JP2006098534A - Device and method for speech processing - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a device and a method for speech processing that can surely extract only a speech that a person utters, using a simple constitution. <P>SOLUTION: The speech processor which processes a speech of dialog etc., between the person and a device is equipped with a sound-collecting means 10, a specific frequency passing means 20 of passing a specific frequency region of the speech collected by the sound-collecting means 10, a speech processing means 22 of performing speech processing, based on the speech passed through the specific frequency passing means 20, and a sound generating means 11 of generating sound based on the signal from a sound generation controller 22, wherein the specific frequency region of the specific frequency passing means 20 is set to a frequency region different from the frequency region of the sound generated by the sound generating means 11. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a voice processing apparatus and a voice processing method for performing voice processing such as dialogue between a person and a device.

  2. Description of the Related Art In recent years, systems have been developed that perform various processes while interacting with a user in setting a destination in car navigation, communicating with a personal robot, and the like. In such a system, a voice input from a microphone is recognized, a conversation content configured according to the recognized voice is output from a speaker, and a dialogue with the user is performed.

  When performing speech recognition in such an interactive format, the sound output from the system is also input to the microphone together with the sound emitted by the user. For this reason, if the system and the user utter the voice at the same time, the accuracy of the voice recognition is lowered, and the user's voice cannot be accurately recognized. Therefore, wait for the audio output in the system to end, the user will start speaking, or before the user starts speaking, press the switch to interrupt the system's audio output, so that the system voice and the user's voice do not overlap There was a need. In addition, another device such as a headset is required so that sound from the speaker is not input to the microphone.

Patent Document 1 discloses an input voice extraction device that extracts only a voice uttered by a user even when a voice uttered by the user and a sound signal output from the apparatus overlap. In this input voice extraction device, only a user's voice is extracted by subtracting a sound signal prepared in advance on the device side from a voice signal input to a microphone.
JP 10-307595 A

  However, in the case of the input voice extraction device described above, if the frequency component overlaps between the user's voice and the sound signal output from the device, only the user's voice can be extracted for the overlapping frequency component. Since this is not possible, the speech recognition accuracy is reduced. In order to avoid this, when the frequency component of the sound signal output from the device is limited in accordance with the frequency characteristics of the user's voice, the frequency component cannot be partially used, so that the output voice is uncomfortable. In addition, in the input voice extraction device, an algorithmic process for subtracting a sound signal prepared in advance on the device side from a voice signal input to the microphone is necessary, so the process becomes complicated. Furthermore, the input speech extraction device cannot perform processing unless the sound signal to be output is known in advance.

  Therefore, an object of the present invention is to provide a voice processing device and a voice processing method that can reliably extract only a voice uttered by a person with a simple configuration.

  The sound processing apparatus according to the present invention performs sound processing based on the sound collecting means, the specific frequency passing means that passes the specific frequency region of the sound collected by the sound collecting means, and the sound that has passed through the specific frequency passing means. Sound processing means and sound generation means for generating sound based on a signal from the sound generation control device, and the specific frequency region in the specific frequency passing means is different from the frequency region of the sound generated from the sound generation device It is set in the frequency domain.

  This voice processing device performs various voice processing such as voice recognition on voice uttered by a person and generates various sounds such as conversational voice for the person. For this purpose, the sound processing device collects the sound produced by the person by the sound collecting means and generates sound based on the signal from the sound generation control means from the sound generating means. When collecting sound, the sound generated from the sound generating means may be collected. In the sound processing device, the specific frequency passing means passes only the specific frequency region different from the frequency region of the sound generated from the sound generating means in the collected sound. Therefore, when the sound generated by the person and the sound generated by the sound generating means are simultaneously generated, the sound collected by the sound collecting means includes both frequency components of the sound generated by the person and the sound generated by the sound generating means. The sound that has passed through the passage means does not include sound generated by the sound generation means. That is, only the frequency component of the voice uttered by a person is extracted. Therefore, in the voice processing apparatus, the voice processing unit performs voice processing such as voice recognition based on the voice that has passed through the specific frequency passing unit. At this time, since the sound processing is performed only on the frequency component of the sound uttered by a person, the processing accuracy is improved. In particular, when performing speech recognition as speech processing, speech recognition accuracy is improved and human speech can be accurately recognized. As described above, according to this sound processing device, the human speech is generated with a simple configuration in which only the frequency region different from the frequency region of the sound generated from the sound generating unit is allowed to pass from the sound collected by the sound collecting unit. It is possible to reliably extract only the voice that has been played. As a result, even when a person utters a sound while generating sound from the sound generating means, it is possible to perform processing on the sound uttered by the person as if the person uttered the sound alone. .

  In the audio processing apparatus of the present invention, the second specific frequency passing means for passing the second specific frequency region of the sound based on the signal from the sound generation control apparatus is provided between the sound generation control apparatus and the sound generation means. The specific frequency region in the specific frequency passing means and the second specific frequency region in the second specific frequency passing means may be set to different frequency regions.

  In this sound processing apparatus, only the second specific frequency region different from the specific frequency region in the specific frequency passing means among the sound based on the signal from the sound generation control device is passed by the second specific frequency passing means. In the sound processing apparatus, only the sound that has passed through the second specific frequency passing means is generated by the sound generating means. The frequency component of the sound generated from the sound generating means includes only the second specific frequency region, but the frequency component of the sound emitted by a person includes all frequency regions from low frequency to high frequency. Therefore, when the sound from the person and the sound from the sound generation means are generated simultaneously, the sound from the person and the sound from the sound generation means overlap in the second specific frequency region, but the frequency is different from that of the second specific frequency region. In a region (including a specific frequency region), only human speech exists. For this reason, since the sound that has passed through the specific frequency passing means does not include the sound generated by the sound generating means, it is possible to extract only the sound emitted by a person more reliably.

  In the audio processing apparatus of the present invention, it is preferable that the specific frequency region in the specific frequency passing means and the second specific frequency region in the second specific frequency passing means are alternately set.

  In this sound processing device, the specific frequency region for allowing the collected sound to pass through and the second specific frequency region for allowing the sound based on the signal from the sound generation control device to pass alternately are set to change from a low frequency to a high frequency. The specific frequency region and the second specific frequency region can be set over a wide frequency region. That is, the specific frequency region and the second specific frequency region are not arranged in a predetermined frequency band, but can be arranged in a wide frequency region. Therefore, since it is possible to perform speech processing on a wide frequency range as frequency components of speech uttered by a person, the processing accuracy is improved (when speech recognition is performed as speech processing, the accuracy of speech recognition is improved. ). On the other hand, since the frequency component of the sound generated from the sound generating means is also composed of components in a wide frequency range, it is possible to generate a wide variety of sounds (when generating sound, it is possible to generate sound with no sense of incongruity). ).

  The sound processing apparatus of the present invention includes person specifying means for specifying a person who has emitted the sound collected by the sound collecting means, and the specific frequency region in the specific frequency passing means is the sound of the person specified by the person specifying means. It is good also as a structure changed according to a characteristic.

  In this sound processing device, the person who has uttered the sound is specified by the person specifying means. In the sound processing device, the specific frequency region to be passed by the specific frequency passing means is set according to the characteristics of the sound of the specified person. In other words, the frequency characteristics of uttered voices vary from person to person due to individual differences (for example, the frequency characteristics are clearly different between people with high frequency and those with low frequency). When the voice processing is performed, the processing is performed mainly on the characteristic frequency region of the person, and the voice processing accuracy is improved. Since the sound in the specific frequency region is used instead of the entire frequency region, the amount of information when performing the sound processing is reduced, but since the characteristic frequency region is processed, the sound processing accuracy (particularly, the speech recognition accuracy) ) Does not decrease.

  In the voice processing apparatus according to the present invention, the person specifying means may be a feature detecting means for detecting a feature of the sound collected by the sound collecting means.

  In this sound processing apparatus, the person who is emitting the sound is specified by detecting the characteristics of the sound collected by the sound collecting means. The feature may be detected directly from the sound collected by the sound collecting means, or the feature may be detected from the sound component that has passed through the specific frequency region of the collected sound by the specific frequency passing means. Also good.

  In the sound processing apparatus of the present invention, it is preferable to switch between the specific frequency region in the specific frequency passing means and the second specific frequency region in the second specific frequency passing means according to the passage of time.

  In this sound processing device, a specific frequency region for allowing the collected sound to pass through and a second specific frequency region for allowing the sound based on the signal from the sound generation control device to pass are switched over time. That is, when a certain frequency region is A and a frequency region different from A is B, the specific frequency region is A at a certain time and the second specific frequency region is B, and the specific frequency region is the next time. And B, the second specific frequency region is A, and at the next time, the specific frequency region is A and the second specific frequency region is B. When this switching time interval is set to a very short time, the frequency components thinned out at a certain time are supplemented after a very short time, and all frequency components can be obtained in a very short time. Therefore, the sound generated from the sound generation means becomes a sound including all frequency components of the sound based on the signal from the sound generation control device, and outputs a sound (particularly a sound) that is heard by a person and does not feel uncomfortable. Can do. Further, the sound processed by the sound processing means becomes a sound including all frequency components of the sound uttered by a person, and the sound processing accuracy (particularly, the sound recognition accuracy) is improved.

  The sound processing method according to the present invention performs sound processing based on a sound collecting step, a specific frequency passing step that passes a specific frequency region of the sound collected in the sound collecting step, and a sound that has passed the specific frequency passing step. Including a sound processing step and a sound generation step for generating sound based on a signal from the sound generation control device, wherein the specific frequency region in the specific frequency passing step is different from the frequency region of the sound generated in the sound generation step It is set to an area.

  The audio processing method of the present invention includes a second specific frequency passing step for passing a second specific frequency region of sound based on a signal from the sound generation control device, and the specific frequency region and the second specific frequency in the specific frequency passing step. It is good also as a structure set to the frequency area | region different from the 2nd specific frequency area | region in a frequency passage step.

  In the audio processing method of the present invention, it is preferable that the specific frequency region in the specific frequency passing step and the second specific frequency region in the second specific frequency passing step are alternately set.

  The sound processing method of the present invention includes a person specifying step for specifying a person who has emitted the sound collected in the sound collecting step, and the specific frequency region in the specific frequency passing step is the voice of the person specified in the person specifying step. It is good also as a structure changed according to a characteristic.

  In the voice processing method of the present invention, the person specifying step may be configured to detect a feature of the voice collected in the sound collecting step.

  In the audio processing method of the present invention, it is preferable to switch between the specific frequency region in the specific frequency passing step and the second specific frequency region in the second specific frequency passing step according to the passage of time.

  Each voice processing method described above has the same operations and effects as each voice processing apparatus described above.

  According to the present invention, a person uttered by a simple configuration in which only a frequency region different from the frequency region of the sound output from the collected sound is passed and the sound processing is performed based on the passed sound. Only voice can be reliably extracted.

  Hereinafter, embodiments of a sound processing apparatus and a sound processing method according to the present invention will be described with reference to the drawings.

  In the present embodiment, the present invention is applied to a voice processing device in a car navigation system capable of interactive communication with a user in destination setting or the like. The voice processing apparatus according to the present embodiment recognizes voice collected by a microphone, outputs voice of conversation content configured according to the recognized voice from a speaker, and performs dialogue with the user. In particular, in the speech processing apparatus according to the present embodiment, speech recognition is performed using speech components in the first specific frequency region (corresponding to the specific frequency region) out of the frequency components of speech collected by a microphone, and the speech content of the conversation content is recorded. Of the frequency components, only the sound component in the second specific frequency region is generated from the speaker. In this embodiment, there are three forms, the first embodiment is a basic configuration of the speech processing apparatus, and the second embodiment is a first specific frequency according to the characteristics of the user's speech. This is a configuration having a function of changing the region, and the third embodiment is a configuration having a function of switching between the first specific frequency region and the second specific frequency region every extremely short time.

  With reference to FIG. 1, the overall configuration of a speech processing apparatus 1 according to the present embodiment will be described. FIG. 1 is an overall configuration diagram of the speech processing apparatus according to the present embodiment.

  The voice processing device 1 is configured as one device in a car navigation system, and is used to interact with a user to set a destination or the like. Incidentally, the voice processing device 1 may be used as voice output means such as guidance voice for route guidance. The audio processing device 1 mainly includes a microphone 10, a speaker 11, and an audio processing unit 12. In the voice processing device 1, the voice emitted by the user is collected by the microphone 10, and the voice emitted by the user is recognized from the collected voice by the voice processing unit 12. Further, in the voice processing device 1, the voice processing unit 12 constitutes the conversation contents generated according to the recognized voice contents, generates the utterance voice of the conversation contents, and outputs the utterance voice from the speaker 11. In the present embodiment, the microphone 10 corresponds to the sound collecting means described in the claims, and the speaker 11 corresponds to the sound generating means described in the claims.

  With reference to FIGS. 2 to 5, the configuration of the speech processing apparatus 1 </ b> A according to the first embodiment will be described. FIG. 2 is a configuration diagram of the speech processing apparatus according to the first embodiment. FIG. 3 is an explanatory diagram of processing in the speech processing apparatus of FIG. 1 when only the user utters speech. FIG. 4 is a diagram illustrating an audio signal collected by a microphone when the audio emitted by the user and the audio output from the audio processing device overlap. FIG. 5 is an explanatory diagram of processing in the voice processing device in FIG. 1 when the voice uttered by the user and the voice output by the voice processing device overlap.

  Audio processing apparatus 1A is a basic configuration of the audio processing apparatus according to the present embodiment. The speech processing apparatus 1A has a speech recognition function, a conversation composition function, an utterance speech generation function, a speech output function, and the like in order to interact with the user. In particular, the voice processing apparatus 1A can reliably extract only the user's voice and reliably recognize the user's voice even when the user's voice and the output voice are emitted simultaneously. For this purpose, the sound processing device 1A includes a microphone 10, a speaker 11, and a sound processing unit 12A. The sound processing unit 12A includes a first filter 20, a second filter 21, and a sound control device 22.

  In the first embodiment, the first filter 20 corresponds to the specific frequency pass means described in the claims, and the second filter 21 corresponds to the second specific frequency pass means described in the claims. The sound control device 22 corresponds to the sound processing means and the sound generation control device described in the claims.

  The microphone 10 collects sound (air sound produced by a person or sound output from the device) made of air vibrations. In the microphone 10, the collected sound is frequency-converted by fast Fourier transform or the like, and an original input sound signal having an electrical intensity for each frequency is transmitted to the first filter 20. An output audio signal is input to the speaker 11 from the second filter 21. Then, the speaker 11 converts the output sound signal composed of the electric intensity for each frequency by inverse fast Fourier transform or the like, and outputs sound corresponding to the output sound signal. In the graphs showing the user's voice and the voice of the device shown in FIG. 3 and the like, the horizontal axis represents frequency and the vertical axis represents intensity, and represents frequency characteristics at a certain moment of each voice. This frequency characteristic changes with the passage of time. If a person is different, the frequency characteristic becomes different, and the frequency characteristic also changes depending on the content to be emitted.

  The first filter 20 and the second filter 21 are a kind of bandpass filter in which a frequency band that passes from low frequency to high frequency and a frequency band that does not pass through are arranged at a constant narrow frequency interval, as shown in FIG. Comb frequency characteristics. The frequency band passed by the first filter 20 and the frequency band passed by the second filter 21 have the same width and are arranged so as not to overlap each other. That is, the first filter 20 and the second filter 21 are arranged such that a frequency band that does not pass the other frequency band is arranged in one frequency band that is passed, and the frequency band that passes is different. The width of the frequency band to be passed and the width of the frequency band not to be passed are determined by experiments, and are set so that the voice can be recognized sufficiently and the output voice is not uncomfortable when the voice is recognized. In the frequency characteristics of each filter shown in FIG. 3 and the like, the horizontal axis is frequency, and the vertical axis is represented by 0 (does not pass the corresponding frequency component) and 1 (passes the corresponding frequency component).

  When the original input audio signal is input from the microphone 10, the first filter 20 allows only the frequency components in the comb-shaped first specific frequency region shown in FIG. 3 to pass. Then, the first filter 20 transmits an input audio signal composed of frequency components in the first specific frequency region to the audio control device 22. On the other hand, when the original output audio signal is input from the audio control device 22, the second filter 21 passes only the frequency components in the comb-shaped second specific frequency region shown in FIG. Then, the second filter 21 transmits an output audio signal composed of frequency components in the second specific frequency region to the speaker 11.

  The voice control device 22 includes a CPU [Central Processing Unit], a ROM [Read Only Memory], a RAM [Random Access Memory], and the like. The voice control device 22 is a device that receives an input voice signal corresponding to the voice uttered by the user and outputs an original output voice signal for outputting an appropriate voice corresponding to the voice uttered by the user. For this purpose, the voice control device 22 stores various programs for realizing each function in the ROM, and the various programs are loaded into the RAM and executed by the CPU.

  When the voice control device 22 receives the input voice signal from the first filter 20, it performs voice recognition processing based on this input voice signal. Since this input sound signal passes through the first specific frequency region of the first filter 20, it consists of frequency components thinned out in a comb shape. For this reason, the voice control device 22 stores dictionary data created by learning in advance using voice signals of each language information thinned out in the first specific frequency region. The voice control device 22 converts the input voice signal into language information based on the dictionary data, and recognizes the user's voice.

  Furthermore, the voice control device 22 composes a conversation according to the recognized language information. For this purpose, the voice control device 22 stores a database in which conversation contents to be output on the device side are associated with specific words included in the words uttered by the user. For example, in the case of destination setting, when the user utters the word “Destination”, the conversation contents “Please tell me the destination address” and “Please enter the destination phone number” are associated. ing.

  Subsequently, the voice control device 22 performs voice synthesis on the constructed conversation and generates an original output voice signal. This original output audio signal includes frequency components of all frequencies from low frequency to high frequency. Instead of performing speech synthesis in the apparatus, an original output speech signal corresponding to the conversation content may be stored in the apparatus in advance.

  Next, an operation in the sound processing apparatus 1A when the user sets a destination will be described. Here, a case where only the user's voice is uttered with reference to FIG. 3 and a case where the user's voice and the apparatus's voice are simultaneously emitted with reference to FIG. 5 will be described.

  First, a case where only the user utters voice will be described. For example, when the user issues “destination setting” to set the destination, the microphone 10 collects the sound. The collected sound is frequency-converted and passed through the first filter 20 as an original input sound signal. The input audio signal that has passed through the first specific frequency region of the first filter 20 is transmitted to the audio control device 22. This input voice signal includes only the frequency component of the voice uttered by the user, and is thinned out in a comb shape as the frequency component.

  The voice control device 22 acquires language information from the input voice signal based on dictionary data and recognizes the voice. The input speech signal has frequency components thinned out from the original input speech signal. Since dictionary data created using the thinned speech signal is used, speech recognition can be sufficiently performed. Further, the voice control device 22 sets conversation contents according to the language information. Here, according to the “destination”, for example, a conversation content “tell me the address of the destination” is set. Subsequently, the voice control device 22 generates an original output voice signal according to the content of the conversation.

  This original output audio signal is passed through the second filter 21. The output audio signal that has passed through the second specific frequency region of the second filter 21 is transmitted to the speaker 11. The speaker 11 converts the output audio signal and outputs it as audio. The output audio signal is a signal in which frequency components are thinned out in a comb shape, but since the thinned intervals are narrow and frequency components are distributed over all frequencies from low frequency to high frequency, when interacting with the user The sound is uncomfortable.

  Next, a case where the user and the sound processing apparatus 1A simultaneously emit sound will be described. As shown in FIG. 4, when the user and the sound processing apparatus 1 </ b> A simultaneously emit sound, the frequency component of the sound by the apparatus overlaps with the frequency component of the sound by the user, and changes from the frequency component of only the user's sound. At this time, since the frequency component output from the sound processing apparatus 1A has a thinned comb shape, the overlapped portion is skipped every certain frequency band.

  For example, if the user starts speaking “Chuo-ku, Tokyo ...” while outputting “Please tell me the address of the destination” from the speech processing apparatus 1A, the frequency components of the two sounds overlap. Into the microphone 10. The microphone 10 collects the overlapped sound and frequency-converts the sound. Then, the original input voice signal of the voice is passed through the first filter 20, and an input voice signal composed of frequency components that have passed through the first specific frequency region is transmitted to the voice control device 22. Since this input audio signal consists only of frequency components in the first specific frequency region, frequency components in the second specific frequency region are not included. That is, since the frequency component of the sound output from the sound processing device 1A (that is, the output sound signal that has passed through the second filter 21) cannot pass through the first filter 20, the output from the sound processing device 1A is not included in the input sound signal. Audio is not included. Therefore, the input voice signal includes only the frequency component of the voice uttered by the user. Therefore, the voice control device 22 can perform voice recognition with the same accuracy as when only the above-mentioned user utters voice. Since the operation after the voice control device 22 is the same as the above, the description thereof is omitted.

  According to the sound processing apparatus 1A, even when the user and the apparatus are simultaneously producing sound, only the sound emitted by the user can be extracted, and thus the user's sound can be recognized with high accuracy. Thereby, a smooth dialogue between the user and the voice processing device 1A can be performed. In order to reliably extract only the voice uttered by the user, the voice processing device 1A places the first filter 20 between the microphone 10 and the voice control device 22, and between the voice control device 22 and the speaker 11. This is a simple configuration in which the second filter 21 is arranged so that the first specific frequency region of the first filter 20 and the second specific frequency region of the second filter 21 do not overlap.

  Furthermore, according to the speech processing apparatus 1A, since the frequency bands to be passed through the specific frequency regions of the filters 20 and 21 are alternately arranged over all frequencies, when performing speech recognition, the recognition accuracy can be improved. In the case of outputting sound, it is possible to output sound with no sense of incongruity.

  With reference to FIG.6 and FIG.7, the structure of the audio | voice processing apparatus 1B which concerns on 2nd Embodiment is demonstrated. FIG. 6 is a configuration diagram of the speech processing apparatus according to the second embodiment. FIG. 7 is an example of the frequency characteristic of the filter according to the characteristics of the user's voice, where (a) is the frequency characteristic of the first filter and (b) is the frequency characteristic of the second filter.

  The sound processing device 1B is substantially the same device as the sound processing device 1A according to the first embodiment, but sets a first specific frequency region and a second specific frequency region according to a user who interacts. For this purpose, the audio processing device 1B includes a microphone 10, a speaker 11, and an audio processing unit 12B. The audio processing unit 12B includes a first variable filter 30, a second variable filter 31, an audio feature detection device 32, a specific frequency. An area database 33, a specific frequency area selection device 34, and a voice control device 35 are included.

  In the second embodiment, the first variable filter 30 corresponds to the specific frequency passing means described in the claims, and the second variable filter 31 is the second specific frequency passing means described in the claims. The voice feature detection device 32 corresponds to the person specifying means and the feature detection device described in the claims, and the voice control device 35 corresponds to the voice processing means and the sound generation control device described in the claims. To do.

  Humans have individual differences in voices to be uttered, and the frequency characteristics of voices vary from person to person. For example, a person with a low voice tends to have a frequency component biased toward a low frequency range, and a person with a high voice tends to have a frequency component biased toward a high frequency range. Therefore, when the speech is recognized, the amount of information increases and the recognition accuracy increases when the recognition processing is performed mainly on the frequency domain component having the features of the speech. In view of this, in the audio processing device 1B, the user is specified from the characteristics of the audio, and the first specific frequency region and the second specific frequency region are set according to the specified user.

  The first variable filter 30 and the second variable filter 31 are filters that can change a frequency band that passes from a low frequency to a high frequency according to the first selection signal and the second selection signal from the specific frequency region selection device 34. It is. The frequency band passed by the first variable filter 30 and the frequency band passed by the second variable filter 31 are arranged so as not to overlap each other, but the width varies depending on the user.

  In the first variable filter 30, when an input sound signal is input from the microphone 10, only the sound component in the first specific frequency region is allowed to pass, and the input sound signal including the first specific frequency region is transmitted to the sound control device 35. . On the other hand, in the second variable filter 31, when the original output audio signal is input from the audio control device 35, only the audio component in the second specific frequency region is allowed to pass, and the output audio signal including the second specific frequency region is transmitted to the speaker 11. Send to. Until the user is specified, the first variable filter 30 and the second variable filter 31 are in the specific frequency regions in the initial state (for example, the first filter 20 and the second filter 21 in the first embodiment). Specified frequency range).

  The voice feature detection device 32 receives an input voice signal that has passed through the first specific frequency region in the initial state of the first variable filter 30, detects a voice feature based on the input voice signal, and specifies a user. For this purpose, the audio feature detection device 32 uses the audio signal thinned out in the first specific frequency region in the initial state from the audio signal of the audio emitted by each user who may use the vehicle installed in the car navigation system. The frequency analysis is performed on the signal in advance, and the voice characteristics (frequency characteristics) of each user based on the frequency analysis are stored. The voice feature detection device 32 compares the frequency characteristics of the voice characteristics of each user stored in advance and the input voice signal, and identifies the user who emitted the voice. When the voice feature detection device 32 specifies a user, the user information is transmitted as a user signal to the specific frequency domain selection device 34 and the voice control device 35.

  The specific frequency region database 33 stores a first specific frequency region and a second specific frequency region corresponding to each user who may use the vehicle mounted on the car navigation system. The first specific frequency region corresponding to each user is a frequency band in which the sound signal of the sound emitted by each user is subjected to frequency analysis, and a wide frequency band is passed through the frequency region having the characteristics of each user's sound based on the analysis result. Is arranged, and a narrow frequency range is arranged in other frequency ranges. For example, in the case of a user who has a characteristic of bass, as shown in FIG. 7A, a wide frequency band is set in the low frequency range, and a narrow frequency band is set in the high frequency range. . On the other hand, in the second specific frequency region, a frequency band that is not allowed to pass in the first specific frequency region is disposed, and a frequency band that is not allowed to pass in the first specific frequency region is disposed. For example, in the case of a user having a characteristic of bass, as shown in FIG. 7B, a narrow frequency band is set in the low frequency range, and a wide frequency band is set in the high frequency range. . The specific frequency region database 33 may be configured to store only the first specific frequency region corresponding to each user.

  By the way, even in the case of a user who is characterized by low frequencies, providing a frequency band that passes not only in the low frequency range but also in the high frequency range is that the user mainly emits the low frequency range, but the high frequency range This is because some of the sound is emitted. This is because the speech recognition accuracy can be improved by setting the first specific frequency region as the speech recognition target from the low frequency region to the high frequency region. In addition, since the second specific frequency region can be used for sound output from a low frequency region to a high frequency region, the output sound does not feel uncomfortable.

  When the specific frequency domain selection device 34 receives the user signal from the voice feature detection device 32, the specific frequency domain selection device 34 refers to the specific frequency domain database 33 using the user indicated by the user signal as a key, and responds to the user from the specific frequency domain database 33. The first specific frequency region and the second specific frequency region are selected. Then, the specific frequency domain selection device 34 transmits a first selection signal indicating the first specific frequency domain (information of a plurality of frequency bands to be passed) to the first variable filter 30, and the second specific frequency domain (multiple frequency domains). The second selection signal indicating the frequency band information to be transmitted is transmitted to the second variable filter 31. When only the first specific frequency region corresponding to each user is stored in the specific frequency region database 33, the specific frequency region selecting device 34 selects the first specific frequency region corresponding to the user, and the first specific frequency region is selected. A second specific frequency region is set based on the specific frequency region.

  The voice control device 35 has substantially the same configuration as the voice control device 22 according to the first embodiment, but part of the processing is different depending on the identified user. The voice control device 35 includes dictionary data created by learning in advance using speech signals of each language information thinned out in the first specific frequency region in the initial state, and thinning out in the first specific frequency region corresponding to each user. Stored is dictionary data created by learning in advance using the voice signal of each language information. The voice control device 35 converts the input voice signal into language information based on the dictionary data in the first specific frequency region in the initial state until the user signal is received, and recognizes the user's voice. In addition, when receiving the user signal, the voice control device 35 converts the input voice signal into language information based on dictionary data corresponding to the user indicated in the user signal, and recognizes the user's voice.

  Furthermore, when receiving the user signal, the voice control device 35 may constitute the conversation content corresponding to the user indicated by the user signal. For this purpose, the voice control device 35 has a request at the time of dialogue set according to the user (for example, dialogue with female voice, dialogue with Osaka dialect), conversation content according to the user (for example, destination) Stores data such as a Japanese restaurant in the case of a restaurant, a Japanese inn if the destination is an accommodation facility, and a designated series of gas stations if the destination is a gas station.

  With reference to FIG. 6, the operation of the sound processing apparatus 1B will be described. Before the dialogue between the user and the sound processing device 1B starts, the first variable frequency band 30 is set with the first specific frequency region in the initial state, and the second variable filter 31 has the second specific frequency region in the initial state. Is set. For example, when the user issues “destination setting” to set the destination, the microphone 10 collects the sound. The collected sound is frequency-converted and passed through the first variable filter 30 as an original input sound signal. The input audio signal that has passed through the first specific frequency region in the initial state of the first variable filter 30 is transmitted to the audio control device 35 and the audio feature detection device 32. At this time, the voice control device 35 performs the same operation as in the first embodiment by using the input voice signal that has passed through the first specific frequency region in the initial state and the dictionary data in the initial state. Then, the original output audio signal transmitted from the audio control device 35 is passed through the second variable filter 31, and the output audio signal that has passed through the second specific frequency region in the initial state of the second variable filter 31 is transmitted to the speaker 11. The The speaker 11 converts the output audio signal and outputs it as audio.

  The voice feature detection device 32 compares the frequency characteristics of the input voice signal that has passed through the first specific frequency region in the initial state with the frequency characteristics of the voice characteristics of each user, and the frequency characteristics are similar. Detect what to do. Then, the voice feature detection device 32 specifies a user from the detection result, and transmits a user signal indicating the user to the specific frequency domain selection device 34 and the voice control device 35.

  The specific frequency region selection device 34 selects the first specific frequency region and the second specific frequency region according to the user specified from the specific frequency region database 33 based on the user signal. Then, the specific frequency region selection device 34 transmits a first selection signal indicating the selected first specific frequency region to the first variable filter 30, and a second variable selection signal indicating the second specific frequency region is second variable. It transmits to the filter 31.

  The first variable filter 30 changes the frequency band to pass from the first specific frequency region in the initial state to the first specific frequency region corresponding to the user based on the first selection signal. The original input audio signal is passed through the first variable filter 30, and the input audio signal that has passed through the first specific frequency region corresponding to the user is transmitted to the audio control device 35. This input audio signal is a signal that is decimated from the original input audio signal, but mainly includes frequency components in the frequency range that are characteristic of the user's audio. At this time, the voice control device 35 performs the same operation as that of the first embodiment by using the input voice signal that has passed through the first specific frequency range corresponding to the user and the dictionary data corresponding to the user. The voice recognition accuracy in the voice control device 35 is improved as compared with the case where the input voice signal that has passed through the first specific frequency region in the initial state is used because the voice recognition is performed based on the frequency component that is a feature of the user.

  In the second variable filter 31, the frequency band to be passed is changed from the second specific frequency region in the initial state to the second specific frequency region corresponding to the user based on the second selection signal. The original output audio signal transmitted from the audio control device 35 is passed through the second variable filter 31, and the output audio signal that has passed through the second specific frequency region corresponding to the user is transmitted to the speaker 11. The speaker 11 converts the output audio signal and outputs it as audio. Since the output sound includes frequency components over the entire number of frequencies from low frequency to high frequency, there is almost no sense of incongruity.

  According to the voice processing device 1B, the same effect as the voice processing device 1A is obtained, and voice recognition is performed in consideration of the characteristics of the user's voice, so that the voice recognition accuracy is further improved.

  With reference to FIGS. 8 to 11, the configuration of a speech processing apparatus 1 </ b> C according to the third embodiment will be described. FIG. 8 is a configuration diagram of a speech processing apparatus according to the third embodiment. FIG. 9 is an explanatory diagram of processing in the sound processing apparatus of FIG. 8 when the first filter is switched to the sound collection side and the second filter is switched to the sound output side. FIG. 10 is an explanatory diagram of processing in the sound processing apparatus of FIG. 8 when the second filter is switched to the sound collection side and the first filter is switched to the sound output side. FIG. 11 is an example of the frequency characteristics of the sound output from the sound processing apparatus of FIG.

  The sound processing device 1C is substantially the same device as the sound processing device 1A according to the first embodiment, but switches between the first filter and the second filter every very short time. For this purpose, the audio processing device 1C includes a microphone 10, a speaker 11, and an audio processing unit 12C. The audio processing unit 12C includes a first filter 40, a second filter 41, a switching timing generator 42, and a first switching control. The apparatus 43, the 2nd switching control apparatus 44, and the audio | voice control apparatus 45 are provided.

  In the third embodiment, the first filter 40 or the second filter 41 corresponds to the specific frequency passing means described in the claims, and the first filter 40 or the second filter 41 is in the claims. The sound control device 45 corresponds to the sound processing device and the sound generation control device described in the claims.

  The first filter 40 and the second filter 41 are the same filters as the first filter 20 and the second filter 21 according to the first embodiment. The first filter 40 has a first switching control device 43 connected to the input side and a second switching control device 44 connected to the output side. On the other hand, in the second filter 41, the second switching control device 44 is connected to the input side, and the first switching control device 43 is connected to the output side.

  The switching timing generation device 42 transmits a switching timing signal to the first switching control device 43, the second switching control device 44, and the voice control device 45 at regular intervals. The switching timing signal is a signal for indicating whether the first filter 40 and the second filter 41 are connected to the sound collection side or the sound output side. The fixed time for transmitting the switching timing signal is set to an extremely short time that is obtained by experiments and does not cause discomfort in the sound output from the sound processing apparatus 1C.

  The first switching control device 43 is disposed between the microphone 10 and the speaker 11 and the first filter 40 and the second filter 41. The microphone 10 and the speaker 11 are connected to one side, and the first filter 40 and the first filter are connected to the other side. Two filters 41 are connected. In the first switching control device 43, when the switching timing signal indicates that the first filter 40 is connected to the sound collection side (and / or the second filter 41 is connected to the sound output side), the first filter 40 is connected to the microphone 10. And the second filter 41 is connected to the speaker 11 (see FIG. 9), and the switching timing signal instructs the first filter 40 to be connected to the audio output side (and / or the second filter 41 to the sound collection side). If so, the second filter 41 is connected to the microphone 10 and the first filter 40 is connected to the speaker 11 (see FIG. 10).

  The second switching control device 44 is disposed between the first filter 40 and the second filter 41 and the sound control device 45, the first filter 40 and the second filter 41 are connected to one side, and the sound control device is connected to the other side. 45 input terminals and output terminals are connected. In the second switching control device 44, when the switching timing signal indicates that the first filter 40 is connected to the sound collection side, the first filter 40 is connected to the input end of the voice control device 45 and the second filter 41 is connected. Is connected to the output terminal of the voice control device 45 (see FIG. 9), and the second filter 41 is input to the voice control device 45 when the switching timing signal indicates that the first filter 40 is connected to the voice output side. The first filter 40 is connected to the output end of the voice control device 45 while being connected to the end (see FIG. 10).

  The voice control device 45 has substantially the same configuration as the voice control device 22 according to the first embodiment, but part of the processing is different when performing voice recognition. Since the first filter 40 and the second filter 41 are switched in the sound processing device 1C, the input sound signal input to the sound control device 45 is a signal obtained by thinning the original input sound signal into a comb shape, but is thinned out. The frequency band changes every certain time. Therefore, it is necessary to switch the dictionary data when converting the input voice signal into the language information. Therefore, the voice control device 45 includes the first dictionary data created by learning in advance using the voice signal of each language information thinned out in the first specific frequency region of the first filter 40 and the second filter 41. 2 Stores second dictionary data created by learning in advance using speech signals of each language information thinned out in a specific frequency region. The voice control device 45 converts the input voice signal into language information based on the first dictionary data when the switch timing signal instructs the first filter 40 to be connected to the sound collection side, and the switch timing signal When it is instructed to connect the first filter 40 to the voice output side, the input voice signal is converted into language information based on the second dictionary data, and the user's voice is recognized.

  With reference to FIGS. 8 to 11, the operation of the sound processing apparatus 1 </ b> C will be described. In the initial state, the first switching control device 43 connects the first filter 40 to the microphone 10 and the second filter 41 to the speaker 11, and the second switching control device 44 connects the first filter 40 to the voice control device 45. And the second filter 41 is connected to the output end of the voice control device 45 (see FIG. 9).

  When the conversation between the user and the voice processing device 1C is started, the switching timing generation device 42 generates a switching timing signal indicating the connection destination of the first filter 40 and the second filter 41 for each extremely short time. The data is transmitted to the switching control device 43, the second switching control device 44, and the voice control device 45.

  When it is instructed to connect the first filter 40 to the sound collection side as a switching timing signal, the first switching control device 43 switches the connection destination of the first filter 40 from the microphone 10 to the speaker 11 and connects the second filter 41. The second switch control device 44 switches the connection destination of the first filter 40 from the input end to the output end of the voice control device 45 and the connection destination of the second filter 41 to the voice control device 45. Are switched from the output terminal to the input terminal (see FIG. 10).

  The original input audio signal collected by the microphone 10 and frequency-converted is transmitted to the first switching control device 43 and passed through the second filter 41 via the first switching control device 43. The input audio signal that has passed through the second specific frequency region of the second filter 41 is transmitted to the audio control device 45 via the second switching control device 44.

  At this time, the voice control device 45 performs voice recognition using the input voice signal that has passed through the second specific frequency region and the second dictionary data. Further, the voice control device 45 performs the same operation as in the first embodiment, and transmits the original output voice signal to the second switching control device 44.

  The original output audio signal transmitted from the audio control device 45 is passed through the first filter 40 via the second switching control device 44. The output audio signal that has passed through the first specific frequency region of the first filter 40 is transmitted to the speaker 11. The speaker 11 converts the output audio signal and outputs it as audio.

  When it is instructed to connect the first filter 40 to the sound collection side as a switching timing signal, the first switching control device 43 switches the connection destination of the first filter 40 from the speaker 11 to the microphone 10 and connects the second filter 41. The second switch control device 44 switches the connection destination of the first filter 40 from the output end to the input end of the voice control device 45 and the connection destination of the second filter 41 to the voice control device 45. Is switched from the input end to the output end (see FIG. 9).

  The original input audio signal collected by the microphone 10 and frequency-converted is transmitted to the first switching control device 43 and passed through the first filter 40 via the first switching control device 43. The input audio signal that has passed through the first specific frequency region of the first filter 40 is transmitted to the audio control device 45 via the second switching control device 44.

  At this time, the voice control device 45 performs voice recognition using the input voice signal that has passed through the first specific frequency region and the first dictionary data. Further, the voice control device 45 performs the same operation as in the first embodiment, and transmits the original output voice signal to the second switching control device 44.

  The original output audio signal transmitted from the audio control device 45 is passed through the second filter 41 via the second switching control device 44. The output audio signal that has passed through the second specific frequency region of the second filter 41 is transmitted to the speaker 11. The speaker 11 converts the output audio signal and outputs it as audio.

  In the sound processing apparatus 1C, the above operation is repeated every extremely short time. Therefore, the original output audio signal output from the audio control device 45 is alternately switched between the first filter 40 and the second filter 41 in response to switching between the first filter 40 and the second filter 41 every extremely short time. To pass through. At this time, if the switching timing is executed at a sufficiently high speed, the first filter 40 and the second filter 41 are switched before the frequency component of the original output audio signal generated by the audio control device 45 changes significantly. Therefore, as shown in FIG. 11, the original output audio signal OS output at a certain moment passes through the second filter 41 and becomes the output audio signal OS2 thinned out by the second specific frequency region. Then, sound corresponding to the output sound signal OS2 is output from the speaker 11. The original output audio signal output at the next moment hardly changes from the original output audio signal OS output at a certain moment, passes through the first filter 40, and is thinned out by the first specific frequency region. The output audio signal OS1. Then, sound corresponding to the output sound signal OS1 is output from the speaker 11. As a result, the sound heard by the user is a sound corresponding to the sound signal OS3 obtained by synthesizing the output sound signal OS1 and the output sound signal OS2. The audio signal OS3 is substantially the same signal as the original output audio signal including all frequency components.

  Since the input side also performs voice recognition alternately for both the input voice signal thinned out by the first specific frequency region and the input voice signal thinned out by the second specific frequency region, This is almost equivalent to performing speech recognition on the original input speech signal including frequency components.

  According to the voice processing device 1C, the same effect as that of the voice processing device 1A can be obtained, and voice that does not feel uncomfortable for the user can be output, and voice recognition accuracy is improved.

  As mentioned above, although embodiment which concerns on this invention was described, this invention is implemented in various forms, without being limited to the said embodiment.

  For example, although the present embodiment is applied to a car navigation system, the present invention can also be applied to other systems that perform interactive voice processing, such as a robot that can interact with a user. Further, in the present embodiment, the voice control device is configured to perform each process of voice recognition, conversation configuration, and utterance voice generation. However, the processing content to be performed may be changed according to the purpose of the system to be applied. Also, not only when one type of voice is output on the system side, but also when multiple personal robots have different timbres, or when there are multiple agents on the car navigation system, The present invention can also be applied when outputting sound.

  In the present embodiment, the second filter is provided between the sound control device and the speaker. However, the second filter is not provided, and the second filter is the same as when the second filter is passed through the sound control device. You may comprise so that the audio | voice which consists of a frequency component of a specific frequency area | region may be produced | generated.

  Also, in this embodiment, the speech output for speech from the speech processing device is applied to the conversation, but applied to various sounds such as mechanical sounds such as “boo” and “pea” and music. Is possible.

  In the present embodiment, the first specific frequency region and the second specific frequency region are alternately arranged at a predetermined frequency interval. However, the first specific frequency region and the second specific frequency region are arranged in a specific frequency band. It is good also as a structure arrange | positioned collectively. For example, in the case of mechanical sounds such as “boo” and “pea”, the generated sound is in a specific narrow frequency band, so the second specific frequency region is arranged in this frequency band, and other frequency bands The first specific frequency region is arranged in

  Further, in the first embodiment, the frequency bands that are allowed to pass through and the frequency bands that are not allowed to pass in the first filter and the second filter are alternately arranged with the same width, but they may have different widths or may be passed through. If the bands are arranged so as not to overlap, the frequency bands not allowed to pass may overlap.

  Further, in the second embodiment, the characteristics of the collected voice are detected and the person is specified. However, the ID number assigned to each individual is input and the person is identified by collating the ID number. The person may be specified by various methods such as a configuration for specifying, a configuration for specifying a person by capturing a face of a person and recognizing the face image, or a configuration for specifying a person from a fingerprint or an iris.

1 is an overall configuration diagram of a speech processing apparatus according to the present embodiment. It is a block diagram of the speech processing unit which concerns on 1st Embodiment. It is explanatory drawing of the process in the audio | voice processing apparatus of FIG. 1 when only a user utters an audio | voice. It is a figure which shows the audio | voice signal collected by a microphone when the audio | voice which a user utters, and the audio | voice which an audio processing apparatus outputs overlap. It is explanatory drawing of the process in the audio | voice processing apparatus of FIG. 1 in case the audio | voice which a user utters, and the audio | voice which an audio processing apparatus outputs overlap. It is a block diagram of the speech processing unit which concerns on 2nd Embodiment. It is an example of the frequency characteristic of the filter according to the characteristic of a user's voice, (a) is the frequency characteristic of the 1st filter, and (b) is the frequency characteristic of the 2nd filter. It is a block diagram of the audio | voice processing apparatus which concerns on 3rd Embodiment. It is explanatory drawing of the process in the audio | voice processing apparatus of FIG. 8 when the 1st filter switches to the sound collection side and the 2nd filter switches to the audio | voice output side. It is explanatory drawing of the process in the audio | voice processing apparatus of FIG. 8 when the 2nd filter switches to the sound collection side and a 1st filter switches to the audio | voice output side. It is an example of the frequency characteristic of the audio | voice output from the audio | voice processing apparatus of FIG.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1,1A, 1B, 1C ... Audio processing apparatus, 10 ... Microphone, 11 ... Speaker, 12, 12A, 12B, 12C ... Audio processing unit, 20, 40 ... First filter, 21, 41 ... Second filter, 22, 35, 45 ... voice control device, 30 ... first variable filter, 31 ... second variable filter, 32 ... voice feature detection device, 33 ... specific frequency domain database, 34 ... specific frequency domain selection device, 42 ... switching timing generation device 43 ... 1st switching control device, 44 ... 2nd switching control device

Claims (12)

Sound collecting means,
Specific frequency passing means for passing a specific frequency region of the sound collected by the sound collecting means;
Sound processing means for performing sound processing based on the sound that has passed through the specific frequency passing means;
Sound generating means for generating sound based on a signal from the sound generation control device,
The audio processing apparatus according to claim 1, wherein the specific frequency region in the specific frequency passing unit is set to a frequency region different from the frequency region of the sound generated from the sound generating unit. Between the sound generation control device and the sound generation means, a second specific frequency passage means for passing a second specific frequency region of the sound based on a signal from the sound generation control device,
The audio processing apparatus according to claim 1, wherein the specific frequency region in the specific frequency passing unit and the second specific frequency region in the second specific frequency passing unit are set to different frequency regions.   The audio processing apparatus according to claim 2, wherein the specific frequency region in the specific frequency passing means and the second specific frequency region in the second specific frequency passing means are alternately set. Comprising a person specifying means for specifying a person who has emitted the sound collected by the sound collecting means,
The sound according to any one of claims 1 to 3, wherein the specific frequency region in the specific frequency passing means is changed according to the characteristics of the voice of the person specified by the person specifying means. Processing equipment.   The voice processing apparatus according to claim 4, wherein the person specifying unit is a feature detection unit that detects a feature of the voice collected by the sound collection unit.   4. The audio processing according to claim 2, wherein the specific frequency region in the specific frequency passing means and the second specific frequency region in the second specific frequency passing means are switched according to the passage of time. apparatus. Collecting steps;
A specific frequency passing step for passing the specific frequency region of the sound collected in the sound collecting step;
A voice processing step for performing voice processing based on the voice that has passed through the specific frequency passing step;
A sound generation step for generating sound based on a signal from the sound generation control device,
The specific frequency region in the specific frequency passage step is set to a frequency region different from the frequency region of the sound generated in the sound generation step. Including a second specific frequency passing step of passing a second specific frequency region of sound based on a signal from the sound generation control device;
The audio processing method according to claim 7, wherein the specific frequency region in the specific frequency passing step and the second specific frequency region in the second specific frequency passing step are set to different frequency regions.   The audio processing method according to claim 8, wherein the specific frequency region in the specific frequency passing step and the second specific frequency region in the second specific frequency passing step are alternately set. Including a person specifying step of specifying a person who has emitted the sound collected in the sound collecting step,
The sound according to any one of claims 7 to 9, wherein the specific frequency region in the specific frequency passing step is changed according to the characteristics of the voice of the person specified in the person specifying step. Processing method.   The voice processing method according to claim 10, wherein the person specifying step detects a feature of the voice collected in the sound collecting step.   10. The audio processing according to claim 8, wherein the specific frequency region in the specific frequency passing step and the second specific frequency region in the second specific frequency passing step are switched according to the passage of time. Method.

Images