JP2008147914A - Echo canceler and echo canceling method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an echo canceler and an echo canceling method capable of eliminating echo while reducing cost. <P>SOLUTION: A talker voice spectrum F1 (ω) corresponding to an echo can be eliminated from a received voice spectrum F2 (ω) without providing a complicated and expensive device apparatus such as an adaptive filter by eliminating the talker voice spectrum F1 (ω) from the received voice spectrum F2 (ω) on the basis of a characteristic spectrum waveform inherent to the voice of the talker voice spectrum F1 (ω) generated on the basis of a talker voice, the cost of the entire device can be reduced to the extent that an expensive adaptive filter for echo elimination is not provided in this way, and the echo can be eliminated. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to an echo canceller and an echo cancellation method. For example, a hands-free call system such as a video conference system or a hands-free phone, a voice recognition function and a dialog control function are mounted, and simple daily conversations with a user can be performed. The present invention is suitable for application to entertainment robots and the like.

  In a hands-free call system such as a video conference system or a hands-free phone, a microphone and a speaker are arranged in the same room for bidirectional audio transmission. For this reason, the participant's voice passes through the transmission path from the microphone, then wraps around from the speaker on the other party's side to the microphone, and returns through the transmission path again. Due to such acoustic coupling, a time delay occurs in the returning voice, which is detected as an echo.

  In particular, in a hands-free call system in which a voice signal is converted into an IP (Internet Protocol) packet and voice communication is performed via a network such as the Internet, a delay occurs when the IP packet passes through a router on the network. Based on this, echo due to acoustic coupling may occur.

As an echo canceller for removing echo due to such acoustic coupling, an adaptive filter that generates an echo-like pseudo echo signal based on an audio signal supplied to a speaker is provided, and this pseudo-acoustic signal is collected from a received sound signal collected by a microphone. An apparatus that adaptively removes echoes by subtracting echo signals has been proposed (see, for example, Patent Document 1). Further, such an echo canceller is considered to be used not only in a hands-free call system but also in various fields such as a robot apparatus on the voice conversation side.
Japanese Patent No. 2618888

  However, the echo canceller having such a configuration needs to control each circuit unit in a complicated manner, which increases the scale of the apparatus and the software, and requires the discrimination of the audio signal using a complicated signal detector. There was a problem that the manufacturing cost was high.

  The present invention has been made in consideration of the above points, and an object thereof is to provide an echo canceller and an echo cancellation method capable of removing echoes while reducing costs.

  An echo canceller according to a first aspect of the present invention is an echo canceller for removing echoes generated when sound output through a speaker wraps around a microphone and is collected, and is output from the speaker. An echo component acquisition means for acquiring an echo component spectrum obtained by frequency analysis of the sound to be an echo component, and obtaining a sound reception spectrum obtained by frequency analysis of a sound reception signal obtained by collecting the sound with the microphone; Echo removal means for removing the echo component spectrum from the sound reception spectrum based on a characteristic spectrum waveform unique to speech possessed by the component spectrum is provided.

  In the echo canceller according to claim 2 of the present invention, the voice signal transmitted to the communication terminal on the receiver side through the network is output as a voice from the speaker of the communication terminal, circulates to the microphone, and returns through the network again. An echo canceller for removing an echo generated by the transmission, and an echo component acquisition means for acquiring an echo component spectrum obtained by frequency analysis of the voice signal transmitted to the communication terminal; and the communication through the network An echo removing unit that obtains a sound reception spectrum obtained by frequency analysis of a sound reception signal received from a terminal, and removes the echo component spectrum from the sound reception spectrum based on a characteristic spectrum waveform peculiar to the voice of the echo component spectrum; It is characterized by providing.

  According to a third aspect of the present invention, the echo canceller includes a delayed echo component generation unit that generates a plurality of delayed echo component spectra that are shifted in timing by delaying the echo component spectrum on a time axis, and the echo removal unit includes: It is characterized in that it is determined whether or not each characteristic spectrum waveform is included in the sound reception spectrum at each timing of the echo component spectrum and the plurality of delayed echo component spectra.

  According to a fourth aspect of the present invention, there is provided an echo canceller comprising: a frame delay echo component generation means for generating a plurality of frame delay echo component spectra obtained by shifting the echo component spectrum by shifting on a time axis in units of frames of a predetermined length. And the echo removing means determines whether each of the received spectrum has the characteristic spectrum waveform at each timing of the echo component spectrum and the plurality of frame delay echo component spectra. To do.

An echo canceling method according to claim 5 of the present invention is an echo canceling method for removing echoes generated when sound output from a speaker wraps around and is collected by a microphone. An echo component acquisition step of acquiring an echo component spectrum obtained by frequency analysis of the sound to be output as an echo component;
A sound reception spectrum acquisition step of acquiring a sound reception spectrum obtained by frequency analysis of a sound reception signal obtained by collecting the sound with the microphone, and a voice-specific characteristic spectrum waveform of the echo component spectrum from the sound reception spectrum And an echo removal step to be removed.

  In the echo canceling method according to claim 6 of the present invention, an audio signal transmitted to a communication terminal on the receiver side through a network is output as a sound from a speaker of the communication terminal and circulates to a microphone, and again passes through the network. An echo cancellation method for removing an echo generated by returning, an echo component acquisition step of acquiring an echo component spectrum obtained by frequency analysis of the audio signal transmitted to the communication terminal, and through the network A sound reception spectrum acquisition step of acquiring a sound reception spectrum obtained by frequency analysis of a sound reception signal received from the communication terminal; and an echo removal step of removing a characteristic spectrum waveform peculiar to the voice of the echo component spectrum from the sound reception spectrum; It is characterized by comprising

  The echo cancellation method according to claim 7 of the present invention includes a delayed echo component spectrum generation step for generating a plurality of delayed echo component spectra with the timing shifted by delaying the echo component spectrum on a time axis, The step is characterized by determining whether or not each characteristic spectrum waveform is included in the sound reception spectrum at each timing of the echo component spectrum and the plurality of delayed echo component spectra.

  According to an eighth aspect of the present invention, there is provided an echo cancellation method for generating a plurality of frame delay echo component spectra by delaying the echo component spectrum by shifting on a time axis in units of frames of a predetermined length. A step of determining whether the echo spectrum has the characteristic spectrum waveform in the received spectrum at each timing of the echo component spectrum and the plurality of frame delay echo component spectra. It is a feature.

  According to the echo canceller of claim 1 and the echo cancellation method of claim 5 of the present invention, the echo is detected from the received spectrum based on the characteristic spectral waveform of the echo component spectrum obtained by frequency analysis of the voice. By removing the component spectrum, it is possible to remove the echo component spectrum corresponding to the echo from the received spectrum without providing a complicated and expensive device such as an adaptive filter. Echoes can be removed while reducing the cost of the entire apparatus by the amount that no adaptive filter is provided.

  According to the echo canceller according to claim 2 and the echo cancellation method according to claim 6 of the present invention, the voice signal transmitted to the communication terminal on the receiver side through the network is output as a voice from the speaker of the communication terminal, and the microphone is used. Even if an echo is generated by going back to the network and returning again through the network, the echo can be removed without providing complicated and expensive equipment such as an adaptive filter. The cost of the entire apparatus can be reduced by the amount that the expensive adaptive filter is not provided.

  According to the echo canceller according to claim 3 and the echo cancellation method according to claim 7 of the present invention, even if the echo component spectrum appears late in the received sound spectrum, the timing of the plurality of delayed echo component spectra It is possible to determine whether or not there is a characteristic spectrum waveform, and to reliably remove the echo component spectrum that is an echo component.

  According to the echo canceller according to claim 4 and the echo cancellation method according to claim 8 of the present invention, even if a delay occurs on the network and an echo component spectrum appears at a timing delayed in frame units in the received spectrum. It is possible to determine whether or not there is a characteristic spectrum waveform at the timing of a plurality of delayed echo component spectra, and to reliably remove the echo component spectrum that is an echo component.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

(1) First Embodiment In FIG. 1, reference numeral 1 denotes a hands-free call system as a whole. This hands-free call system 1 includes a plurality of call terminals 2A and 2B installed at different points, for example, Internet lines. Are connected to each other via the network 3.

  Incidentally, although these call terminals 2A and 2B have the same configuration, for convenience of explanation, one call terminal 2A is set as the receiver side, and the other call terminal 2B is set as the caller side. In the first embodiment, The following description will focus on the call terminal 2A on the receiver side. In the case of this embodiment, the communication terminals 2A and 2B have the speaker 4 and the microphone 5 arranged in the vicinity of the vicinity, so that the communication terminals 2A and 2B as a whole are reduced in size.

  In this case, the call terminal 2A receives the received signal S1 received from the network 3 via an antenna (not shown) by performing frequency analysis processing and demodulation processing in the signal processing unit 7 in units of frames of a predetermined length. This is obtained after obtaining data, performing predetermined data processing (for example, processing for detecting and masking a code error occurring during transmission) and decoding, and digital / analog conversion of the resulting audio signal. The transmitted speaker voice spectrum F1 (ω) is output as the voice of the speaker (speaker voice) through the speaker.

  In addition to such a configuration, the call terminal 2A includes an echo canceller 12 including a voice spectrum acquisition unit 10 and an echo removal unit 11 in the main body 13. The echo canceller 12 is sequentially sent to the voice spectrum acquisition unit 10 as the echo component acquisition means when the speaker voice spectrum F1 (ω) is sent from the signal processing unit 7 to the speaker 4. The voice spectrum acquisition unit 10 acquires the speaker voice spectrum F1 (ω). When the voice spectrum acquisition unit 10 acquires the talker voice spectrum F1 (ω) as the echo component spectrum, the voice spectrum acquisition unit 10 sends it to the echo removal unit 11.

  On the other hand, the microphone 5 collects the voice of the receiver (the voice of the receiver) or the voice of the sender when the speaker voice wraps around the microphone 5 from the speaker 4. When the microphone 5 collects the voice of the receiver and the voice of the sender, the microphone 5 sends them to the spectrum processing unit 15 as a received sound signal. The spectrum processing unit 15 performs frequency analysis processing on the sound reception signal S2, and then transmits the obtained sound reception spectrum F2 (ω) to the echo removal unit 11.

  When the echo removal unit 11 receives the sound reception spectrum F2 (ω) from the spectro processing unit 15, the sound reception spectrum F2 (ω) using the transmitter voice spectrum F1 (ω) received from the voice spectrum acquisition unit 10 in advance. Is subjected to echo cancellation processing described later, thereby removing the spectrum component corresponding to the transmitter voice spectrum F1 (ω) from the received voice spectrum F2 (ω), and the obtained speaker voice spectrum F3 (ω ) To the signal processing unit 7.

  Here, the echo removing unit 11 is configured to perform spectrum separation processing by band selection (that is, binary mask method) as echo removal processing. The received sound spectrum F2 (ω) and the transmitter voice spectrum F1 (ω). And the magnitude of the power is compared for each frequency band between the transmitter voice spectrum F1 (ω) and the sound reception spectrum F2 (ω).

  Here, since the speaker voice spectrum F1 (ω) is generated based on the voice of the sender, the power distribution on the time axis has a characteristic that sparse irregularities are formed (sparking). It becomes a characteristic spectrum waveform peculiar to voice. As a result, the echo removing unit 11 is almost the same as the transmitter voice spectrum F1 (ω) in the received spectrum F2 (ω) based on the distribution of the power of the transmitter voice spectrum F1 (ω) on the time axis. It is determined whether or not it has a frequency band portion that becomes a characteristic spectrum waveform peculiar to speech.

  Here, in this embodiment, since the speaker 4 and the microphone 5 are disposed in the vicinity, the microphone 5 outputs the talker voice output from the speaker 4 to the microphone 5 almost simultaneously with the output from the speaker 4. A roundabout sound is collected. For this reason, the echo removing unit 11 receives the received sound spectrum F2 (ω) from the spectrum processing unit 15 at substantially the same timing when receiving the transmitter speech spectrum F1 (ω) from the voice spectrum acquiring unit 10.

  That is, when the voice of the speaker output from the speaker 4 goes around the microphone 5 and is collected, when the sound reception spectrum F2 (ω) is received from the spectrum processing unit 15, the sound reception spectrum F2 is received. (Ω) includes a characteristic spectrum waveform that is almost the same as the speaker voice spectrum F 1 (ω).

  At this time, the echo removing unit 11 removes the transmitter speech spectrum F1 (ω) corresponding to the echo component from the received sound spectrum F2 (ω) based on the binary mask method, and the remaining from the received sound spectrum F2 (ω). Can be extracted and sent to the signal processing unit 7.

  The signal processing unit 15 receives the listener's voice spectrum F3 (ω) from the echo removing unit 11, performs analog / digital conversion on the receiver's voice spectrum F3 (ω), and then encrypts the obtained speaker's voice signal to the obtained speaker's voice signal. As a result, transmission data is obtained in frame units of a predetermined length, and a transmission signal S3 is generated by performing modulation processing and frequency analysis processing on the transmission data, and this is transmitted via the antenna and the network 3 The data is transmitted to the call terminal 2B on the sender side.

  In the above configuration, the echo canceller 12 obtains the transmitter voice spectrum F1 (ω) output from the speaker 4 in advance, and the received sound spectrum F2 (frequency analysis of the received sound signal output from the microphone 5). ω) and the power of the speaker's voice spectrum F1 (ω) are compared, and a characteristic spectrum waveform almost identical to the power distribution waveform on the time axis of the speaker's voice spectrum F1 (ω) is echoed. As a component, it is removed from the sound reception spectrum F2 (ω).

  As described above, the echo canceller 12 uses the speaker voice spectrum F1 (ω) generated based on the voice of the talker and having a characteristic spectrum waveform unique to the voice, thereby receiving the voice based on the characteristic spectrum waveform. From the sound spectrum F2 (ω), the sender voice spectrum F1 (ω) as an echo component can be easily discriminated. Thus, the echo canceller 12 can reliably remove only the transmitter voice spectrum F1 (ω) as an echo component from the received spectrum F2 (ω) and extract the receiver voice spectrum F3 (ω). it can.

  In this way, the echo canceller 12 transmits only clear receiver speech without echoes to the talker terminal 2B on the transmitter side without providing an expensive adaptive filter that generates an echo-like pseudo echo signal as in the prior art. can do.

  Further, when echo is removed using a conventional adaptive filter, the adaptive filter requires a predetermined time for convergence, and thus an echo may occur in a transient state. However, in the echo canceller 12 of the present invention, a microphone is used. The occurrence of echoes can be prevented immediately after starting to collect sound from 5.

  According to the above configuration, based on the characteristic spectrum waveform unique to the voice of the speaker voice spectrum F1 (ω) generated based on the voice of the sender, the transmission of the voice from the received spectrum F2 (ω). By removing the speaker voice spectrum F1 (ω), the transmitter voice spectrum equivalent to the echo from the received sound spectrum F2 (ω) can be obtained without providing complicated and expensive equipment such as an adaptive filter. F1 (ω) can be removed, and thus the cost of the entire apparatus can be reduced and the echo can be removed by the amount that the expensive adaptive filter for removing the echo is not provided.

  In addition, this invention is not limited to said embodiment, A various deformation | transformation implementation is possible. For example, in the first embodiment described above, the binary mask method is used as an echo removal process for removing a characteristic spectrum waveform corresponding to the transmitter voice spectrum F1 (ω) from the received sound spectrum F2 (ω). However, the present invention is not limited to this, and a spectral subtraction (SS) method may be used.

  Here, when the spectral subtraction (SS) method is used instead of the binary mask method, the echo canceller 12 uses the power γ of the received spectrum F2 (ω) for each frequency band, and the speaker voice spectrum F1. A value (K × δ) obtained by multiplying the power δ of (ω) by a coefficient K is subtracted, thereby generating a receiver voice spectrum F3 (ω) from which the speaker voice spectrum F1 (ω) is removed.

  The coefficient K is, for example, a coefficient that depends on the magnitude of the difference between the power γ of the sound reception spectrum F2 (ω) and the power δ of the transmitter voice spectrum F1 (ω). Further, for example, in the frequency band where the power γ of the sound reception spectrum F2 (ω) is smaller than the value (K × δ) obtained by multiplying the power δ of the speaker voice spectrum F1 (ω) by the coefficient K, For example, a minimum value determined by a certain rule (may be a constant value for each frequency band, or a value proportional to each power value for each frequency band of the speaker voice spectrum F1 (ω)). It may be a calculated value or may be zero. Thereby, the same effect as the above-described embodiment can be obtained.

  In the above-described embodiment, the transmitter voice spectrum F1 (ω) is included in the received spectrum F2 (ω) almost simultaneously with the reception of the transmitter voice spectrum F1 (ω) from the voice spectrum acquisition unit 10. The case where it is determined whether or not there is a substantially identical characteristic spectrum waveform has been described. However, the present invention is not limited to this, and a plurality of transmission voice frames F1 (ω) delayed by a predetermined time unit. A delayed speech spectrum is generated by the speech spectrum acquisition unit 10 as a delayed echo component generating means, and it is determined whether or not there is a characteristic spectrum waveform in the received spectrum F2 (ω) at the timing of the plurality of delayed speech spectra. You may make it do.

  In this case, the speaker 4 and the microphone 5 are arranged apart from each other, and it takes a little time for the sound output from the speaker 4 to enter the microphone 5, and a characteristic spectrum waveform is transmitted in the received sound spectrum F2 (ω). Even if the speaker voice spectrum F1 (ω) appears later than the timing when it is sent to the echo removing unit 11, it is possible to determine whether or not there is a characteristic spectrum waveform at the timings of a plurality of delayed voice spectra, thus echoing. Only the speaker voice spectrum F1 (ω) as a component can be surely removed.

(2) Second Embodiment In FIG. 2, in which parts corresponding to those in FIG. 1 are assigned the same reference numerals, 31 denotes a hands-free call system according to the second embodiment, which is the same as in the first embodiment. The configuration of the echo canceller 32 is different. Incidentally, although these call terminals 2A and 2B have the same configuration, for convenience of explanation, one call terminal 2A is set as the receiver side and the other call terminal 2B is set as the caller side. In the second embodiment, The following description will focus on the call terminal 2B on the other transmitter side.

  The call terminal 2B collects the talker's voice using the microphone 5, and sends the obtained voice signal S2a to the spectrum processing unit 15. The spectrum processing unit 15 performs frequency analysis processing on the voice signal S2a, and then sends the obtained speaker voice spectrum F4 (ω) to the signal processing unit 7 and the echo canceller 32. The echo canceller 32 has a voice spectrum acquisition unit 33 and an echo removal unit 34, and the voice spectrum acquisition unit 33 acquires the speaker voice spectrum F4 (ω) as an echo component spectrum transmitted from the spectrum processing unit 15. I will do it.

  In addition to this, the voice spectrum acquisition unit 33 keeps the characteristic spectrum waveform peculiar to the voice of the talker voice spectrum F4 (ω) as it is and delays the delayed voice spectrum F5a sequentially in time by one frame. (Ω), F5b (ω),... Can be generated.

  The speech spectrum acquisition unit 33 serving as the delayed echo component generating means and the frame delayed echo component generating means is configured to transmit the speaker sound characteristics F4 (ω) and the delayed speech spectra F5a (ω), F5b as the frame delayed echo component spectra. (Ω),... Are sent to the echo removing unit 34. The echo removing unit 34 temporarily stores the speaker voice spectrum F4 (ω) and a plurality of delayed voice spectra F5a (ω), F5b (ω),. It is made to do.

  When the signal processing unit 7 receives the speaker voice spectrum F4 (ω) from the spectrum processing unit 15, the signal processing unit 7 performs analog / digital conversion on the speaker voice spectrum F4 (ω) and then performs predetermined data processing such as encryption. As a result, transmission data is obtained in a frame unit of a predetermined length, and a transmission signal S3 obtained by performing modulation processing and frequency analysis processing on the transmission data is received via an antenna (not shown) and the network 3 To the call terminal 2A.

  In addition, when the communication terminal 2B on the transmitter side receives the received sound signal S4 transmitted from the communication terminal 2A of the receiver through the antenna, the communication terminal 2B transmits the received signal S4 to the signal processing unit 7.

  By the way, in the communication terminal 2A on the receiver side, the receiver's voice is collected by the microphone 5. At this time, when the speaker's voice output from the speaker 4 wraps around the microphone 5, the speaker's voice becomes the speaker 4's. Is output to the microphone 5 and collected at almost the same time. For this reason, the sender's communication terminal 2 </ b> B returns to the sender's communication terminal 2 </ b> A through the network 3 again with the speaker's voice that has entered the microphone 5 superimposed on the receiver's voice.

  The communication terminal 2B on the transmitter side performs a frequency analysis process and a demodulation process on the received sound signal S4 received from the network 3 via an antenna (not shown) in the signal processing unit 7, thereby a frame having a predetermined length. Received data is obtained in units, predetermined data processing (for example, processing for detecting and masking a code error occurring during transmission) and decoding, and the resulting audio signal is digital / analog converted and then obtained. The received sound spectrum F6 (ω) is sent to the echo removing unit 34.

  When the echo removal unit 34 receives the received sound spectrum F6 (ω) from the signal processing unit 7, the echoer sound spectrum F4 (ω) received from the sound spectrum acquisition unit 33 and the plurality of delayed sound spectra F5a (ω), Echo removal processing is performed on the sound reception spectrum F6 (ω) using any one of F5b (ω),... The characteristic spectrum waveform corresponding to any one of the delayed speech spectra F5a (ω), F5b (ω),... Is removed, and the speaker speech spectrum F7 (ω) obtained thereby is output to the speaker 4. ing.

  In practice, the echo removal unit 34 performs spectrum separation processing by band selection as echo removal processing, and the power of each frequency band between the sound reception spectrum F6 (ω) and the transmitter voice spectrum F2 (ω) is obtained. After comparing the magnitude, the magnitude of power is compared for each frequency band between the received sound spectrum F6 (ω) and the plurality of delayed speech spectra F5a (ω), F5b (ω),.

  Here, the speaker's voice passes through the network 3 from the microphone 5 of the communication terminal 2B on the transmitter side, and then enters the microphone 5 from the speaker 4 of the communication terminal 2A on the receiver side. When there is no delay at the time of returning, the sound reception received from the signal processing unit 7 almost at the same time when the echo removal unit 34 receives the transmitter speech spectrum F2 (ω) from the speech spectrum acquisition unit 33. In the spectrum F6 (ω), a characteristic spectrum waveform peculiar to speech corresponding to the speaker speech spectrum F2 (ω) is formed.

  At this time, the echo removing unit 34 removes the speaker voice spectrum F2 (ω) corresponding to the echo component from the sound reception spectrum F6 (ω), and extracts only the speaker voice spectrum F7 (ω). It can be output to the speaker 4.

  On the other hand, when a large number of relay points (routers) in the network 3 are routed, the transit time of the router increases by the number of routers. For example, the communication terminal 2A on the receiver side passes through the network 3. Thus, a delay is likely to occur when the received sound spectrum F6 (ω) reaches the communication terminal 2B on the transmitter side.

  For this reason, when a delay occurs, for example, by a predetermined number of frames on the network 3 of the hands-free call system 31, it is determined whether or not a characteristic spectrum waveform exists in the received sound spectrum F6 (ω). Due to the shift in timing, the transmitter voice spectrum F2 (ω) can be discriminated from the received spectrum F6 (ω) when the echo removing unit 34 receives the received spectrum F6 (ω). Absent.

  At this time, the echo removing unit 34 compares the plurality of previously stored delayed speech spectrums F5a (ω), F5b (ω),... With the received spectrum F6 (ω), and the received received spectrum F6 ( It is determined whether or not there is a delayed speech spectrum F5a (ω), F5b (ω),... at a corresponding timing from among these delayed speech spectra F5a (ω), F5b (ω),.

  That is, the echo removing unit 34 first selects a delayed speech spectrum F5a (ω) shifted by one frame from the delayed received sound spectra F5a (ω), F5b (ω),... As delayed echo component spectra. Then, it is determined whether or not there is a characteristic spectrum waveform in the sound reception spectrum F6 (ω) at the timing of the delayed sound spectrum F5a (ω).

  When the echo removing unit 34 determines that the characteristic spectrum waveform does not exist in the received sound spectrum F6 (ω) at the timing of the delayed sound spectrum F5a (ω) shifted by one frame, the delayed sound shifted by two frames next. The spectrum F5b (ω) is selected, and it is determined whether or not the received spectrum F6 (ω) has a characteristic spectrum waveform at the timing of the delayed speech spectrum F5b (ω).

  In this manner, the echo removing unit 34 sequentially compares the plurality of delayed sound spectra F5a (ω), F5b (ω),... With the sound reception spectrum F6 (ω) received from the signal processing unit 7, and The characteristic spectrum waveform of the delayed speech spectrum F5a (ω), F5b (ω),... At any timing of the delayed speech spectrum F5a (ω), F5b (ω),. Judge whether there is.

  The echo removing unit 34 then includes the delayed speech spectrum F5a (ω), at a timing corresponding to any one of the plurality of delayed speech spectra F5a (ω), F5b (ω),. If it is determined that there is F5b (ω),..., The delayed speech spectrums F5a (ω), F5b (ω),... Are removed from the received spectrum F6 (ω), and the obtained speaker speech spectrum F7 ( ω) is output from the speaker 4.

  In the above configuration, the echo canceller 32 acquires the speaker voice spectrum F2 (ω) corresponding to a frame unit of a predetermined length, and the voice-specific characteristic spectrum waveform of the speaker voice spectrum F2 (ω). Are generated as a plurality of delayed speech spectra F5a (ω), F5b (ω),..., Which are sequentially delayed by one frame at a time, and the sender speech spectrum F2 (ω) and a plurality of delays are generated. The audio spectrum F5a (ω), F5b (ω),... Is stored in the echo removing unit 34.

  Further, in the echo canceller 32, at the timing when the echo removing unit 34 receives the sound reception spectrum F6 (ω) from the signal processing unit 7, the speaker voice spectrum F2 (ω) in the sound reception spectrum F6 (ω). Whether or not it has a characteristic spectral waveform corresponding to is determined.

  In the echo canceller 32, when the echo removing unit 34 receives the speaker voice spectrum F2 (ω) from the voice spectrum acquiring unit 33, the received sound spectrum F6 (ω) received from the signal processing unit 7 almost simultaneously. If it is determined that the voice has a characteristic spectrum waveform unique to the voice corresponding to the voice spectrum F2 (ω), the voice spectrum F2 (ω) corresponding to the echo component is removed from the received voice spectrum F6 (ω). Thus, only the speaker voice spectrum F7 (ω) can be extracted from the sound reception spectrum F6 (ω).

  In addition, in the echo canceller 32, the sound reception spectrum F6 (ω) is received at the timing of a plurality of delayed speech spectra F5a (ω), F5b (ω),... It is determined whether or not it has a characteristic spectrum waveform corresponding to any of the delayed speech spectrums F5a (ω), F5b (ω),.

  Therefore, in the echo canceller 32, when the transmission signal S3 arrives from the communication terminal 2B on the sender side through the network 3 to the communication terminal 2A on the receiver side, or from the communication terminal 2A on the receiver side through the network 3. When the received signal S4 arrives at the communication terminal 2B on the transmitter side, even if a delay on various networks 3 such as a delay caused by passing through a number of routers in the network 3 occurs, the frame unit , By using a plurality of delayed sound spectra F5a (ω), F5b (ω),... With different timings, the characteristic spectrum waveforms shifted in time in the received sound spectrum F6 (ω). And the delayed speech spectrums F5a (ω), F5b (ω),..., Which are echo components, can be reliably removed from the received sound spectrum F6 (ω).

  Therefore, in this echo canceller 32, even if a delay on the network 3 occurs, the echo from the speaker voice spectrum F2 (ω) is echoed without providing a complicated and expensive device such as an adaptive filter. The corresponding delayed speech spectrums F5a (ω), F5b (ω),... Can be removed, and thus the cost of the entire apparatus can be reduced by providing no expensive adaptive filter for removing the echoes. Can be removed.

(3) Third Embodiment In FIG. 3, reference numeral 41 denotes a speech processing apparatus according to a third embodiment which is incorporated in the robot apparatus as a whole. In this case, the overall view of the robot apparatus is omitted, and only the voice processing apparatus 41 will be described.

  The voice processing device 41 generates a synthesized voice spectrum F8 (ω) by performing predetermined data processing on the synthesized voice generated by the voice processing unit 42, and sends this to the speaker 4 and the voice spectrum acquisition unit 10. As a result, the speech processing apparatus 41 sends the synthesized speech spectrum F8 (ω) to the echo removing unit 11 via the speech spectrum acquisition unit 10 and outputs the synthesized speech spectrum F8 (ω) from the speaker 4 as synthesized speech.

  Here, in the case of this embodiment, since the speaker 4 and the microphone 5 are disposed in the vicinity, when the user's voice is collected by the microphone 5, the synthesized voice is output from the speaker 4. Then, the synthesized voice goes around the microphone 5 and is collected by the microphone 5 almost simultaneously with the output from the speaker 4. Therefore, the echo removing unit 11 receives the received sound spectrum F2 (ω) from the spectrum processing unit 15 almost simultaneously when receiving the synthesized speech spectrum F8 (ω) from the sound spectrum acquiring unit 10.

  The echo removing unit 11 has substantially the same characteristics as the synthesized speech spectrum F8 (ω) in the received sound spectrum F2 (ω) when the synthesized speech output from the speaker 4 wraps around the microphone 5 and is collected. The result of determination that there is a target spectrum waveform is obtained.

  At this time, the echo removing unit 11 removes the synthesized speech spectrum F8 (ω) corresponding to the echo component from the received sound spectrum F2 (ω), and the remaining users who have removed the echo component from the received sound spectrum F2 (ω). Only the speech spectrum F9 (ω) can be extracted and sent to the speech processing unit 42.

  As a result, the voice processing device 41 sends the user voice spectrum F9 (ω) consisting of the clear user voice from which the echo component has been removed by the echo canceller 12 to a voice recognition processing unit (not shown) or the like. The robot apparatus is configured to reliably perform various operations based on the above.

  In the above configuration, the echo canceller 12 stores in advance the synthesized speech spectrum F8 (ω) output from the speaker 4 and having a characteristic spectrum waveform specific to speech, and the characteristic of the synthesized speech spectrum F8 (ω). Since the synthesized speech spectrum F8 (ω) can be easily determined from the received sound spectrum F2 (ω) based on the spectrum waveform, the synthesized speech spectrum F8 (ω) that is an echo component from the received sound spectrum F2 (ω). Can only be removed reliably.

  As a result, the echo canceller 12 can obtain only the clear user's voice without echo without providing an adaptive filter that generates a pseudo echo signal similar to the conventional echo. Etc. can be used.

  Although the first to third embodiments of the present invention have been described above, the present invention is not limited to the first to third embodiments, and various modifications can be made.

It is the schematic which shows the whole structure of the hands-free call system by 1st Embodiment. It is the schematic which shows the whole structure of the hands-free call system by 2nd Embodiment. It is the schematic which shows the whole structure of the speech processing unit by 3rd Embodiment.

Explanation of symbols

10 Speech spectrum acquisition unit (echo component acquisition means, delayed echo component generation means)
33 Speech spectrum acquisition unit (echo component acquisition means, delayed echo component generation means, frame delay echo component generation means)
11 Echo remover
12, 32 Echo canceller

Claims (8)

An echo canceller for removing echo generated when sound output through a speaker wraps around a microphone and is collected,
An echo component acquisition means for acquiring an echo component spectrum obtained by frequency analysis of the sound to be an echo component output from the speaker;
A sound reception spectrum obtained by frequency analysis of a sound reception signal obtained by collecting the sound with the microphone is obtained, and the echo component is obtained from the sound reception spectrum on the basis of a characteristic spectrum waveform unique to the voice of the echo component spectrum. An echo canceller comprising: an echo removing means for removing a spectrum. The voice signal transmitted to the communication terminal on the receiver side through the network is output as a voice from the speaker of the communication terminal, circulates to the microphone, and removes echo generated by returning through the network again. An echo canceller,
Echo component acquisition means for acquiring an echo component spectrum obtained by frequency analysis of the audio signal transmitted to the communication terminal;
A sound reception spectrum obtained by frequency analysis of a sound reception signal received from the communication terminal through the network is acquired, and the echo component is obtained from the sound reception spectrum on the basis of a characteristic spectrum waveform unique to the voice of the echo component spectrum. An echo canceller comprising: an echo removing means for removing a spectrum. Delay echo component generation means for generating a plurality of delayed echo component spectra that are shifted in timing by delaying the echo component spectrum on the time axis,
The echo removal means determines whether each of the characteristic spectrum waveforms is included in the sound reception spectrum at each timing of the echo component spectrum and the plurality of delayed echo component spectra. The echo canceller according to 1 or 2. Frame delay echo component generation means for generating a plurality of frame delay echo component spectra obtained by shifting the echo component spectrum by shifting on the time axis in units of a frame of a predetermined length,
The echo removal means determines whether or not each of the characteristic spectrum waveforms is included in the sound reception spectrum at each timing of the echo component spectrum and the plurality of frame delay echo component spectra. Item 3. The echo canceller according to Item 2. An echo canceling method for removing echo generated when sound output through a speaker wraps around a microphone and is collected,
An echo component acquisition step of acquiring an echo component spectrum obtained by frequency analysis of the sound to be an echo component output from the speaker;
A sound reception spectrum acquisition step of acquiring a sound reception spectrum obtained by frequency analysis of a sound reception signal obtained by collecting sound with the microphone;
An echo cancellation method comprising: an echo removal step of removing a characteristic spectrum waveform peculiar to speech possessed by the echo component spectrum from the received spectrum. The voice signal transmitted to the communication terminal on the receiver side through the network is output as a voice from the speaker of the communication terminal, circulates to the microphone, and removes echo generated by returning through the network again. An echo cancellation method,
An echo component acquisition step of acquiring an echo component spectrum obtained by frequency analysis of the audio signal to be transmitted to the communication terminal;
A sound reception spectrum obtaining step of obtaining a sound reception spectrum obtained by frequency analysis of a sound reception signal received from the communication terminal through the network;
An echo cancellation method comprising: an echo removal step of removing a characteristic spectrum waveform peculiar to speech possessed by the echo component spectrum from the received spectrum. A delayed echo component spectrum generating step for generating a plurality of delayed echo component spectra with the timing shifted by delaying the echo component spectrum on the time axis;
The echo removal step determines whether or not each of the characteristic spectrum waveforms is included in the sound reception spectrum at each timing of the echo component spectrum and the plurality of delayed echo component spectra. The echo cancellation method according to 5 or 6. A frame delay echo component generation step of generating a plurality of frame delay echo component spectra by delaying the echo component spectrum by shifting on the time axis in units of frames of a predetermined length;
The echo removal step determines whether or not each of the characteristic spectrum waveforms is included in the sound reception spectrum at each timing of the echo component spectrum and the plurality of frame delay echo component spectra. Item 7. The echo cancellation method according to item 6.

Images