JP2009031809A - Speech recognition apparatus - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To improve accuracy of speech recognition by appropriately removing a noise component included in a speech signal. <P>SOLUTION: A speech recognition apparatus includes an adaptive filter for removing the noise component from the speech signal which is input from a microphone. An LMS learning section repeatedly learns a filter coefficient based on an LMS method (S220), and the filter coefficient obtained as the result of the learning is set to the adaptive filter (S230). When a learning prohibition instruction is input from a control section concurrently with the start of the speech recognition, learning of the filter coefficient is stopped. Thereafter, when the speech signal required for the speech recognition is given to the speech recognition section, the filter coefficient is learned and updated again, according to a learning resumption instruction which is input from the control section (S260). <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to a speech recognition apparatus that recognizes speech input to a microphone by a user based on a speech signal obtained from a microphone.

  Conventionally, a voice uttered by a user is collected by a microphone, and compared with a voice pattern stored in advance as a recognized word, a recognized word with a high degree of matching is recognized as a vocabulary spoken by the user. A voice recognition device is known. This type of speech recognition device is incorporated in, for example, a car navigation device.

  It is well known that the speech recognition rate (accuracy rate of speech recognition) of such a speech recognition device depends on the amount of noise included in the speech signal input from the microphone. However, there is a problem that music or the like to be reproduced is collected by a microphone together with the user's voice as noise during operation of the in-vehicle audio device.

  To solve this problem, conventionally, the volume of music played on the in-vehicle audio device is adjusted during the speech recognition process by linking the in-vehicle audio device and the voice recognition device (for example, muting the in-vehicle audio device). In other words, the reproduced music or the like is not input to the microphone, and a voice recognition rate of a certain level or more is ensured. Since such prior art is well known and publicly used, related documents are not disclosed.

  However, the conventional speech recognition apparatus adjusts the volume of music or the like played on the in-vehicle audio device, so the music or the like is temporarily inaudible to the user, which causes the user to be dissatisfied. There was a possibility.

  Therefore, the present inventors learn a noise component included in an audio signal obtained from a microphone based on a reference signal obtained from a noise source (vehicle audio device), and remove noise learned from the audio signal. The removal unit is provided in the speech recognition apparatus.

  However, in a known learning method such as the least mean square (LMS) method, learning is repeated in such a direction that the signal after noise reduction becomes smaller, so that noise is generated while the user is inputting sound into the microphone. When the component learning is repeated, there is a problem in that the noise removing unit erroneously learns due to the influence of the utterance by the user, and the noise component included in the voice signal cannot be removed appropriately. Therefore, even if such a noise removing unit is introduced into the speech recognition apparatus, there is a limit to improving speech recognition accuracy.

  The present invention has been made in view of these problems, and an object of the present invention is to provide a speech recognition apparatus capable of accurately recognizing speech by appropriately removing a noise component contained in a speech signal to be recognized.

  According to the speech recognition apparatus of claim 1 made to achieve the above object, the noise removal signal generating means generates the reference signal input from the noise source in accordance with a preset filter coefficient. Using the noise removal signal, the noise removal means removes a noise component included in the voice signal input from the microphone, and outputs the voice signal after the noise removal. In addition, the speech recognition apparatus includes a coefficient updating unit. The coefficient updating unit learns a filter coefficient to be set in the noise removal signal generation unit based on the voice signal output from the noise removal unit. The obtained filter coefficient is set for the noise removal signal generating means.

  On the other hand, when an operation start command is input from the outside (for example, an operation switch such as a PTT switch), the voice recognition unit acquires a voice signal output from the noise removal unit for a predetermined period, and based on the voice signal, Recognizes the voice input to the microphone.

  In this speech recognition apparatus, the learning speed switching means causes the coefficient updating means to learn the filter coefficient at the first learning speed when the speech recognition means is not operating, and the speech recognition means is output from the noise removal means. While the voice signal is being acquired, the coefficient updating means is made to learn the filter coefficient at a second learning speed lower than the first learning speed.

  Unlike the normal and quasi-stationary sounds, the user's voice input to the microphone is a non-stationary voice that occurs suddenly. Therefore, the voice signal output from the noise removal unit by the voice recognition unit If the learning rate of the filter coefficient is slowed during the period when the filter coefficient is acquired, the influence of the user's voice during learning of the filter coefficient can be suppressed, and erroneous learning of the filter coefficient by the coefficient updating means can be suppressed. .

  That is, according to the speech recognition apparatus of the first aspect, the coefficient update unit can appropriately perform learning of the filter coefficient, and the noise removal accuracy can be improved as compared with the conventional case. Therefore, according to the present invention, the accuracy of speech recognition in the speech recognition apparatus can be improved.

  The learning speed switching means is configured to cause the coefficient updating means to learn the filter coefficient at the second learning speed at least from the time when the voice recognition means starts to acquire the voice signal from the noise removal means to the time when it ends. For example, after the acquisition of the audio signal is completed, the learning of the filter coefficient is updated at the second learning speed until the recognition of the audio by the audio recognition unit is completed and the operation of the audio recognition unit stops. The means may be executed.

  The learning speed switching means may be configured to cause the coefficient updating means to learn the filter coefficient at the first learning speed at least when the speech recognition means is not operating. The filter updating unit may be made to learn the filter coefficient at the first learning speed during the operation of the unit and during the period when the voice recognition unit does not acquire the voice signal output from the noise removing unit. . That is, the learning speed switching means learns the filter coefficient to the coefficient updating means at the first learning speed immediately after the voice recognition means acquires the voice signal regardless of whether or not the voice recognition means performs voice recognition. It may be configured to perform.

  In addition, the speech recognition means in the speech recognition apparatus of the present invention acquires the speech signal output from the noise removal means only during the utterance period in which the user uttered after the operation start command was input from the outside. It is good to be configured to do. If the speech recognition means has such a configuration, it is not necessary to use a speech signal in a noise section that does not include the content of the utterance by the user at the time of speech recognition, and the accuracy of speech recognition is improved.

  Further, when the voice recognition unit selectively acquires the voice signal during the utterance period by the user as described above, the utterance period during which the user uttered is changed based on the voice signal output from the noise removal unit. It is only necessary to provide in the apparatus an acquisition control means for making the voice recognition means selectively obtain only the voice signal corresponding to the utterance period among the voice signals output from the noise removal means.

  By the way, when the voice recognition unit does not acquire a voice signal from the noise removal unit at the same time as the operation start command, only the period during which the voice recognition unit acquires the voice signal output from the noise removal unit is obtained at the second learning speed. If the filter coefficients are learned, the apparatus configuration may become complicated.

  Therefore, in the above-described voice recognition device, the operation start command is input to the voice recognition unit, and at the same time, the coefficient update unit learns the filter coefficient at the second learning speed, and the voice recognition unit acquires the voice signal. The learning speed switching means may be configured so that the coefficient update means continues to learn the filter coefficient at the second learning speed during the period until the end.

  According to the speech recognition apparatus of the second aspect configured as described above, the speech recognition means performs speech to the speech recognition means from the noise removal means only by monitoring whether or not an operation start command is externally input to the speech recognition means. When acquiring the signal, the coefficient updating means can learn the filter coefficient at the second learning speed. That is, according to this voice recognition device, the learning speed of the coefficient updating means can be appropriately switched with a simple device configuration (control).

  In addition, the above-described invention may be applied to a speech recognition apparatus in which the coefficient updating unit learns a filter coefficient to be set in the noise removal signal generating unit using the LMS method.

  When the LMS method is used, if the voice input to the microphone includes voice generated from a sound source other than the noise source (that is, user voice), erroneous learning of the filter coefficient is likely to occur. Therefore, if the present invention (Claim 1 or Claim 2) is applied to a speech recognition apparatus that performs learning using the LMS method as described in Claim 3, the accuracy of speech recognition is effectively improved. be able to.

In addition, the invention described in claims 1 to 3 may be applied to a speech recognition apparatus in which the noise source is an audio device as described in claim 4.
According to the voice recognition device of the fourth aspect, it is possible to perform voice recognition with high accuracy without adjusting the volume of music or the like reproduced from the speaker by the operation of the audio device. is there.

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram illustrating a schematic configuration of the speech recognition apparatus 1.
A voice recognition device 1 according to this embodiment shown in FIG. 1 is connected to a car navigation device 3, recognizes a user's voice input to a microphone 5, and inputs an operation signal according to the voice to the car navigation device 3. Thus, an operation according to the user's voice is performed on the car navigation device 3.

  The speech recognition apparatus 1 mainly includes an audio canceller unit 20, a speech extraction unit 31, and a speech recognition unit connected to the microphone 5 and the in-vehicle audio device 7 via analog-digital converters (ADC) 11 and 13. 33, a PTT (Push to Talk) switch 35, a control unit 37, and a speech synthesis unit 39.

  The audio canceller unit 20 is mainly composed of an adaptive filter 21, a subtracting unit 23, and an LMS learning unit 25, and a subtracting unit for the audio signal y (t) input from the microphone 5 via the ADC 11. The audio signal x (t) input to the speaker 9 from the in-vehicle audio device 7 is acquired from the ADC 13 and the audio signal x (t) is input to the adaptive filter 21.

The adaptive filter 21 includes a register (not shown) that stores the filter coefficient w.
w = (w [0], w [1], ..., w [J]) ^T ... Formula (1)
The superscript T means a transposed matrix. The parameter J + 1 represents the tap length.

  The adaptive filter 21 includes a filter coefficient w set in a register in advance by an operation of the LMS learning unit 25 (details will be described later), and the audio signal x (t) obtained as a reference signal from the in-vehicle audio device 7 as a noise source. Is calculated by substituting into the following equation, the audio signal x (t) is filtered according to the filter coefficient w, and the noise removal signal c (t) for removing the noise component from the audio signal y (t) is generated. . Then, the noise removal signal c (t) is input to the subtraction unit 23.

c (t) = x ^T · w Equation (2)
However, the parameter x is a time series vector of the audio signal x (t) represented by the following equation. The parameter t here is a time parameter with a sampling period as a unit.

x = (x (t), x (t−1),
x (t-2), ... x (t-J)) ^T ... Formula (3)
On the other hand, the subtracting unit 23 subtracts the noise removal signal c (t) from the audio signal y (t) input from the microphone 5 via the ADC 11, so that the noise component included in the audio signal y (t) (that is, The audio component reproduced from the speaker 9 by the operation of the in-vehicle audio device 7 is removed, and the audio signal z (t) after noise removal is obtained.

z (t) = y (t) −c (t) (formula 4)
Further, the subtractor 23 inputs the audio signal z (t) after noise removal obtained as a result of the subtraction to the audio extractor 31.

  The voice extraction unit 31 is configured to start operation upon receiving an operation start command from the control unit 37. When the operation starts, the voice signal z (t) after noise removal input from the audio canceller unit 20 is started. Is a signal in a speech section (ie, a speech period in which the user utters) or a noise section signal that does not include the user's voice and does not belong to the speech section, If it is determined that the signal is a signal, the voice signal z (t) is input to the voice recognition unit 33. When the voice section ends, the operation is stopped.

  As a determination method, for example, the short-time power of the input signal is extracted at fixed time intervals, and is a voice interval or a noise interval depending on whether or not the short-time power equal to or greater than a predetermined threshold continues for a certain time. The method of determining whether or not is often adopted.

  On the other hand, the voice recognition unit 33 starts operation according to the operation start command input from the control unit 37 and acquires the voice signal z (t) output from the voice extraction unit 31, so that the voice recognition unit 33 passes through the voice extraction unit 31. Thus, the signal z (t) of the voice section is selectively acquired from the subtracting unit 23. Further, the voice recognition unit 33 acoustically analyzes the voice signal z (t) after obtaining the voice signal z (t), and extracts a feature amount (for example, a cepstrum) from the voice signal z (t). Get quantity time-series data.

  After that, the speech recognition unit 33 compares the time-series data of the feature amount with a speech pattern registered in a speech dictionary (not shown) provided in itself using a well-known technique, and the vocabulary corresponding to the speech pattern having a high degree of coincidence. Is recognized as a vocabulary spoken by the user, the recognition result is input to the control unit 37, and then the operation is stopped.

  The control unit 37 is configured to monitor the timing when the PTT switch 35 is pressed or returned, and determines that the PTT switch 35 is pressed and an operation start command signal is input from the PTT switch 35 (S100). Yes), a learning prohibition command is input to the LMS learning unit 25 of the audio canceller unit 20 (S110), and then an operation start command is input to the voice recognition unit 33 and the voice extraction unit 31, thereby the voice recognition unit. 33 and the voice extraction unit 31 are operated to start voice recognition (S120). FIG. 2 is a flowchart showing the processing operation of the control unit 37.

  After that, the control unit 37 determines whether or not the voice section is finished and the voice recognition unit 33 has completed the acquisition of the voice signal based on the operation state of the voice extraction unit 31 (S130), and the voice section is finished. When the determination is made (Yes in S130), a learning restart command is input to the LMS learning unit 25 (S140), and a recognition result is acquired from the speech recognition unit 33 (S150). Then, in order to inquire the user whether or not the recognition result is correct, the recognition result is talked back (S160).

  That is, the control unit 37 controls the voice synthesis unit 39 to cause the voice synthesis unit 39 to generate a voice signal according to the recognition result and input the voice signal to the speaker 9. The voice synthesizer 39 synthesizes a voice signal based on a voice output instruction from the controller 37 using a voice waveform stored in a waveform database (not shown) and outputs the synthesized voice signal to the speaker 9. Therefore, in S160, the recognition result is notified to the user by voice.

  Thereafter, the control unit 37 determines whether or not a recognition result confirmation signal indicating that the recognition result is correct is input from an operation switch such as the PTT switch 35 by the user's operation (S170), and the recognition result confirmation signal is input. If it is determined (Yes in S170), post-determination processing is executed (S180). On the other hand, if it is determined that the recognition result confirmation signal has not been input (No in S170), the process is terminated without executing the post-confirmation process.

  In the post-confirmation process performed in S180, the control unit 37 inputs an operation signal according to the recognition result to the car navigation device 3. Such post-determination processing uses a well-known technique and will not be described in detail.

  Next, the processing operation of the LMS learning unit 25 of the audio canceller unit 20 will be described with reference to FIG. FIG. 3 is a flowchart showing a learning process executed by the LMS learning unit 25 at the same time when the speech recognition apparatus 1 is turned on.

  When the LMS learning unit 25 starts executing the learning process, first, the LMS learning unit 25 performs initial setting on the adaptive filter 21 (S210). That is, the LMS learning unit 25 sets predetermined filter coefficients (initial values) in the adaptive filter 21.

  Thereafter, the LMS learning unit 25 uses the audio signal z (t) output from the subtracting unit 23 to calculate the coefficient w ′ according to the following equation based on the LMS method, and thereby the filter to be set to the adaptive filter 21 next. The coefficient w ′ is learned (S220).

ここで、代入する係数ｗは、既に適応フィルタ２１に設定したフィルタ係数ｗの値である。また、αは、係数ｗ’が発散するのを防止するための忘却係数であり、βは、除数がゼロになるのを防止するための正の定数である。その他μは、ステップサイズパラメータと呼ばれるものであり、フィルタ係数の学習速度に対応するパラメータである。

Here, the coefficient w to be substituted is the value of the filter coefficient w that has already been set in the adaptive filter 21. Further, α is a forgetting factor for preventing the coefficient w ′ from diverging, and β is a positive constant for preventing the divisor from becoming zero. Other μ is called a step size parameter, and is a parameter corresponding to the learning rate of the filter coefficient.

  When the calculation of the filter coefficient w ′ in S220 is completed, the LMS learning unit 25 sets the filter coefficient w ′ calculated in S220 as the new filter coefficient w in the adaptive filter 21 (S230).

  Thereafter, the LMS learning unit 25 determines whether or not a learning prohibition command is input from the control unit 37 (S240). If not input (No in S240), the LMS learning unit 25 learns by turning off the power of the device, an error, or the like. It is determined whether or not a processing end command is input from the control unit 37 (S250). If an end command is input (Yes in S250), the process ends. If no end command is input (No in S250), the process returns to S220 to learn the filter coefficient w ′. Thereafter, the filter coefficient is updated (S230).

  In S240, if it is determined that the learning prohibition command is input from the control unit 37 (Yes in S240), the LMS learning unit 25 moves the process to S260, and the learning restart command is input from the control unit 37. Determine whether or not. If a learning restart command is not input (No in S260), it is determined in S270 that an end command is input. If an end command is input (Yes in S270), the process ends. If the end command is not input (No in S270), the process returns to S260 and waits until the learning restart command is input from the control unit 37.

  If it is determined that a learning restart command has been input (Yes in S260), the process returns to S220, the filter coefficient w ′ is learned, and the filter coefficient w ′ obtained as a result is set as a new filter coefficient w. The adaptive filter 21 is set (S230).

  By repeating such an operation, the LMS learning unit 25 stops the filter coefficient learning operation from when the PTT switch 35 is pressed (turned on) until the end of the speech period, as shown in FIG. . Further, when the speech interval ends and the learning restart command is input, the filter coefficient learning is continued until the next learning prohibition command is input again. FIG. 4 is a time chart showing the operation switching timing of the LMS learning unit 25.

  The speech recognition device 1 according to the present embodiment has been described above. In this speech recognition device 1, the speech recognition unit 33 acquires a speech signal from the audio canceller unit 20 via the speech extraction unit 31 by the operation of the control unit 37. During this period, the learning of the filter coefficient by the LMS learning unit 25 is prohibited, so that the learning update of the filter coefficient w is performed when the voice uttered by the user for voice recognition is input to the microphone 5. Can be prevented.

  Therefore, according to the voice recognition apparatus 1, when the voice signal is acquired by the voice recognition unit 33, the filter coefficient is prevented from being inappropriately learned and updated due to the influence of the voice of the user input to the microphone 5. Therefore, the noise component can be accurately removed from the speech signal that is the target of speech recognition. As a result, according to the present embodiment, the accuracy of speech recognition in the speech recognition apparatus 1 can be increased, and a high speech recognition rate can be realized.

  In addition, in this embodiment, based on the audio signal z (t) output from the audio canceller unit 20, the audio extraction unit 31 determines the utterance period during which the user has made an utterance, and the audio canceller unit 20 outputs the utterance period. Since only the speech signal corresponding to the speech period is selectively input to the speech recognition unit 33, the speech signal in the noise section that does not include the utterance content by the user is It is not necessary to input the signal to the recognition unit 33, and the voice recognition unit 33 can perform accurate voice recognition without being affected by noise. In this embodiment, since the voice extraction unit 31 automatically determines the utterance period, it is convenient that the user does not input information regarding the utterance period from the operation switch.

  Further, in this embodiment, at the same time when an operation start command is input from the PTT switch 35, the voice extraction unit 31 ends detection of the voice section from that time until the voice recognition unit 33 ends the acquisition of the voice signal. Since the control unit 37 is configured to prohibit the learning of the filter coefficient by the LMS learning unit 25 during this period, it is possible to stop the learning of the filter coefficient during the user's utterance period with simple control. .

  In the above embodiment, the speech recognition apparatus 1 has been improved in performance by prohibiting the learning of filter coefficients. However, during the user's utterance period, the learning speed of the filter coefficients is slowed down, so that the comparison with the prior art is achieved. Thus, it is possible to perform voice recognition with high accuracy.

  Next, a modified speech recognition apparatus having such a configuration will be described. Note that the modified speech recognition apparatus has a configuration in which the partial processing operations of the control unit 37 and the LMS learning unit 25 are different, and the configuration of each other part in the apparatus is the same as that of the speech recognition apparatus 1 described above. Therefore, below, description of each part of the same structure as the speech recognition apparatus 1 described above will be omitted, and only the operation of the control unit 37 and the LMS learning unit 25 will be described with reference to FIGS. 5 and 6.

  FIG. 5 is a flowchart showing the processing operation of the control unit 37 in the voice recognition device according to the modification. As shown in FIG. 5, when the control unit 37 determines that the operation start command signal is input from the PTT switch 35 (Yes in S300), the audio canceller unit 20 sends a low speed learning command for slowing down the learning rate of the filter coefficient. Is input to the LMS learning unit 25 (S310), and then the speech recognition unit 33 and the speech extraction unit 31 are operated to start speech recognition (S320).

  Thereafter, the control unit 37 determines whether or not the voice section is finished and the voice signal input from the voice extraction unit 31 to the voice recognition unit 33 is completed based on the operation state of the voice extraction unit 31 (S330). If it is determined that the section has ended (Yes in S330), a normal learning command for changing the learning speed of the filter coefficient to the normal learning speed is input to the LMS learning unit 25 (S340). At the same time, a recognition result is acquired from the voice recognition unit 33 (S350). Then, in order to inquire the user whether or not the recognition result is correct, the recognition result is talked back (S360).

  Thereafter, the control unit 37 determines whether or not a recognition result confirmation signal indicating that the recognition result is correct is input from an operation switch such as the PTT switch 35 by the user's operation (S370), and the recognition result confirmation signal is input. If it is determined (Yes in S370), post-confirmation processing is executed (S380). On the other hand, if it is determined that the recognition result confirmation signal has not been input (No in S370), the process is terminated without executing the post-confirmation process.

  Next, the processing operation of the LMS learning unit 25 in the voice recognition device according to the modification will be described with reference to FIG. FIG. 6 is a flowchart showing a learning process executed by the LMS learning unit 25 of the modified example at the same time when the voice recognition apparatus is turned on.

When the learning process is started, the LMS learning unit 25 sets a predetermined predetermined filter coefficient (initial value) for the adaptive filter 21 as an initial setting and uses it in calculating the filter coefficient w ′ in S410. The parameter μ in the equation (5) is set to the initial value μ _H (μ = μ _H ).

  Thereafter, the LMS learning unit 25 uses the audio signal z (t) output from the subtraction unit 23 to calculate the coefficient w ′ according to Expression (5) based on the LMS method (S420). By this operation, the LMS learning unit 25 learns the filter coefficient w ′ to be set next in the adaptive filter 21, and sets the filter coefficient w ′ in the adaptive filter 21 as a new filter coefficient w in S 430.

Subsequently, the LMS learning unit 25 determines whether or not a low-speed learning command is input from the control unit 37 (S440). If it is determined that the low-speed learning command is input (Yes in S440), the learning speed is represented in S450. the parameter mu, sets the predetermined value _{μ L (μ = μ L)} . Incidentally, the value mu _L and the value mu _H, the relationship of inequality μ _L <μ _H is established.

As can be understood from the expression (5), when the value of the parameter μ is reduced, the amount of change in the filter coefficient w ′ can be reduced. That is, by reducing the parameter μ, the time until the filter coefficient w ′ converges can be lengthened, and the learning speed can be suppressed. LMS learning section 25, by thus setting the parameter mu Usually smaller value mu _L, with each other to lower the learning speed of the filter coefficients.

Thereafter, the control unit 37 moves the processing to S420, in accordance with Equation (5) is a mu = mu _L, and calculates the filter coefficient w ', then update the filter coefficients w (S430).
On the other hand, when determining that the low speed learning command is not input in S440 (No in S440), the control unit 37 determines whether or not the normal learning command is input from the control unit 37 in S460.

Here, the normal learning command is judged to be inputted (Yes in S460), the control unit 37, at S470, a parameter mu representing the learning speed is changed to _{μ H (μ = μ H)} . Then, the process again proceeds to S420, the filter coefficient w ′ is calculated according to the equation (5) where μ = μ _H , and then the filter coefficient w is updated (S430).

  In addition, when the control unit 37 determines No in S440 and S460, the control unit 37 determines whether an instruction to end the learning process is input from the control unit 37 in S480. If it is determined that an end command has not been input (No in S480), the process returns to S420, the filter coefficient w 'is learned, and then the filter coefficient is updated (S430). On the other hand, if it is determined that an end command has been input (Yes in S480), the learning process ends.

  As shown in FIG. 7, the LMS learning unit 25 performs such processing, so that the learning speed of the filter coefficient is reduced from when the PTT switch 35 is pressed (turned on) until the end of the speech period. Lower. Further, when the normal learning command is input after the end of the speech section, the filter coefficient is learned at the normal learning speed until the next low-speed learning command is input again. FIG. 7 is a time chart showing learning speed switching timing.

  Although the modification has been described above, in the voice recognition device of the modification, the voice recognition unit 33 extracts the voice by inputting the low speed learning command to the LMS learning unit 25 together with the operation of the voice recognition unit 33 and the voice extraction unit 31. While the audio signal is acquired from the audio canceller unit 20 via the unit 31, the LMS learning unit 25 is learning the filter coefficient at a lower learning speed than usual. It is possible to suppress the influence of the user's voice, and to suppress erroneous learning of the filter coefficient in the LMS learning unit 25.

  As a result, according to the speech recognition apparatus of the modification, the LMS learning unit 25 can appropriately learn the filter coefficient, and the accuracy of noise removal in the audio canceller unit 20 can be improved. Therefore, according to the modified example, it is possible to provide a voice recognition device capable of voice recognition with high accuracy.

  In the modification, the LMS learning unit is operated at a normal learning speed immediately after the voice recognition unit 33 acquires the voice signal (that is, immediately after the voice section is finished), in addition to when the voice recognition unit 33 is not operating. 25, the filter coefficient learning is performed. Therefore, even when the operation start command signal is continuously input from the PTT switch 35 and the voice recognition unit 33 is operated, the audio canceller unit 20 performs appropriate noise. Removal is possible.

  In addition, in the modified example, since the voice extraction unit 31 selectively inputs only the voice signal corresponding to the utterance period in which the utterance is made by the user to the voice recognition unit 33, the utterance content by the user is not included. It is not necessary to input the speech signal in the noise section to the speech recognition unit 33, and the speech recognition unit 33 can perform accurate speech recognition without being affected by noise.

  In the modified speech recognition apparatus, since the operation start command signal is input from the PTT switch 35 and the LMS learning unit 25 learns the filter coefficient at a low learning speed at the same time, a simple method is adopted. It is possible to reliably suppress erroneous learning of filter coefficients by the control.

  In addition, according to the speech recognition apparatus of the above embodiment, since the volume of music or the like reproduced from the speaker 9 is not adjusted by the operation of the in-vehicle audio device 7, speech recognition can be performed with high accuracy. Conventional problems such as dissatisfaction with users due to adjustments can be solved.

  Although the embodiments of the present invention have been described above, the noise removal signal generating means of the present invention corresponds to the adaptive filter 21 of this embodiment, and the noise removal means of the present invention corresponds to the subtracting unit 23. The coefficient updating unit corresponds to the LMS learning unit 25, and the speech recognition unit corresponds to the speech recognition unit 33 that acquires the speech signal z (t) in the speech section and performs speech recognition.

In addition, the learning speed switching means is realized by an operation in which the control unit 37 inputs a low speed learning command and a normal learning command to the LMS learning unit 25 at a timing according to the processing shown in FIG. Note that the operation in which the learning speed switching means causes the coefficient updating means to learn the filter coefficient at the second learning speed is the parameter μ = μ _L corresponding to the second learning speed in the LMS learning unit 25 in this embodiment. This is realized by the operation of calculating the filter coefficient w ′. In addition, the operation in which the learning speed switching unit causes the coefficient updating unit to learn the filter coefficient at the first learning speed causes the LMS learning unit 25 to use the parameter μ = μ _H corresponding to the first learning speed in the present embodiment. This is realized by the operation of calculating the filter coefficient w ′.

Moreover, the speech recognition apparatus of the present invention is not limited to the above-described embodiments, and can take various forms.
For example, the control unit 37 is configured to prohibit the operation of the LMS learning unit 25 or reduce the learning rate of the filter coefficient in the LMS learning unit 25 at least during the acquisition period of the audio signal by the audio recognition unit 33. And the learning of the filter coefficient by the LMS learning unit 25 may be prohibited until the speech recognition in the speech recognition unit 33 is completed and the result of speech recognition is obtained. The LMS learning unit 25 may be operated.

  In the above embodiment, an example in which the present invention is applied to a speech recognition apparatus that employs the LMS method, specifically, the Normalized LMS (NLMS) algorithm, as a filter coefficient learning method has been described. The present invention may be applied to a speech recognition apparatus that learns filter coefficients. As a learning method to which the present invention can be applied, in addition to the adaptive algorithm described above, for example, complex LMS algorithm, FastLMS (FLMS) algorithm, projection algorithm, RLS (Recursive Least Square) algorithm, SHARF (Simple Hyperstable Adaptive Recursive). ) Algorithm, adaptive filter using DCT (Discrete Cosine Transform), SAN (Single Frequency Adaptive Notch) filter, neural network, genetic algorithm, and the like.

It is a block diagram showing schematic structure of the speech recognition apparatus 1 of a present Example. 7 is a flowchart illustrating a processing operation of a control unit 37. It is a flowchart showing the learning process which the LMS learning part 25 performs. 3 is a time chart showing operation switching timing of an LMS learning unit 25. It is a flowchart showing the processing operation in the control part 37 of a modification. It is a flowchart showing the learning process which the LMS learning part 25 of a modification performs. It is a time chart showing the switching timing of learning speed.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 ... Voice recognition apparatus, 3 ... Car navigation apparatus, 5 ... Microphone, 7 ... Car-mounted audio equipment, 9 ... Speaker, 11, 13 ... ADC, 20 ... Audio canceller part, 21 ... Adaptive filter, 23 ... Subtraction part, 25 ... LMS learning unit, 31 ... speech extraction unit, 33 ... speech recognition unit, 35 ... PTT switch, 37 ... control unit, 39 ... speech synthesis unit

Claims (4)

A noise removal signal generating means for generating a noise removal signal for removing noise by filtering a reference signal input from a noise source according to a preset filter coefficient;
Using the noise removal signal generated by the noise removal signal generation means, a noise removal means for removing a noise component included in a voice signal input from a microphone and outputting the voice signal after the noise removal;
Coefficient updating means for learning filter coefficients to be set in the noise removal signal generating means based on the audio signal output from the noise removing means, and setting the filter coefficients obtained as a result of learning in the noise removal signal generating means When,
When an operation start command is input from the outside, a voice recognition unit that acquires a voice signal output from the noise removal unit for a predetermined period and recognizes a voice input to the microphone based on the voice signal;
When the voice recognition means is not operating, the coefficient update means learns the filter coefficient at the first learning speed, and the voice recognition means acquires the voice signal output from the noise removal means. Learning speed switching means for causing the coefficient update means to learn the filter coefficient at a second learning speed lower than the first learning speed;
A speech recognition apparatus comprising:   The learning speed switching means is configured to apply the coefficient at the second learning speed for a period until the voice recognition means ends acquisition of the voice signal at the same time when the operation start command is input to the voice recognition means. The speech recognition apparatus according to claim 1, wherein the update means is made to learn the filter coefficient.   2. The coefficient updating unit learns a filter coefficient to be set in the noise removal signal generation unit by using an LMS method based on an audio signal output from the noise removal unit. The speech recognition apparatus according to 2.   The speech recognition apparatus according to claim 1, wherein the noise source is an audio device.

Images