detection signal and the speech signal; and an audibility compensation filter provided between the microphone and the estimation section. A transfer characteristic from the microphone to the audible compensation filter is made approximately to be A characteristic. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to a speech intelligibility improvement system and a speech intelligibility improvement method, and more particularly to a speech intelligibility improvement system and a speech intelligibility improvement method in which a microphone that picks up surrounding noise can be used also as a speaker voice input microphone.

  There is a speech intelligibility improvement system that makes it possible to hear sound output from a speaker clearly even under noise. For example, in an in-vehicle navigation device, sound such as route guidance is output from a speaker to a vehicle interior, but when noise such as engine sound and road noise is high during traveling, it is difficult to hear the speaker output sound due to a masking effect. Therefore, when the noise is louder than the speaker output sound, the speaker output sound is clearly audible even under noise by performing loudness compensation on the speaker output sound and increasing the gain of the entire sound band.

  FIG. 10 is a block diagram showing an example of an in-vehicle device including a conventional speech intelligibility improving system. Reference numeral 1 denotes an in-vehicle navigation device as an audio signal generating means for outputting an audio signal such as route guidance, and reference numeral 2 denotes an audio intelligibility improving system for improving the intelligibility of the audio signal. A guidance voice signal such as a route guidance output from the in-vehicle navigation device 1 is acoustically converted by a speaker 6 through a gain adjustment unit 4 and a power amplifier 5 as a loudness compensation unit for performing loudness compensation of the speech intelligibility improvement system 2. A guidance voice is output in the passenger compartment. Surrounding noise (referred to as ambient sound) N such as guidance voice A, engine sound and road noise is picked up by an omnidirectional microphone 7 having a substantially flat frequency characteristic up to a low frequency and output to an estimation unit 8. The estimation unit 8 estimates the guidance voice power and the ambient sound power from the guidance voice signal and the microphone detection signal. The estimation unit 8 inputs a guidance voice signal to a transmission characteristic simulation filter 9 that simulates a transmission characteristic from the input side of the power amplifier 5 to the output side of the microphone 7 to estimate a guidance voice component, and to the power calculator 10. Guidance voice power is estimated through the guidance voice component. On the other hand, the power of the microphone input is calculated by the power calculator 11 connected to the microphone 7, and the guidance voice power is subtracted from the microphone input power by the adder 12 to estimate the ambient sound power during the microphone input. The loudness compensation control unit 13 determines a gain by which the guidance voice can be heard clearly regardless of the level of the surrounding sound based on the guidance voice power and the ambient sound power, and performs gain adjustment for the gain adjustment unit 4.

By the way, the noise level of the vehicle is very high in the low frequency range, and this low frequency range component exists even when idling while the vehicle is stopped. There is not much difference. On the other hand, the noise level in the mid-high range increases while traveling, but this band overlaps with the speaker output sound, so that the speaker output sound is less audible than when idling while the vehicle is stopped.
In the conventional speech intelligibility improvement system 2 described above, the ambient sound component including the low sound range is estimated using the omnidirectional microphone 7 having the frequency characteristic (C characteristic) substantially flat to the low frequency range. However, when an omnidirectional microphone having a flat frequency characteristic is used, the low-frequency level of automobile noise mainly determines the overall noise level. It becomes difficult to catch changes. In other words, the conventional speech intelligibility improving system 2 has a problem in that it cannot correctly control the gain of the speaker output sound by detecting a change in the noise level in the mid-high range.

Apart from this, various systems that require a microphone for speaker voice input, such as voice recognition systems and hands-free telephones, have recently been introduced in vehicles. In these voice recognition systems and hands-free telephones, unidirectional microphones are mainly used. Unidirectional microphones have a characteristic that cuts low-frequency noise that does not overlap very much with voice while improving the S / N ratio in the mid-high range by forming a directivity called cardioid. This is because many of them are effective in improving the SN ratio. On the other hand, in the conventional speech intelligibility improving system, since the surrounding noise level is estimated including the low frequency range (C characteristic) as described above, the low frequency range noise that does not overlap with the voice is cut off. It is difficult to adopt a microphone with special characteristics. If the microphone of the speech intelligibility improvement system can also be used as the microphone of the speaker speech input system, it is effective to simplify the overall system configuration. There is a problem that a microphone cannot be shared with a free telephone.
As a conventional technique, there is a technique of narrowing the pass frequency band of a microphone input in order to prevent malfunction due to ambient noise when the microphone sensitivity is increased (Patent Document 1). However, this prior art does not solve the above problem.
JP 10-1160999 A

In view of the above, an object of the present invention is to make it possible to share the microphone of the speech intelligibility improving system and the microphone for inputting the speaker voice.
Another object of the present invention is to detect a change in the noise level in the mid-high range so that the gain of the speaker output sound can be correctly controlled.
Another object of the present invention is to provide a speech intelligibility improvement system and a speech intelligibility improvement method that can appropriately perform loudness compensation by accurately detecting changes in the level of ambient sounds in the middle and high frequencies using a speaker input microphone. Is to provide.

-Voice clarity improvement system The first aspect of the present invention is a voice clarity improvement system that controls the gain of a voice signal based on the voice power of the voice signal generated by the voice signal generation unit and the power of surrounding sounds. This speech intelligibility improving system is a unidirectional microphone that detects sound and ambient sound output from a speaker based on the sound signal, and obtains sound power and peripheral sound power from the microphone detection signal and the sound signal. An estimation unit for estimation, and an auditory correction filter provided between the microphone and the estimation unit are provided, and a transfer characteristic from the microphone to the auditory correction filter is set to a substantially A characteristic.
In the above speech intelligibility improvement system, the output of the A characteristic filter when the A characteristic filter is connected to the omnidirectional microphone is represented in the frequency domain as Y (ω), and is placed close to the omnidirectional microphone. When the output of the unidirectional microphone in the frequency domain is X (ω), the transfer characteristic H (ω) of the audibility correction filter is
H (ω) = Y (ω) / X (ω)
To be determined.
-Voice intelligibility improving method The second of the present invention is that the audio signal generated by the audio signal generation unit is acoustically converted by a speaker, the audio and surrounding sounds output from the speaker are detected by a microphone, and the microphone detection signal and the This is a speech intelligibility improving method that estimates speech power and ambient sound power from a speech signal and controls the gain of the speech signal based on the estimated speech power and ambient sound power. In this speech intelligibility improving method, the microphone is unidirectional, and an auditory correction filter is provided between the microphone and the estimation unit for estimating the voice power and the ambient sound power, and the auditory correction filter is provided from the microphone. So that the transmission characteristic up to approximately A characteristic is obtained, and the estimation unit estimates the sound power and the ambient sound power from the microphone detection signal and the sound signal input via the audibility correction filter.
In the above speech intelligibility improving method, the output of the A characteristic filter when the A characteristic filter is connected to the omnidirectional microphone is expressed in the frequency domain as Y (ω), and is set close to the omnidirectional microphone. When the output of the unidirectional microphone in the frequency domain is X (ω), the transfer characteristic H (ω) of the audibility correction filter is
H (ω) = Y (ω) / X (ω)
To be determined.
In the speech intelligibility improving method, an A characteristic filter is connected to an omnidirectional microphone, an adaptive filter is connected to a unidirectional microphone, and a first microphone and a second microphone installed so as to be close to each other are predetermined. Learning the transfer characteristic H (ω) of the audibility correction filter by updating the filter coefficient of the adaptive filter so that the error between the output of the A characteristic filter and the output of the adaptive filter is minimized while the sound wave is incident. The transfer characteristic of the later adaptive filter is used.

According to the present invention, since the transfer characteristic from the microphone to the auditory correction filter is close to the A characteristic based on the human auditory characteristic, the change in the ambient sound level in the middle and high frequencies that strongly affects the ease of hearing of the sound is accurately captured. Therefore, voice clarity improvement control can be performed, and navigation guidance voice and other voices can be easily heard.
Further, according to the present invention, since the microphone of the speech intelligibility improving system can be used also as a speaker input microphone such as a voice recognition, a hands-free telephone, etc., the overall configuration combined with other voice input devices using the speaker input microphone Can be simplified.

(A) Outline Unlike the conventional speech intelligibility improvement system, the present invention does not estimate the surrounding noise level (including the C characteristic) including the low frequency range, but rather influences the ease of hearing of the mid-high frequency range. Mainly observe the noise level. This can be realized by observing the level of a noise or voice signal through a certain high-pass filter. Therefore, the ambient noise and voice levels are estimated through the A-characteristic, which has a characteristic as a high-pass filter and is devised based on a sense of volume felt by humans. Since basic research on loudness compensation based on noise (ambient sound) and speech level estimation values through A-characteristics has already been made as shown in the following document, actual control should be based on this research result. .
[1] Y. Suzuki et al., J. Acoust. Soc. Jpn. (E), 3 (2), pp55-65, 1982
[2] Y. Suzuki et al., J. Acoust. Soc. Jpn. (E), 6 (3), pp161-170, 1985
Next, unidirectional microphones, which are mainly used for voice recognition and hands-free telephones, have characteristics (high-pass filter characteristics) that cut low-frequency noise that does not overlap with speech. Many. Therefore, if a unidirectional microphone is used and appropriate filtering is performed on the output side so that the overall characteristic becomes the A characteristic, this microphone system can also be used as a speech intelligibility improvement system. It becomes possible. Of course, the output of the unidirectional microphone may be used without filtering in voice recognition, hands-free telephone, or the like. In addition, since the unidirectional microphone itself is a high-pass filter, it is connected after the unidirectional microphone, compared to the case where a microphone having a flat characteristic up to the low frequency range and an A-characteristic filter are used. The size of the filter to be performed can be made smaller.
In summary, in the present invention,
(1) Medium-high, which has characteristics as a high-pass filter and has a strong influence on “easy to hear” by estimating ambient noise and sound levels through A-weighting designed based on the sense of volume perceived by humans. Make it easier to detect changes in the noise level of the area.
(2) Use a unidirectional microphone and put an appropriate filter process on the output side so that the overall characteristics become A characteristics, so that the microphone is shared with voice recognition and hands-free telephones. In addition, the size of the filter connected after the unidirectional microphone is smaller than the “method of using a microphone having a flat characteristic up to the low frequency range and an A characteristic filter as in the past”. To do.

(B) Embodiment Next, an embodiment of the present invention will be described with reference to FIG. FIG. 1 is a block diagram showing a configuration of an in-vehicle device including a speech intelligibility improving system according to the present invention, and the same components as those in FIG. 10 are denoted by the same reference numerals.
In FIG. 1, 2A is a speech intelligibility improving system, 7A is a guidance voice A output from a speaker, a microphone for detecting noise N such as engine sound and road noise, and is also used as a microphone for inputting a speaker's voice. Unidirectional. The unidirectional microphone for mass production improves the S / N ratio in the mid-high range by forming the directivity called cardioid, and cuts out the low-frequency noise that does not overlap with the voice because of its structure. Has characteristics (see FIG. 2). Reference numeral 20 denotes a voice recognition device which performs voice recognition based on microphone input (speaker input voice) from the microphone 7A and performs voice control of an audio system, a navigation system, and the like. Reference numeral 21 denotes an auditory correction filter provided between the microphone 7A and the estimation unit 8A. Here, as shown in FIG. 3, the weighting coefficients s1, a11, a21, b01, b11, and b21 are fixed ones of the secondary IIR filters. It has a stage configuration. The transfer characteristic (frequency-amplitude characteristic) of the audibility correction filter 21 is a medium / high frequency band that has a strong influence on the ease of hearing of the sound if the combination of the microphone 7A and the audibility correction filter 21 is determined so as to exhibit approximately A characteristics. It is possible to accurately grasp changes in the ambient sound level of the.

  The estimation unit 8A estimates the guidance voice power and the ambient sound power from the guidance voice signal and the output signal of the audibility correction filter 21. That is, the estimation unit 8A obtains the guidance voice component by inputting the guidance voice signal to the transfer characteristic simulation filter 8A that simulates the transfer characteristic from the input side of the power amplifier 5 to the output side of the audibility correction filter 21, and calculates the power. The guidance voice component is input to the device 10 to estimate the guidance voice power. On the other hand, the power calculator 11 connected to the audibility correction filter 21 calculates the power of the microphone input signal, and the adder 12 subtracts the guidance voice power from the microphone input power to estimate the ambient sound power. The loudness compensation control unit 13 determines a gain by which the guidance voice can be heard clearly regardless of the level of the surrounding sound based on the guidance voice power and the ambient sound power, and performs gain adjustment for the gain adjustment unit 4. At this time, since the perceived sound power has been corrected for hearing, it is possible to accurately capture changes in the mid-high range sound level, which strongly influences the ease of hearing of the sound. When increased, the loudness compensation can be performed properly.

(C) Transfer characteristic setting method of auditory correction filter 21 In the speech intelligibility improvement system, when an omnidirectional microphone having a characteristic (see FIG. 4) that is substantially flat up to a low frequency is used, as shown in FIG. A directional microphone MIC is connected in series with an A characteristic filter AFIL having an A characteristic (see FIG. 6) in consideration of the auditory characteristic of human volume perceived volume, and the output of the A characteristic filter AFIL is input to the estimation unit 8A. . In this way, it is possible to accurately grasp the change in the surrounding sound level in the middle and high range, which strongly influences the ease of hearing of the sound. Note that, when the A characteristic filter AFIL is approximated by an IIR filter, the attenuation in the low band is rapidly increased, and therefore a two-stage configuration of a secondary IIR filter is required as shown in FIG.
Therefore, the transmission characteristic of the audibility correction filter 21 (the characteristic of the combination of the microphone 7A and the audibility correction filter 21 is substantially the same as the transmission characteristic from the omnidirectional microphone MIC to the A characteristic filter AFIL in FIG. If the frequency-amplitude characteristics are set, the combination of the microphone 7A and the audibility correction filter 21 can accurately capture the change in the mid-high range surrounding sound level that strongly affects the ease of hearing of the sound.

(A) First Determination Method of Filter Characteristics of Auditory Correction Filter FIG. 8 is an explanatory diagram of a first determination method of filter characteristics of the auditory correction filter 21.
An omnidirectional microphone MIC and a unidirectional microphone 7A are arranged close to each other toward the speaker SP, and an A characteristic filter 22 is connected to the output side of the microphone MIC. When the white noise (WN) radiated from the speaker SP is detected by the microphone MIC and 7A, the output y of the A characteristic filter AFIL and the output x of the microphone 7A are frequency-analyzed by the FFT analyzer ANL _FFT , and the resulting microphone MIC When the output of the system in which the A characteristic filter AFIL is connected to Y (ω) and the output of the microphone 7A is X (ω), the following expression H (ω) = Y (ω) / X (ω)
A transfer function H (ω) that satisfies the above is obtained. The frequency-amplitude characteristic of the auditory sensation correction filter 21 in FIG. 1 is matched or substantially matched with the frequency-amplitude characteristic of the transfer function H (ω). Specifically, the frequency-amplitude characteristic of the transfer function H (ω) is simplified to determine the frequency-amplitude characteristic of an analog filter (for example, LPF), and the one-stage configuration shown in FIG. 3 is obtained by a technique such as sz conversion. The secondary IIR weighting coefficients s1, a11, a21, b01, b11, b21 are determined.

(B) Second Determination Method of Filter Characteristics of Auditory Correction Filter FIG. 9 is an explanatory diagram of a second determination method of filter characteristics of the auditory correction filter 21.
An omnidirectional microphone MIC and a unidirectional microphone 7A are arranged close to each other toward the speaker SP, an A characteristic filter AFIL is connected to the output side of the microphone MIC, and an adaptive filter ADF is connected to the output side of the microphone 7A To do. While the white noise WN is radiated from the speaker SP, the adder ADD obtains an error e between the output y of the A characteristic filter AFIL and the output x of the adaptive filter ADF, and the coefficient updating unit CRNW obtains the square value of the error e. The learning is performed by updating the coefficient of the adaptive filter ADF by the LMS algorithm or the like so as to minimize. The frequency-amplitude characteristic of the adaptive filter ADF after learning is simplified to determine the frequency-amplitude characteristic of an analog filter (for example, LPF), and the second-order IIR having the one-stage configuration shown in FIG. Each weighting coefficient s1, a11, a21, b01, b11, b21 is determined.

As described above, according to the present invention, the transfer characteristic from the unidirectional microphone 7A also serving as a speaker input to the audibility correction filter 21 is close to the A characteristic imitating the human auditory characteristic. This makes it possible to properly compensate for the loudness compensation by accurately grasping the change in the surrounding sound level in the mid-high range that strongly influences. Moreover, since the microphone 7A of the speech intelligibility improving system 2A can be used also as the speaker voice input microphone of the speech recognition device 20, the microphone configuration is simplified.
Furthermore, since the low-frequency attenuation characteristic of the A-characteristic filter is steep, when implementing the A-characteristic filter with an IIR filter, it is necessary to have a two-stage IIR filter, but the unidirectional microphone 7A has a low-frequency range. Since it has an attenuation characteristic (see FIG. 2), the low frequency attenuation characteristic of the audibility correction filter 21 may be more gradual than that of the A characteristic filter, and can be realized by a one-stage configuration of a secondary IIR filter (see FIG. 3).
In the embodiment described above, the auditory sensation correction filter has a secondary IIR one-stage configuration. However, the present invention is not limited to this, and the secondary IIR has a two-stage configuration or a third-order or higher order. It is good also as a structure of 1 step | paragraph or more of IIR. Moreover, you may comprise with a FIR filter.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram illustrating a configuration of an in-vehicle device including a speech intelligibility improvement system according to the present invention (Example 1). It is a diagram which shows the frequency characteristic of a unidirectional microphone. It is a block diagram which shows the structural example of the auditory sense correction filter in FIG. It is a diagram which shows the frequency characteristic of an omnidirectional microphone. This configuration uses an omnidirectional microphone to accurately capture changes in the ambient sound level in the middle and high range, which strongly influences the ease of hearing. It is a diagram which shows the frequency characteristic of A characteristic filter. It is a block diagram which shows the structural example of an A characteristic filter. It is explanatory drawing of the 1st determination method of the characteristic of a hearing correction filter. It is explanatory drawing of the 2nd determination method of the characteristic of an auditory sensation correction filter. It is a block diagram which shows the structure of the vehicle-mounted apparatus containing the conventional speech clarity improvement system.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Car navigation system 2A Voice clarity improvement system 4 Gain adjustment part 6 Speaker 7A Unidirectional microphone 8A Estimation part 13 Loudness compensation control part

Claims (5)

In the speech intelligibility improvement system that controls the gain of the audio signal based on the audio power of the audio signal generated by the audio signal generation unit and the power of the surrounding sound,
A unidirectional microphone for detecting sound output from a speaker and ambient sound based on the sound signal;
An estimation unit that estimates the sound power and the power of surrounding sounds from the microphone detection signal and the sound signal;
Hearing correction filter provided between the microphone and the estimation unit,
With
A speech intelligibility improvement system characterized in that a transfer characteristic from a microphone to the audibility correction filter is substantially A-characteristic. Y (ω) represents the output of the A characteristic filter in the frequency domain when the A characteristic filter is connected to the omnidirectional microphone, and the output of the unidirectional microphone installed close to the omnidirectional microphone. When what is expressed in the frequency domain is X (ω), the transfer characteristic H (ω) of the hearing correction filter is
H (ω) = Y (ω) / X (ω)
Decide to be,
The speech intelligibility improvement system according to claim 1. The sound signal generated by the sound signal generation unit is acoustically converted by the speaker, the sound output from the speaker and the surrounding sound are detected by the microphone, and the sound power and the surrounding sound power are estimated from the microphone detection signal and the sound signal. In the speech intelligibility improving method for controlling the gain of the audio signal based on the estimated audio power and the ambient audio power,
The microphone is unidirectional,
An audibility correction filter is provided between the microphone and the estimation unit for estimating the sound power and the ambient sound power, and a transfer characteristic from the microphone to the audibility correction filter is substantially A characteristic,
Estimating audio power and ambient sound power from the microphone detection signal and the audio signal input through the auditory correction filter in the estimation unit,
A speech intelligibility improvement method characterized by the above. Y (ω) represents the output of the A characteristic filter in the frequency domain when the A characteristic filter is connected to the omnidirectional microphone, and the output of the unidirectional microphone installed close to the omnidirectional microphone. When what is expressed in the frequency domain is X (ω), the transfer characteristic H (ω) of the hearing correction filter is
H (ω) = Y (ω) / X (ω)
Decide to be,
The method of improving speech intelligibility according to claim 3. An A characteristic filter is connected to an omnidirectional microphone, an adaptive filter is connected to a unidirectional microphone, and a predetermined sound wave is incident on a first microphone and a second microphone installed so as to be close to each other. Learning is performed by updating the filter coefficient of the adaptive filter so that the error between the output of the adaptive filter and the output of the adaptive filter is minimized, and the transfer characteristic H (ω) of the audibility correction filter is changed to the transfer characteristic of the adaptive filter after learning. To
The method of improving speech intelligibility according to claim 3.

Images