JPWO2011043380A1 - Speech recognition apparatus and speech recognition method - Google Patents

Abstract

The speech recognition apparatus includes an information processing unit (10) that performs processing for deriving a word corresponding to an external speech signal, an output unit (20) that sequentially outputs words, and use of resources of the information processing unit A usage rate acquisition unit (30) for acquiring a rate. The information processing unit (10) includes an information generation unit (12) that generates predetermined information based on an audio signal, a temporary storage unit (14) that temporarily stores predetermined information, and a predetermined information. A word derivation unit (16) for deriving a word corresponding to the speech signal and performing speech recognition, and outputting the word to the output unit (20), and processing performed by the word derivation unit (16) are used rate acquisition unit (30) And a control unit (18) for suspending and resuming based on the usage rate acquired by the above. Since the processing performed by the word deriving unit (16) is interrupted and restarted based on the usage rate, inconvenience caused by an excessive usage rate of resources can be suppressed.

Description

  The present invention relates to a speech recognition apparatus and a speech recognition method, and more particularly to a speech recognition apparatus and a speech recognition method that sequentially output results recognized when speech is continuously input.

  In recent years, techniques for generating subtitles for TV programs, creating meeting minutes, and translating speech using a speech recognition system that recognizes speech input by speech and converts it into a symbol string such as text are widely known. ing.

  However, in a normal speech recognition system, since the most probable result cannot be output unless the utterance is once finished, there is a time lag before the result recognized from the actual utterance is output. For this reason, it has been difficult to use in situations where real-time speech recognition is required, such as when subtitles are displayed during live broadcast of a TV program, or when used in a conversation between different languages in combination with a translation system.

  In order to solve such a problem, there has been proposed a continuous speech recognition apparatus that performs speech recognition at regular intervals even during speech and outputs the results sequentially (for example, see Patent Document 1). In this continuous speech recognition device, the most probable words corresponding to the input continuous speech are selected at regular intervals, and the words that can be output stably are extracted and output, so that they are stable in real time. High voice recognition is possible.

Japanese Patent No. 3834169

L. Mangu, et al., “Finding consensus in speech recognition: word error minimization and other applications of confusion network,” Computer Speech and Language, vol.14, no.4, pp.373-400, 2000.

However, in general, processing for performing speech recognition uses a large amount of resources such as a CPU and a memory, which may cause a shortage of resources. When such a resource shortage occurs, if the speech recognition processing is performed sequentially as in the continuous speech recognition device described in Patent Document 1, each processing of the speech recognition processing stagnates due to the lack of resources, and speech input There is a risk of inconvenience such as spilling over.
A situation where such a resource shortage occurs is likely to occur when speech recognition is performed by a small device such as a mobile phone in which it is difficult to mount a relatively high-performance resource. Even when a device equipped with relatively high-performance resources is used, it is likely to occur when speech recognition is performed in combination with a system with a large amount of processing such as a speech synthesis system or an automatic translation system.

  The present invention has been made in order to solve the above-described problem, and has made it possible to suppress the loss of an external audio signal even when the resource usage rate becomes relatively large. An object of the present invention is to provide a voice recognition device and a voice recognition method.

  The speech recognition apparatus according to the present invention includes an information processing unit that performs processing for deriving a word corresponding to an external speech signal, an output unit that sequentially outputs the word, and a resource usage rate of the information processing unit A usage rate acquisition unit for acquiring the information, and the information processing unit generates an information generation unit based on the audio signal, and temporarily stores the predetermined information from the information generation unit. A storage unit, a word derivation unit that derives a word corresponding to a speech signal based on predetermined information temporarily stored by the temporary storage unit, performs speech recognition, and outputs the word to the output unit; And a control unit that suspends and resumes the processing performed by the word deriving unit based on the usage rate acquired by the usage rate acquisition unit.

  The speech recognition method according to the present invention includes a step of generating predetermined information based on an external audio signal, a step of temporarily storing the predetermined information, and a voice based on the temporarily stored predetermined information. Deriving a word corresponding to the signal and performing speech recognition, outputting the word, deriving a word corresponding to the speech signal and obtaining a usage rate of resources for performing speech recognition, and performing speech recognition And a step of suspending and resuming processing based on the usage rate acquired.

According to the speech recognition apparatus and speech recognition method of the present invention, the process of deriving a word corresponding to an external speech signal is interrupted and restarted based on the resource usage rate of the information processing unit. It is possible to suppress inconvenience resulting from an excessive rate, that is, a delay in processing for generating predetermined information based on an external audio signal.
Moreover, since the generated predetermined information can be temporarily stored, even when the process of deriving a word corresponding to an external audio signal is interrupted, the generated predetermined information is lost without being processed. Can be suppressed.

FIG. 1 is a block diagram showing a configuration example of a speech recognition apparatus according to Embodiment 1 of the present invention. FIG. 2 is a flowchart showing an example of processing executed by the speech recognition apparatus according to Embodiment 1 of the present invention. FIG. 3 is a block diagram showing a configuration example of a speech recognition apparatus according to Embodiment 2 of the present invention. FIG. 4 is a flowchart showing an example of processing executed by the speech recognition apparatus according to Embodiment 2 of the present invention. FIG. 5 is a flowchart showing an example of processing for setting the flag F1 executed by the control unit of the speech recognition apparatus according to Embodiment 2 of the present invention. FIG. 6 is a flowchart showing an example of processing executed by the speech recognition apparatus according to Embodiment 3 of the present invention. FIG. 7 is a block diagram showing a configuration example of a speech recognition apparatus according to a modification of the present invention.

[Embodiment 1]
Hereinafter, the speech recognition apparatus according to Embodiment 1 of the present invention will be described in detail with reference to the drawings.
The speech recognition apparatus according to Embodiment 1 of the present invention derives words corresponding to external speech signals and sequentially outputs them.

First, the configuration of the speech recognition apparatus according to Embodiment 1 of the present invention will be described with reference to FIG.
As shown in FIG. 1, the speech recognition apparatus according to Embodiment 1 of the present invention performs information processing based on a speech signal from the outside, and information processed by the information processing unit 10. Output unit 20 and a usage rate acquisition unit 30 that acquires the usage rate of resources constituting the information processing unit 10.
The information processing unit 10 includes an information generation unit 12 that generates predetermined information based on an external audio signal, a temporary storage unit 14 that temporarily stores predetermined information generated by the information generation unit 12, and a temporary storage unit The word deriving unit 16 for deriving predetermined information from 14 and deriving a word corresponding to an external audio signal, and the processing performed by the word deriving unit 16 based on the signal from the usage rate acquiring unit 30 are interrupted or restarted. The control unit 18 is included.
The output unit 20 sequentially outputs the words derived by the word deriving unit 16 as sound or video.
The usage rate acquisition unit 30 acquires the usage rate of resources such as a CPU and a memory (not shown) included in the information processing unit 10 every predetermined time (for example, every several tens of msec or every several hundreds of msec). To be sent.
Each function of the speech recognition apparatus according to the first embodiment of the present invention described above is obtained by reading a program recorded on a recording medium by a computer having a CPU, a memory, and various interfaces, so that hardware resources of the computer are read. Can be realized through collaboration between software and software.

  Next, the operation of the speech recognition apparatus according to Embodiment 1 of the present invention will be described with reference to FIG.

First, the information generation unit 12 inputs an audio signal from the outside and a resource usage rate U from the usage rate acquisition unit 30 (step S100).
Next, predetermined information is generated based on the input audio signal, and the predetermined information is stored in the temporary storage unit 14 (step S110).
Subsequently, the control unit 18 determines whether or not the resource usage rate U acquired by the usage rate acquisition unit 30 is normal (step S120).

When it is determined that the usage rate U is smaller than an arbitrary predetermined value and within the normal range (step S120: YES), the word deriving unit 16 generates an audio signal based on the predetermined information stored in the temporary storage unit 14. The word corresponding to is derived (step S130), and the process of outputting the word by the output unit 20 is performed (step S140).
Subsequently, it is determined whether or not the process needs to be continued (step S150). When the audio signal is continuously input or when there is an unprocessed audio signal, the process returns to step S100 and the processes of steps S100 to S140 are repeated (step S150: NO), and the input of the audio signal is completed. And after confirming that there is no unprocessed audio | voice signal (step S150: YES), a series of processes are complete | finished.

  On the other hand, when it is determined that the usage rate U is equal to or greater than an arbitrary predetermined value and is abnormal (step S120: NO), the word deriving unit 16 in step S130 does not execute the process of deriving a word corresponding to the audio signal. Then, the process returns to step S100. For this reason, the process of steps S100 and S110 in which a voice signal is input again from the outside, predetermined information is generated and temporarily stored in the temporary storage unit 14 is repeated, and the usage rate U is smaller than an arbitrary predetermined value and normal. (Step S120: YES), the word corresponding to the external audio signal is derived based on the predetermined information temporarily stored in the temporary storage unit 16, and the output unit The processing output from 20 is executed (steps S130 and S140).

  Note that, when the usage rate U is equal to or greater than an arbitrary predetermined value and is determined to be abnormal, the processing of steps S100 to S120 is not performed, and the word derivation is particularly performed without executing the processing of steps S130 and S140. This is because the resource usage rate U may be further increased when the unit 16 executes the process of step S130. Inconvenience caused by further increasing the usage rate U by interrupting these processes, that is, a step of inputting an audio signal from the outside, generating predetermined information, and temporarily storing it in the temporary storage unit 14 It is possible to prevent the processing of S100 and S110 from being delayed. In addition, since the predetermined information generated by the process of step S110 is temporarily stored in the temporary storage unit 14, the generated information is prevented from being lost without being processed, so the usage rate U becomes normal. Even after waiting for this, the processing of steps S130 and S140 can be executed.

As described above, according to the speech recognition apparatus according to Embodiment 1 of the present invention, the process of deriving a word corresponding to an external speech signal is interrupted based on the resource usage rate U of the information processing unit 10. Therefore, it is possible to suppress the inconvenience caused by the excessive usage rate U of the resource, that is, the inconvenience that the process of generating the predetermined information by inputting an external audio signal is delayed.
Moreover, since the generated predetermined information can be temporarily stored, the generated predetermined information remains unprocessed even when the process of deriving a word corresponding to the external audio signal is interrupted. Can be suppressed.

[Embodiment 2]
Next, a speech recognition apparatus according to Embodiment 2 of the present invention will be described in detail with reference to the drawings.
The speech recognition apparatus according to Embodiment 2 of the present invention also derives and sequentially outputs words corresponding to external speech signals.

First, the configuration of the speech recognition apparatus according to Embodiment 2 of the present invention will be described with reference to FIG.
As shown in FIG. 3, the speech recognition apparatus according to Embodiment 2 of the present invention performs information processing based on an external audio signal, information processing unit 110 that performs predetermined information processing, and information processed by information processing unit 110. Output unit 20 that outputs information, a usage rate acquisition unit 130 that acquires the usage rate of a CPU (not shown) included in the information processing unit 110, an acoustic model 140 that stores data related to phonemes and syllables, and a recognition dictionary that stores data related to vocabulary 142. Here, as an external audio signal, an audio signal generated by a microphone (not shown) can be input, and an audio signal generated at a remote place can also be input via a network.
The information processing unit 110 analyzes an external audio signal and generates a feature amount arranged in time series, and a temporary storage unit that temporarily stores the feature amount generated by the feature amount generation unit 112 114, a word deriving unit 116 that derives a feature value from the temporary storage unit 114 and compares the feature value with data stored in the acoustic model 140 and the recognition dictionary 142 to derive a word corresponding to an external audio signal; A control unit 118 is included that interrupts or restarts the processing performed by the word deriving unit 116 based on a signal from the usage rate acquisition unit 130.
In addition, the word deriving unit 116 obtains a plurality of word hypotheses in association with the likelihood, a word hypothesis acquisition unit 116a, and externally based on the likelihood associated with each of the word hypotheses acquired by the word hypothesis acquisition unit 116a. Includes a word selection unit 116b that selects one word hypothesis as a word corresponding to the voice signal.
The output unit 120 sequentially outputs the words selected by the word selection unit 116b on the display.
The usage rate acquisition unit 130 acquires the CPU usage rate U of the information processing unit 110 and transmits it to the control unit 118 every predetermined time (for example, every several tens of msec or every few hundreds of msec).
Each function of the speech recognition apparatus according to the second embodiment of the present invention described above is obtained by reading a program recorded on a recording medium by a computer having a CPU, a memory, and various interfaces, so that the hardware resources of the computer Can be realized through collaboration between software and software.

  Next, the operation of the speech recognition apparatus according to Embodiment 2 of the present invention will be described with reference to FIG.

First, the CPU usage rate U is input from the outside from the audio signal and usage rate acquisition unit 130 (step S200).
Next, the input audio signal is analyzed to generate a feature amount sequence arranged in time series, and is stored in the temporary storage unit 114 (step S210). Specifically, the input audio signal is divided into frames every several tens of msec, and a feature amount such as a cepstrum is generated for each frame and stored in the temporary storage unit 114.

Subsequently, the control unit 118 sets a flag F1 indicating whether or not the usage rate U is within a normal range based on the input usage rate U (step S220). Here, the flag F1 is set to a value of 0 in the initial state, and is set to a value of 1 when it is determined that the usage rate U is excessive.
The process for setting the flag F1 will be described with reference to FIG. The usage rate U is compared with a predetermined threshold value Ulow and a threshold value Uhi greater than the threshold value Ulow (step S221), and when the usage rate U is greater than or equal to the threshold value Ulow and less than the threshold value Uhi, the value of the flag F1 is maintained in the existing state. To do. On the other hand, when the usage rate U is equal to or greater than the threshold value Uhi, it is determined that the usage rate U is excessive and a value 1 is set in the flag F1 (step S222). Conversely, when the usage rate U is less than the threshold Ulow, it is determined that there is a margin in the usage rate U and a value 0 is set in the flag F1 (step S223). Here, for example, a value such as 80% or 90% can be used as the threshold value Uhi, and a value such as 50% or 60% can be used as the threshold value Ulow. Note that the threshold Uhi serving as a reference for setting the value 1 to the flag F1 and the threshold Ulow serving as a reference for setting the value 0 to the flag F1 are set to values different from each other. This is because by giving hysteresis to the relationship between U and the value of the flag F1, the value of the flag F1 is prevented from being frequently changed between “0” and “1”.

Next, the control unit 118 determines whether or not the value of the flag F1 is 0 (step S230).
When the flag F1 is 0 (step S230: YES), processing for deriving a word corresponding to the input external voice is executed.

First, two word hypotheses are acquired in descending order of the likelihood P representing the probability, and a value 1 is set to the flag F2 indicating whether or not the process of deriving the word corresponding to the external speech is being executed. (Step S240).
Specifically, the word hypothesis acquisition unit 116 a and the likelihood setting unit 116 b extract the feature amount series arranged in a time series stored in the temporary storage unit 114 and store the feature amount series and the acoustic model 140. The distance between the phoneme and the syllable data is calculated, and based on the calculation result, the two word hypotheses are known in order from the vocabulary stored in the recognition dictionary 142 in descending order of the likelihood P. It is acquired using a technique such as a hidden Markov model (HMM), and is temporarily stored in a memory (not shown). Note that a technique for calculating a distance and obtaining a word hypothesis having a high likelihood P in speech recognition is a well-known technique, and thus a detailed description thereof is omitted (for example, see Non-Patent Document 1).

  Next, since the likelihood P is set for each of the two word hypotheses acquired in step S240, from the likelihood P1 of the word hypothesis having the maximum likelihood P among the two word hypotheses, A likelihood difference ΔP obtained by subtracting the likelihood P2 of the word hypothesis with the second largest likelihood P is calculated (step S250).

  Thereafter, the likelihood difference ΔP is compared with a threshold value ΔPthre (step S260). Since the threshold value ΔPthr is set to a relatively large value ΔP0 as an initial value, the likelihood difference ΔP is usually less than the threshold value ΔPthr (step S260: NO), and the threshold value ΔPthr is set to Δ (for example, an initial value). A value obtained by subtracting one hundredths or thousands of values from ΔP0 is reset to the threshold ΔPthre (step S262), and the process returns to step S200. Here, the value obtained by subtracting Δ from the threshold value ΔPthre is newly set to the threshold value ΔPthre because the condition for deriving the word corresponding to the external speech with time elapses, and the unprocessed word is not determined. This is to prevent the feature amount from increasing unnecessarily.

  In this manner, while repeating the processes of steps S200 to S262 described above, the feature amount extracted from the temporary storage unit 114 increases, and the likelihood P1 of the word hypothesis having the maximum likelihood P also increases. Since the threshold value ΔPthr is set to be gradually smaller in the process of S262, the probability that the likelihood difference ΔP becomes the threshold value Pthre gradually increases. As a result, when the likelihood difference ΔP is equal to or greater than the threshold value ΔPthre (step S260: YES), the word hypothesis having the maximum likelihood P is regarded as a word corresponding to the input speech signal and output by the output unit 120. Then, the value 0 is reset to the flag F2 set to the value 1 in the process of step S240, and the initial value ΔP0 is reset to the threshold value ΔPthr changed in the process of step S262 (step S270). At this time, the memory can be effectively used by deleting information such as the feature amount generated in the processing of steps S200 to S270 and the acquired word hypothesis from the memory.

  Subsequently, it is determined whether or not the process needs to be continued based on whether or not the audio signal is continuously input and whether or not there is an unprocessed audio signal (step S280). When the audio signal is continuously input or when there is an unprocessed audio signal, the processing of steps S200 to S270 is repeated (step S280: NO), the input of the audio signal is completed, and the unprocessed audio signal is processed. After confirming that there is no audio signal (step S280: YES), the series of processing is terminated.

On the other hand, when the flag F1 indicating whether or not the usage rate U is within the normal range is a value 1 indicating that the usage rate U is excessive (step S230: NO), a word corresponding to the external voice is further generated. It is determined whether or not the flag F2 indicating whether or not the process for deriving is being executed is 0 (step S240).
When the flag F2 is 1, that is, when the process of deriving a word is being executed (step S232: YES), the process of deriving a word corresponding to an external audio signal after step S240 is continued.
Then, after confirming that the flag F2 has a value of 0, that is, confirming that a word corresponding to an external audio signal has been output in the process of step S280 (step S232: NO), step The processing returns to step S200 without executing the processing after S240.
For this reason, an external audio signal is input, a feature value is generated and temporarily stored in the temporary storage unit 114, and the processing of steps S200 to S220 for setting the flag F1 based on the usage rate U is repeated and used. After confirming that the rate U is less than the predetermined value Ulow and the flag F1 is set to 0 (step S230: YES), the processing from step S240 is executed.

The reason why the processing after step S240 is not executed when the flag F1 is 1 and the flag F2 is 0 is based on the following two reasons. The first reason is that, particularly, when the word hypothesis acquisition unit 116a executes the process of step S240, the usage rate U is further increased. The second reason is that when the process of deriving a word corresponding to an external audio signal is being executed, the acquired information regarding the two word hypotheses cannot be erased from the memory. This is because it is desirable to stop the process once deriving from the process.
Discontinuing these processes causes an inconvenience caused by an excessive usage rate U, that is, an external audio signal is input, a feature amount is generated, and temporarily stored in the temporary storage unit 114 (step S200). , S210 can be prevented from being delayed. In addition, since the feature amount generated by the process of step S210 is temporarily stored in the temporary storage unit 114 and is prevented from disappearing without being processed, the normal usage rate U is less than the threshold value Low. Even after waiting for the value 0 to be set in the flag F1, the processes in and after step S240 can be executed.

As described above, according to the speech recognition apparatus according to Embodiment 2 of the present invention, when it is determined that the CPU usage rate U of the information processing unit 110 is excessive, a word corresponding to an external speech signal is derived. Since the process is interrupted or restarted, the inconvenience caused by the excessive usage rate U of the CPU, that is, the inconvenience that the process of generating the feature amount by inputting the external audio signal is delayed is suppressed. it can.
Also, when the process of deriving a word corresponding to an external audio signal is being executed, the process of deriving a word is not interrupted even when the CPU usage rate U is determined to be excessive. It is possible to suppress inconvenience caused by interrupting the process of deriving the word while storing the information in the memory, that is, the excessive usage rate of the memory.
In addition, since the generated feature values can be temporarily stored, even when the process of deriving a word corresponding to an external audio signal is interrupted, the feature values are lost without being processed. Can be suppressed.

[Embodiment 3]
Next, a speech recognition apparatus according to Embodiment 3 of the present invention will be described in detail with reference to the drawings.
The speech recognition apparatus according to Embodiment 3 of the present invention also derives and sequentially outputs words corresponding to external speech signals. Since the constituent elements are common to the constituent elements of the speech recognition apparatus according to the second embodiment, the same reference numerals are used and detailed description thereof is omitted.

The operation of the speech recognition apparatus according to Embodiment 3 of the present invention will be described with reference to FIG.
Since the processes in steps S300 to S330 for generating feature amounts after inputting an external audio signal are the same as the processes in steps S200 to S230 described in the second embodiment of the present invention, detailed description is omitted. To do. On the other hand, the process of steps S340 to S370 for acquiring a plurality of word hypotheses, selecting and outputting one word hypothesis from the acquired plurality of word hypotheses is the step described in the second embodiment of the present invention. Since the processing is different from the processing of S240 to S270, the contents will be described below.

  In the speech recognition apparatus according to Embodiment 2 of the present invention, at least two word hypotheses may be acquired in descending order of likelihood, but more in the speech recognition apparatus according to Embodiment 3 of the present invention ( For example, 5 or 10 word hypotheses are preferably acquired in association with the likelihood P (step S340).

After acquiring a plurality of word hypotheses in association with the likelihood P in step S340, the word selection unit 116b selects a group of words having the same meaning as one word hypothesis having the maximum likelihood P among these word hypotheses. A word hypothesis is extracted, and a hypothesis occupancy H that is a ratio of the extracted group of word hypotheses to the total number of word hypotheses is calculated (step S350).
The hypothesis occupation rate H will be described using a specific example.
For example, as a result of generating a feature value from an input voice signal, it is determined that a voice consisting of “ki”, “f”, and “b” is inputted, and “machine” and “instrument” are ordered in descending order of likelihood. Consider a case where five word hypotheses, “opportunity”, “strange”, and “underworld” are acquired. In this case, the word hypothesis “instrument” is synonymous with the word hypothesis “machine” having the maximum likelihood P. Therefore, the total number of word hypotheses is five, whereas the number of a group of word hypotheses having the same meaning as the word hypothesis having the maximum likelihood P is two, so the hypothesis occupancy H is 2 / 5 or 40%.
As another example, as a result of generating a feature amount from the input speech signal and analyzing the speech, the three word hypotheses “color”, “color”, and “color” are ordered in descending order of likelihood. Consider the case of acquisition. In this case, the word hypothesis “color” is synonymous with the word hypothesis “color” having the maximum likelihood. For this reason, the total number of word hypotheses is three, whereas the number of a group of word hypotheses having the same meaning as the word hypothesis having the maximum likelihood P is two, so the hypothesis occupancy H is 2 / 3, ie 60%.
Whether or not one word hypothesis and another word hypothesis have the same meaning is stored in the recognition dictionary 142 as synonymous word data for each word, and is determined by referring to this data. What should I do?

  Next, the hypothesis occupancy H is compared with a threshold Hthre (step S360). Since the threshold value Hthre is set to a relatively large value H0 (for example, 50%, 60%, etc.) as an initial value, the hypothesis occupancy Hthre is usually less than the threshold value Hthre (step S360: NO), and Δ from the threshold value Hthre A value obtained by subtracting (for example, one hundredth or one thousandth of the initial value H0) is reset to the threshold value Hthre (step S362), and the process returns to step S300. Here, the value obtained by subtracting Δ from the threshold value Hthre is newly set as the threshold value Hthre because the condition for deriving the word corresponding to the external speech with the passage of time is relaxed, and the unprocessed word is not determined. This is to prevent the feature amount from increasing unnecessarily.

  Thus, as the process of steps S300 to S362 is repeated several times, the threshold value Hthre is gradually set smaller in the process of step S362, so the probability that the hypothesis occupancy H becomes equal to or higher than the threshold value Hthre gradually increases. As a result, when the hypothesis occupancy H is equal to or higher than the threshold Hthre (step S360: YES), the word hypothesis having the maximum likelihood P is regarded as a word corresponding to the input speech signal and output by the output unit 120. Then, the value 0 is reset to the flag F2 set to the value 1 in the process of step S340, and the initial value H0 is reset to the threshold value Hthre changed in the process of step S362 (step S370). At this time, the feature amount generated in the process of step S310 and the information such as the word hypothesis acquired in the process of step S340 are deleted from the memory, so that the memory can be effectively used.

  In the speech recognition apparatus according to the third exemplary embodiment of the present invention, the word corresponding to the speech signal input from the outside as one word hypothesis among the acquired word hypotheses by using the concept of hypothesis occupancy H Is output. This is because when the hypothesis occupancy H is relatively large, that is, when the acquired word hypothesis is occupied by a word hypothesis having the same meaning as the word hypothesis with the maximum likelihood P, the word with the maximum likelihood P This is because it is rare that a problem occurs even if the hypothesis is regarded as a word corresponding to an external audio signal. As a result, even when the difference between the word hypothesis with the maximum likelihood P and the likelihood of the word hypothesis with the second largest likelihood P is small, the word corresponding to the external speech signal can be determined early. In this way, when it is difficult to determine the word hypothesis corresponding to the external audio signal, the process of step S370 for outputting the word corresponding to the external audio signal even though the usage rate U of the CPU is excessive. It is possible to prevent the CPU usage rate U from being further increased due to the later execution time of the process and the later suspension of the processes in steps S340 to S380.

  Subsequently, it is determined whether or not the processing needs to be continued based on whether or not the audio signal is continuously input and whether or not there is an unprocessed audio signal (step S380). When the audio signal is continuously input or when there is an unprocessed audio signal, the processes of steps S300 to S370 are repeated (step S380: NO), the input of the audio signal is completed, and the unprocessed audio signal is processed. After confirming that there is no audio signal (step S380: YES), the series of processing is terminated.

As described above, also by the speech recognition apparatus according to Embodiment 3 of the present invention, the same effects as those of the speech recognition apparatus according to Embodiment 2 of the present invention can be obtained.
Further, since a word corresponding to an external speech signal is derived and output using the concept of hypothesis occupancy H, the likelihood of the word hypothesis with the maximum likelihood P and the word with the second largest likelihood P Even when the difference from the hypothesis likelihood is small, it is possible to derive and output a word corresponding to an external audio signal at a relatively early stage.
For this reason, although the usage rate U of the CPU is excessive, the timing of executing the process of step S370 for outputting a word corresponding to the external audio signal is delayed, and the processes of steps S340 to S380 are interrupted. It can be suppressed that the CPU usage rate U is further increased due to the late timing.

[Modification]
In the voice recognition device according to the second and third embodiments of the present invention described above, the acoustic model 140 and the recognition dictionary 142 are used to derive words corresponding to an external voice signal. A language model may be used in combination in order to consider linguistic accuracy.

  Moreover, in the speech recognition apparatus according to the second and third embodiments of the present invention described above, the usage rate U has been described as the usage rate of the CPU, but may be the usage rate of the memory.

  Furthermore, in the speech recognition apparatus according to the first to third embodiments of the present invention described above, the words derived by the word deriving units 16 and 116 are described as being sequentially output to the output units 20 and 120, but FIG. As shown in FIG. 7, the image processing apparatus may further include a translation unit 219, which translates the document into another language using the translation unit 219 and sequentially outputs the output to the output unit 220.

  Moreover, although Embodiment 1-3 of this invention mentioned above demonstrated as a form of the speech recognition apparatus, it may be the form of a speech recognition method, and is good also as a form of the computer-readable recording medium which recorded the program.

  The present invention is not limited to the embodiment described above, and can be implemented in various forms without departing from the gist of the present invention.

  This application claims the priority on the basis of Japanese application 2009-235302 for which it applied on October 9, 2009, and takes in those the indications of all here.

  The present invention can be used in the speech recognition device manufacturing industry.

  DESCRIPTION OF SYMBOLS 10 ... Information processing part, 12 ... Information generation part, 14 ... Temporary storage part, 16 ... Word derivation part, 18 ... Control part, 20 ... Output part, 30 ... Usage rate acquisition part.

Claims (13)

An information processing unit that performs processing for deriving a word corresponding to an external audio signal;
An output unit that sequentially outputs words;
A usage rate acquisition unit that acquires the usage rate of the resources of the information processing unit,
The information processing unit
An information generator that generates predetermined information based on the audio signal;
A temporary storage unit that temporarily stores predetermined information from the information generation unit;
A word deriving unit for deriving a word corresponding to an audio signal based on predetermined information temporarily stored by the temporary storage unit, performing speech recognition, and outputting the word to the output unit;
A speech recognition apparatus comprising: a control unit that suspends and resumes processing performed by the word derivation unit based on the usage rate acquired by the usage rate acquisition unit. The speech recognition apparatus according to claim 1, wherein the resource is at least one of a CPU and a memory. The control unit interrupts the processing performed by the word derivation unit when the usage rate acquired by the usage rate acquisition unit is equal to or greater than a first predetermined value, and a second usage rate is smaller than the first predetermined value. The speech recognition apparatus according to claim 1, wherein the processing performed by the word derivation unit is resumed after waiting for the value to become less than a predetermined value. The control unit does not interrupt the process performed by the word deriving unit until the word deriving unit completes the process of deriving a word when the word deriving unit performs the process of deriving a word corresponding to the audio signal. The speech recognition apparatus according to claim 1. The word derivation unit
A word hypothesis acquisition unit that acquires a plurality of word hypotheses corresponding to a speech signal input based on predetermined information temporarily stored by the temporary storage unit in association with a likelihood representing the probability;
The speech recognition according to claim 1, further comprising: a word selection unit that selects one word hypothesis as a word corresponding to the speech signal based on a likelihood set for each of the plurality of word hypotheses. apparatus. The word selection unit selects two word hypotheses from a plurality of word hypotheses having the highest likelihood, and if the difference between the likelihoods set for each is less than a threshold, the word selection corresponding to the speech signal is forgotten. The speech recognition apparatus according to claim 5, wherein when the difference between the likelihoods set for each is equal to or greater than a threshold, the word hypothesis having the maximum likelihood is selected as a word corresponding to the speech signal. The speech recognition apparatus according to claim 6, wherein the word selection unit sets a threshold value as the elapsed time from the start of the process of recognizing a word corresponding to the speech signal becomes longer. The word selection unit selects a word hypothesis having the maximum likelihood from a plurality of word hypotheses, and extracts a group of word hypotheses having the same meaning as the selected word hypothesis from the plurality of word hypotheses. When the hypothesis occupancy ratio, which is the ratio of the extracted group of word hypotheses to the total number of the plurality of word hypotheses, is less than a predetermined occupancy ratio, the selected word hypothesis is selected as a word corresponding to the speech signal. 6. The speech according to claim 5, wherein the selected word hypothesis is selected as a word corresponding to the speech signal when the hypothesis occupancy of the extracted group of word hypotheses is greater than or equal to a predetermined occupancy. Recognition device. The speech recognition apparatus according to claim 8, wherein the word selection unit sets the predetermined occupancy rate as the elapsed time from the start of the process of deriving a word corresponding to the speech signal increases. The speech recognition apparatus according to claim 1, wherein the information generation unit generates a feature amount representing a feature of the speech signal based on the speech signal. The said information processing part is further provided with the translation part which translates the word selected by the said word selection part, and outputs this translated word to the said output part sequentially. Voice recognition device. Generating predetermined information based on an external audio signal;
Temporarily storing predetermined information;
Deriving a word corresponding to the voice signal based on the temporarily stored predetermined information, performing voice recognition, and outputting the word;
Deriving a word corresponding to the speech signal and obtaining a resource usage rate for speech recognition;
And a step of suspending and resuming based on the usage rate obtained by the processing for performing speech recognition. Generating predetermined information based on an external audio signal;
Temporarily storing predetermined information;
Deriving a word corresponding to the voice signal based on the temporarily stored predetermined information, performing voice recognition, and outputting the word;
Deriving a word corresponding to the speech signal and obtaining a resource usage rate for speech recognition;
A computer-readable recording medium on which a program is recorded, which causes a computer to execute a step of suspending and resuming based on a usage rate obtained by processing for performing speech recognition.

Images