JP6478727B2 - Audio processing apparatus, audio processing method and program - Google Patents

Description

  Embodiments described herein relate generally to a voice processing device, a voice processing method, and a program.

  With recent increase in communication demand, wireless communication devices (wireless mobile stations and wireless base stations) are rapidly spreading. In order to maintain order and effectively use radio waves in such a situation, it is necessary to use each wireless communication device under certain conditions. However, it cannot be said that all the wireless communication devices are operated in a condition due to a failure or illegal operation of the wireless communication device. If these wireless communication devices are left unattended, there is a risk that the operation of a wireless communication device that is operating normally may be disturbed. Therefore, it is important to monitor the use of radio waves and prevent the occurrence of abnormal radio waves. ing. However, the frequency band of radio signals is wide, and it is costly for a person to constantly monitor the entire frequency band. Therefore, attention is being focused on a technique for automatically detecting a target signal from a radio signal.

  Here, it is assumed that voice communication using abnormal radio waves is detected. In this case, the target signal is voice (human speech). As a technique for automatically detecting a human utterance from an acoustic signal, a technique called “speech section detection” is known. Speaking section detection is a technique mainly used in speech recognition and the like, and various methods have been developed so far. This technique for detecting an utterance section is considered to be a useful technique when detecting voice communication using abnormal radio waves.

  The frequency band used by the abnormal radio wave communication person cannot usually be known in advance. Therefore, in order to specify a frequency band in which abnormal radio waves exist, a method using a filter bank composed of a plurality of band pass filters having different frequency bands (pass bands) to be passed can be considered. The passband of the bandpass filter in which speech (speech) is detected by dividing the radio signal into multiple subband signals using a filter bank and performing speech segment detection on the demodulated signal obtained by demodulating each subband signal Therefore, it is possible to specify the frequency band used by the abnormal radio wave communication person.

JP 2007-17620 A

  However, depending on the configuration of the filter bank, such as when the passbands of adjacent bandpass filters overlap, one abnormal radio wave may pass through a plurality of bandpass filters. Here, when an abnormal radio wave includes an utterance, the same utterance is detected from a plurality of demodulated signals corresponding to the plurality of bandpass filters. For this reason, for example, when a person listens to and confirms a detected utterance, the same utterance is listened to many times, and the confirmation work becomes complicated, making it impossible to efficiently detect voice communication using abnormal radio waves. there were.

  The problem to be solved by the present invention is to provide a voice processing device, a voice processing method, and a program capable of efficiently detecting voice communication using abnormal radio waves.

The speech processing apparatus according to the embodiment includes a dividing unit, a demodulating unit, a detecting unit, a determining unit, and a selecting unit. The dividing unit divides the received radio signal into a plurality of subband signals using a filter bank including a plurality of bandpass filters having different pass bands. The demodulating unit individually demodulates the plurality of subband signals to generate a plurality of demodulated signals respectively corresponding to the plurality of bandpass filters. The detection unit detects an utterance from each of the plurality of demodulated signals based on a reliability score representing the likelihood of the utterance. The determination unit sets the utterance detected from the demodulated signal of interest among the plurality of demodulated signals as the first utterance, and another bandpass adjacent in the frequency direction to the bandpass filter corresponding to the demodulated signal of interest When the utterance detected from the other demodulated signal corresponding to the filter is the second utterance, when there is one or more second utterances that overlap at least part of the time with respect to the first utterance, It is determined whether or not the first utterance and the second utterance are the same utterance. When it is determined that the first utterance and the second utterance are the same utterance, the selection unit selects one of the first utterance and the second utterance.

The block diagram which shows the functional structural example of the audio processing apparatus of embodiment. The flowchart which shows an example of the process sequence by the audio | voice processing apparatus of embodiment. The figure explaining the structural example of a filter bank. The figure which represented on the time-frequency plane an example of the speech which exists in a radio signal. The figure which shows the example of a waveform of a demodulation signal. The figure which shows the example of a waveform of a demodulation signal. The figure which shows the example of a waveform of a demodulation signal. The figure which shows the result of having detected speech from the demodulated signal shown in FIG. The figure which shows the result of having detected speech from the demodulated signal shown in FIG. The figure which shows the result of having detected speech from the demodulated signal shown in FIG. The figure which shows an example of the information regarding speech. The block diagram which shows the hardware structural example of an audio processing apparatus.

  Hereinafter, an audio processing device, an audio processing method, and a program according to embodiments will be described in detail with reference to the accompanying drawings. The speech processing apparatus according to the present embodiment detects a human utterance (speech) from a radio wave signal and outputs information related to the detected utterance.

  First, the configuration of the speech processing apparatus according to the present embodiment will be described with reference to FIG. FIG. 1 is a block diagram illustrating a functional configuration example of the speech processing apparatus 1 according to the present embodiment. As shown in FIG. 1, the audio processing apparatus 1 includes a dividing unit 11, a plurality of demodulation units 12_1, 12_2,..., 12_n (hereinafter collectively referred to as a demodulation unit 12) and a plurality. , 13_n (hereinafter collectively referred to as the detection unit 13), a determination unit 14, a selection unit 15, and an output unit 16.

  The dividing unit 11 divides the received radio wave signal into a plurality of subband signals using a filter bank composed of a plurality of bandpass filters having different pass bands. In the plurality of bandpass filters constituting the filter bank, part of the passband may overlap with the passband of the adjacent bandpass filter.

  The demodulator 12 demodulates the subband signal divided by the divider 11 to generate a demodulated signal. Each of the plurality of demodulation units 12 corresponds to an individual subband signal. That is, each of the plurality of demodulation units 12 individually demodulates a subband signal that is a signal that has passed through each bandpass filter constituting the filter bank. As a result, a plurality of demodulated signals respectively corresponding to the plurality of bandpass filters constituting the filter bank are generated. As a method of modulating the radio signal and demodulating the subband signal by the demodulator 12, for example, digital modulation such as frequency shift keying (FSK) or phase modulation (PSK) may be used, or amplitude modulation (AM). Or analog modulation such as frequency modulation (FM).

  In FIG. 1, assuming that demodulation for a plurality of subband signals is performed by parallel processing, the same number of demodulations as the number of subband signals (the same number as the number of bandpass filters constituting the filter bank). Although the structure provided with the part 12 is illustrated, it is not restricted to this. For example, a configuration may be employed in which demodulation of at least a part of a plurality of subband signals is performed in time series by a single demodulation unit 12 or a number of demodulation units 12 smaller than the number of subband signals. Each of the plurality of subband signals divided by the dividing unit 11 and the plurality of demodulated signals demodulated by the demodulating unit 12 is added with common time information indicating the position of the signal component in the time direction. To do.

  The detection unit 13 calculates a reliability score indicating the likelihood of the utterance of the demodulated signal generated by the demodulation unit 12 along the time direction of the demodulated signal, and utters the utterance from the demodulated signal based on the calculated reliability score. Is detected. Each of the plurality of detection units 13 corresponds to an individual demodulated signal. That is, each of the plurality of detection units 13 is provided corresponding to the individual demodulation unit 12, and performs processing for individually detecting an utterance on the demodulated signal generated by each demodulation unit 12. As a method for detecting an utterance based on the reliability score, a known method such as a method described in Patent Document 1 can be used.

  In FIG. 1, the number of subband signals is the same as the number of subband signals (the same number as the number of bandpass filters constituting the filter bank) on the assumption that speech detection for a plurality of demodulated signals is performed by parallel processing. Although the structure provided with the detection part 13 is illustrated, it is not restricted to this. For example, a configuration may be adopted in which utterances are detected in time series for at least some of a plurality of demodulated signals by a single detector 13 or a number of detectors 13 smaller than the number of demodulated signals.

  The determination unit 14 determines the identity of utterances respectively detected from different demodulated signals by different detection units 13 based on the utterance detection results by the plurality of detection units 13. Here, the utterance detected from the demodulated signal of interest among the plurality of demodulated signals is referred to as a “first utterance”, and the bandpass filter corresponding to the demodulated signal of interest has another bandpass filter close to the passband. The utterance detected from the corresponding demodulated signal is defined as a “second utterance”. Note that the other bandpass filter referred to here may be a bandpass filter adjacent in the frequency direction to the bandpass filter corresponding to the demodulated signal of interest.

  First, the determination unit 14 searches for a second utterance in which at least a part of time overlaps with each of the first utterances detected from the demodulated signal of interest. As a result of the search, if one or more second utterances that overlap at least part of the time are found with respect to the first utterance, are the first and second utterances the same utterance? Determine whether or not. The determination unit 14 repeatedly performs the above processing on each of the plurality of demodulated signals while switching the demodulated signal of interest.

  Whether or not the first utterance and the second utterance are the same utterance can be determined based on, for example, the degree of overlap of the times when the respective utterances exist. Specifically, for example, the difference between the start times of the first utterance and the second utterance is within a predetermined time, and the difference between the end times of the first utterance and the second utterance is within a predetermined time. In some cases, it is determined that these utterances are the same utterance.

  Whether or not the first utterance and the second utterance are the same utterance is determined, for example, by the evaluation result of the similarity between the feature amount extracted from the first utterance and the feature amount extracted from the second utterance Can be determined based on Examples of the feature amount used here include logarithmic power and MFCC (Mel-Frequency Cepstrum Coefficient). Further, the reliability score calculated by the detection unit 13 may be used as a feature amount. As a method for evaluating the similarity of feature amounts, for example, a correlation coefficient (for example, an inner product) of feature amounts calculated for each time (for example, a frame) from the first utterance and the second utterance may be used. it can. At this time, if the times of the first utterance and the second utterance do not completely match and some times overlap, the similarity of the feature quantities is evaluated using the time of the overlapped portion. Alternatively, the similarity between feature quantities may be evaluated using a time including two utterances.

  Of the utterances detected from the plurality of demodulated signals by the plurality of detection units 13 using the result of determination by the determination unit 14, the selection unit 15 is an utterance whose information is to be output by the output unit 16 described later. Select. That is, when the determination unit 14 determines that the first utterance and the second utterance are the same utterance, the selection unit 15 selects one of the first utterance and the second utterance. An utterance, for example, an utterance having a higher reliability score calculated by the detection unit 13 is selected. Further, the selection unit 15 does not find at least one second utterance in which at least a part of the time overlaps with the first utterance in the process by the determination unit 14, or at least one for the first utterance. If one or more second utterances with the same part time are found, but it is determined that the first utterance and the second utterance are not the same utterance, the first utterance is selected.

  In addition, when the selection unit 15 selects one of the first utterance and the second utterance determined as the same utterance by the determination unit 14, the selected utterance and the time of the utterance not selected If they do not completely match, the selected utterance may be integrated with a part of the utterance that was not selected and that does not overlap the selected utterance. In this case, the integrated utterance is set as an utterance for which information is output by the output unit 16 described later.

  The output unit 16 outputs information related to the utterance selected by the selection unit 15. The information output from the output unit 16 may be, for example, an audio signal of an utterance selected by the selection unit 15, a bandpass filter number corresponding to a demodulated signal from which an utterance has been detected, Information for identifying the selected utterance, such as the time when the utterance exists, may be used. Moreover, you may output combining these information. Furthermore, the reliability score of the selected utterance may be added and output.

  Further, the output unit 16 may output information related to the utterance that has not been selected together with information related to the utterance selected by the selection unit 15. For example, in combination with the voice signal of the utterance selected by the selection unit 15, not only the number of the bandpass filter corresponding to the demodulated signal from which the selected utterance has been detected, but also the demodulated signal from which the utterance not selected has been detected The number of the bandpass filter corresponding to may also be output.

  Next, the operation of the speech processing apparatus 1 of this embodiment will be described with reference to FIG. FIG. 2 is a flowchart illustrating an example of a processing procedure performed by the voice processing device 1. A series of processes shown in the flowchart of FIG. 2 is repeatedly executed at a predetermined cycle by the audio processing apparatus 1 together with reception of the radio signal.

  When the processing shown in the flowchart of FIG. 2 is started, first, the dividing unit 11 divides the received radio wave signal into a plurality of subband signals (step S101). The plurality of subband signals divided by the dividing unit 11 are respectively supplied to the plurality of demodulating units 12.

  Next, each of the plurality of demodulation units 12 individually demodulates the subband signals supplied from the division unit 11 to generate a plurality of demodulation signals (step S102). The plurality of demodulated signals generated by the plurality of demodulation units 12 are respectively supplied to the corresponding plurality of detection units 13.

  Next, each of the plurality of detection units 13 detects the utterance from the demodulated signal supplied from the demodulation unit 12 based on the reliability score representing the likelihood of the utterance (step S103). The detection results of the utterances by the plurality of detection units 13 are supplied to the determination unit 14 and the selection unit 15.

  Next, the determination unit 14 performs a process of determining the identity of the utterances detected from the different demodulated signals by the different detection units 13 based on the utterance detection results by the plurality of detection units 13. That is, the determination unit 14 first searches for a second utterance in which at least a part of the time overlaps with the first utterance for each of the first utterances detected from the demodulated signal of interest (step S104). . If there is one or more second utterances that overlap at least part of the time with respect to the first utterance, the determination unit 14 determines whether the first utterance and the second utterance are the same utterance. It is determined whether or not (step S105). The determination unit 14 repeatedly performs the processing of step S104 and step S105 for all utterances detected by the plurality of detection units 13 while switching the demodulated signal of interest. The result of determination by the determination unit 14 is supplied to the selection unit 15.

  Next, the selection unit 15 is a target to output information by the output unit 16 among the utterances detected from the plurality of demodulated signals by the plurality of detection units 13 using the determination result by the determination unit 14. An utterance is selected (step S106). The result of selection by the selection unit 15 is supplied to the output unit 16.

  Finally, the output unit 16 acquires information related to the utterance selected by the selection unit 15 from each unit from the division unit 11 to the determination unit 14, for example, an output device such as a display or a speaker, a file storage device such as an HDD, The data is output to a communication I / F connected to the network (step S107).

  As described above, the sound processing apparatus 1 according to the present embodiment divides the received radio wave signal into a plurality of subband signals, and detects speech from the demodulated signals obtained by demodulating each subband signal. At this time, when the same utterance is detected from different demodulated signals, one of the utterances is selected by the processing of the determination unit 14 and the selection unit 15, and information about the selected utterance is output. Therefore, for example, when a person listens to and confirms a detected utterance, the same utterance is not listened to many times, and the time required for the confirmation work can be reduced, so voice communication using abnormal radio waves can be detected efficiently. Can do.

  Next, an example of processing by the speech processing apparatus 1 of the present embodiment will be described with specific examples. First, a specific example of a radio wave signal to be processed and a processing result up to the detection unit 13 will be described.

  FIG. 3 is a diagram illustrating a configuration example of a filter bank in the dividing unit 11. The filter bank illustrated in FIG. 3 is configured by arranging a plurality of bandpass filters having a passband width of 8000 hertz every 6000 hertz. In each bandpass filter constituting the filter bank, a part of the passband overlaps the passband of the adjacent bandpass filter.

  FIG. 4 is a diagram illustrating an example of an utterance existing in a radio signal on a time-frequency plane. In this example, it is assumed that the utterance U11 and the utterance U12 exist in the radio signal. The filter bank illustrated in FIG. 3 is shown on the left side of FIG. When this radio wave signal is divided using this filter bank, the signal component of the utterance U11 is included in the subband signal that has passed through the bandpass filter F1 and the subband signal that has passed through the bandpass filter F2. In addition, the signal component of the utterance U12 is included in the subband signal that has passed through the bandpass filter F3.

  FIG. 5 is a diagram illustrating a waveform example of a demodulated signal obtained by demodulating the subband signal that has passed through the bandpass filter F1. FIG. 6 is a diagram illustrating a waveform example of a demodulated signal obtained by demodulating the subband signal that has passed through the bandpass filter F2. FIG. 7 is a diagram illustrating a waveform example of a demodulated signal obtained by demodulating the subband signal that has passed through the bandpass filter F3. T0 and Tn in the figure each indicate a common time.

  FIGS. 8 to 10 are diagrams showing the results of detecting speech by the detection unit 13 for the demodulated signals shown in FIGS. The graph in the figure represents a time series of reliability scores calculated by the detection unit 13. In this example, the detection unit 13 detects a section in which the reliability score exceeds a threshold as an utterance. As a result, from the demodulated signal corresponding to the bandpass filter F1, a section from 26.4 seconds to 30.3 seconds is detected as an utterance U21 as shown in FIG. Further, from the demodulated signal corresponding to the band pass filter F2, as shown in FIG. 9, a section from 26.1 seconds to 29.9 seconds is detected as the utterance U22. Further, from the demodulated signal corresponding to the bandpass filter F3, as shown in FIG. 10, a section from 18.4 seconds to 38.1 seconds is detected as the utterance U23.

  FIG. 11 is a diagram illustrating an example of information related to the utterances U21, U22, and U23 detected by the detection unit 13 in this example. The average reliability score in the figure indicates the average reliability score in the utterance interval.

  Next, behaviors of the determination unit 14, the selection unit 15, and the output unit 16 in this example will be described.

  For each detected utterance, the determination unit 14 in this example is an utterance detected from another demodulated signal adjacent to the corresponding demodulated signal in which the corresponding utterance is detected, and the corresponding bandpass filter is the utterance Search for utterances that overlap at least part of the time. According to this method, for the speech U21 detected from the demodulated signal corresponding to the bandpass filter F1, the determination unit 14 firstly demodulates the demodulated signal corresponding to the adjacent bandpass filter F0 and the demodulated signal corresponding to the bandpass filter F2. The target utterance is searched from the result of the detection unit 13 for. In this example, the utterance is not detected from the demodulated signal corresponding to the bandpass filter F0, and the utterance U22 is detected from the demodulated signal corresponding to the bandpass filter F2. And the utterance U22 overlaps with the utterance U21 in the section from 26.4 seconds to 29.9 seconds. Therefore, the determination unit 14 writes the two utterances U21 and U22 together in the storage unit in order to determine later whether or not the utterance U21 and the utterance U22 are the same utterance.

  Next, for the utterance U22 detected from the demodulated signal corresponding to the bandpass filter F2, the determining unit 14 detects the demodulated signal corresponding to the adjacent bandpass filter F1 and the demodulated signal corresponding to the bandpass filter F3. From the result, the target utterance is searched. In this example, the utterance U21 is detected from the demodulated signal corresponding to the bandpass filter F1, and the utterance U23 is detected from the demodulated signal corresponding to the bandpass filter F3. And the utterance U23 overlaps with the utterance U22 in the section from 26.1 seconds to 29.9 seconds. Therefore, the determination unit 14 writes the two utterances U22 and U23 as a set to the storage unit in order to determine later whether or not the utterance U22 and the utterance U23 are the same utterance. In addition, since the set of the utterance U21 and the utterance U22 has already been written in the storage unit, new writing is not performed here in order to avoid duplication.

  Next, the determination unit 14 determines whether or not both utterances are the same utterance for each utterance set written in the storage unit. The determination unit 14 in this example calculates the correlation coefficient of the reliability score using the time at which two utterances overlap, and has the correlation coefficient exceeded a predetermined threshold (here, 0.60)? It is determined whether or not two utterances are the same utterance.

  First, for the set of the utterance U21 and the utterance U22, the reliability score calculated from the demodulated signal corresponding to the bandpass filter F1 in the section from 26.4 seconds to 29.9 seconds, which is the time when both utterances overlap. And a correlation coefficient with the reliability score calculated from the demodulated signal corresponding to the bandpass filter F2. As a result, the calculated correlation coefficient is 0.91, which exceeds the threshold value of 0.60. Therefore, the determination unit 14 determines that these two utterances are the same utterance. Next, for the set of the utterance U22 and the utterance U23, the reliability calculated from the demodulated signal corresponding to the bandpass filter F2 in the interval from 26.1 seconds to 29.9 seconds, which is the time when both utterances overlap. A correlation coefficient between the score and the reliability score calculated from the demodulated signal corresponding to the bandpass filter F3 is obtained. As a result, the calculated correlation coefficient is 0.08, which is below the threshold value of 0.60, so the determination unit 14 determines that these two utterances are not the same utterance.

  In response to the determination result of the determination unit 14, the selection unit 15 in this example selects the utterance for the utterance for which the utterance determined to be the same does not exist, and for the utterance for which the utterance determined to be the same exists. The utterance having the highest reliability score (for example, average reliability score) is selected from the utterances determined to be the same. As a result, the utterance U23 for which there is no utterance determined to be the same is selected as the utterance to be output by the output unit 16. For the utterance U21 and the utterance U22 determined to be the same, the utterance U21 having an average reliability score higher than that of the utterance U22 is selected as an utterance to be output by the output unit 16.

  In response to the result of the selection unit 15, the output unit 16 in this example outputs information on the selected utterance. For example, the output unit 16 includes the number of the band pass filter F1 corresponding to the demodulated signal from which the utterance U21 is detected, the time when the utterance exists, the average reliability score, and the like as information on the selected utterance U21. Output audio signals. Further, the output unit 16 includes the utterance U23 as information on the selected utterance U23 together with the number of the bandpass filter F3 corresponding to the demodulated signal from which the utterance U23 is detected, the time when the utterance exists, the average reliability score, and the like. Audio signal is output.

  As described above, in this example, two utterances U21 and U22 indicating the same utterance U11 in the radio signal are obtained from a plurality of demodulated signals obtained by decoding a plurality of subband signals divided from the radio signal. Was detected. Here, when a person listens to confirm the detected utterance, the two speeches U21 and U22 indicating the same utterance U11 are repeatedly listened to by simply applying the conventional speech segment detection as it is. Becomes complicated. On the other hand, according to the present embodiment, one utterance U21 of two utterances U21 and U22 indicating the same utterance U11 is selected as an information output target, so confirmation is performed without repeatedly listening to the same utterance. be able to. As a result, it is possible to reduce the time and effort required for the confirmation work, and it is possible to efficiently detect voice communication using abnormal radio waves.

  The speech processing apparatus 1 according to the present embodiment uses, for example, a general-purpose computer system as basic hardware, and executes a predetermined program (software) on the computer system, whereby the above-described units (dividing unit 11, demodulating unit) 12, the detection unit 13, the determination unit 14, the selection unit 15, and the output unit 16) can be realized.

  FIG. 12 is a block diagram illustrating a hardware configuration example of the voice processing device 1 according to the present embodiment. For example, as shown in FIG. 12, the audio processing device 1 is a peripheral device that mediates data input / output between a processor such as a CPU 101, a storage device such as a RAM 102 and a ROM 103, and a peripheral device such as a display 110 and a speaker 120. It has a hardware configuration of a normal computer device including an I / F 104, a file storage device such as an HDD 105, and a communication I / F 106 that communicates with the outside via a network.

  At this time, the above programs are, for example, magnetic disks (flexible disks, hard disks, etc.), optical disks (CD-ROM, CD-R, CD-RW, DVD-ROM, DVD ± R, DVD ± RW, Blu-ray ( (Registered trademark) Disc, etc.), a semiconductor memory, or a similar recording medium. The recording medium for recording the program may be in any form as long as the computer system can read the recording medium. Further, the program may be configured to be installed in advance in the computer system, or the program distributed via a network may be configured to be installed in the computer system as appropriate.

  The program executed in the above computer system is the above-described units (dividing unit 11, demodulating unit 12, detecting unit 13, determining unit 14, selecting unit 15) that are functional components in the audio processing device 1 of the present embodiment. And a module configuration including the output unit 16), and the processor reads and executes the program as appropriate, so that the above-described units are generated on the main memory such as the RAM 102.

  Note that each of the above-described units (the dividing unit 11, the demodulating unit 12, the detecting unit 13, the determining unit 14, the selecting unit 15, and the output unit 16) of the audio processing device 1 according to the present embodiment is realized only by a program (software). Alternatively, a part or all of them can be realized by dedicated hardware such as ASIC (Application Specific Integrated Circuit) or FPGA (Field-Programmable Gate Array).

  The voice processing apparatus 1 according to the present embodiment may be configured as a network system in which a plurality of computers are communicably connected, and the above-described units may be distributed and realized in a plurality of computers. For example, one computer having the function of the dividing unit 11 and a plurality of computers having the functions of the corresponding one demodulating unit 12 and the detecting unit 13 among the plurality of demodulating units 12 and the plurality of detecting units 13. The speech processing apparatus 1 according to the present embodiment may be connected to a single computer having the functions of the determination unit 14, the selection unit 15, and the output unit 16 in a communicable manner.

  As mentioned above, although embodiment of this invention was described, embodiment described here is shown as an example and is not intending limiting the range of invention. The novel embodiments described herein can be implemented in various other forms, and various omissions, replacements, and changes can be made without departing from the spirit of the invention. The embodiments and modifications described herein are included in the scope and gist of the invention, and are also included in the invention described in the claims and the equivalents thereof.

DESCRIPTION OF SYMBOLS 1 Audio | voice processing apparatus 11 Dividing part 12 (12_1,12_2, ..., 12_n) Demodulating part 13 (13_1,13_2, ..., 13_n) Detection part 14 Determination part 15 Selection part 16 Output part U11, U12 (Radio signal Speech F1, F2, F3 Bandpass filters U21, U22, U23 (detected from the demodulated signal)

Claims (11)

A dividing unit that divides a received radio wave signal into a plurality of subband signals using a filter bank composed of a plurality of bandpass filters having different passbands;
A demodulator that individually demodulates the plurality of subband signals and generates a plurality of demodulated signals respectively corresponding to the plurality of bandpass filters;
A detection unit that detects an utterance from each of the plurality of demodulated signals based on a reliability score representing the likelihood of the utterance;
The utterance detected from the demodulated signal of interest among the plurality of demodulated signals is defined as the first utterance and corresponds to another bandpass filter adjacent in the frequency direction to the bandpass filter corresponding to the demodulated signal of interest. When the utterance detected from other demodulated signals is set as the second utterance, when there are one or more second utterances that overlap at least a part of the time with respect to the first utterance, these first utterances exist. A determination unit for determining whether the utterance and the second utterance are the same utterance;
A voice processing comprising: a selection unit that selects one of the first utterance and the second utterance when it is determined that the first utterance and the second utterance are the same utterance apparatus.   The said determination part determines whether these 1st utterance and 2nd utterance are the same utterance based on the overlapping degree of the time of a 1st utterance and a 2nd utterance. The voice processing apparatus according to 1.   Based on the evaluation result of the similarity between the feature amount extracted from the first utterance and the feature amount extracted from the second utterance, the determination unit determines that the first utterance and the second utterance are the same utterance. The speech processing apparatus according to claim 1, wherein it is determined whether or not there is.   When it is determined that the first utterance and the second utterance are the same utterance, the selection unit selects an utterance having a high reliability score from the first utterance and the second utterance. The speech processing apparatus according to any one of claims 1 to 3.   The selection unit has one second utterance in which at least part of time overlaps with the first utterance when there is no second utterance in which at least part of time overlaps with the first utterance. The voice processing according to any one of claims 1 to 4, wherein the first utterance is selected when it is determined that the first utterance and the second utterance are not the same utterance although there is the above utterance. apparatus.   The speech processing apparatus according to any one of claims 1 to 5, wherein the selection unit integrates, with respect to the selected utterance, a part of the utterance that is not selected and a time that does not overlap with the selected utterance.   The speech processing apparatus according to any one of claims 1 to 6, further comprising an output unit configured to output information related to the selected utterance.   The speech processing apparatus according to claim 7, wherein the output unit further outputs information about an unselected utterance together with information about the selected utterance.   The determination unit sets the utterance detected from the demodulated signal of interest among the plurality of demodulated signals as a first utterance, and another band adjacent in the frequency direction to the bandpass filter corresponding to the demodulated signal of interest. When the utterance detected from another demodulated signal corresponding to the pass filter is the second utterance, when there is one or more second utterances that overlap at least part of the time with respect to the first utterance The speech processing apparatus according to claim 1, wherein the first utterance and the second utterance are determined to be the same utterance. A voice processing method executed by a voice processing device,
Dividing a received radio wave signal into a plurality of subband signals using a filter bank composed of a plurality of bandpass filters having different passbands;
Demodulating the plurality of subband signals individually to generate a plurality of demodulated signals respectively corresponding to the plurality of bandpass filters;
Detecting an utterance from each of the plurality of demodulated signals based on a confidence score representing the likelihood of the utterance;
The utterance detected from the demodulated signal of interest among the plurality of demodulated signals is defined as the first utterance and corresponds to another bandpass filter adjacent in the frequency direction to the bandpass filter corresponding to the demodulated signal of interest. When the utterance detected from other demodulated signals is set as the second utterance, when there are one or more second utterances that overlap at least a part of the time with respect to the first utterance, these first utterances exist. Determining whether the utterance and the second utterance are the same utterance;
A step of selecting one of the first utterance and the second utterance when it is determined that the first utterance and the second utterance are the same utterance. . On the computer,
A function of dividing a received radio wave signal into a plurality of subband signals using a filter bank composed of a plurality of bandpass filters having different passbands;
A function of individually demodulating the plurality of subband signals to generate a plurality of demodulated signals respectively corresponding to the plurality of bandpass filters;
A function of detecting an utterance from each of the plurality of demodulated signals based on a confidence score representing the likelihood of the utterance;
The utterance detected from the demodulated signal of interest among the plurality of demodulated signals is defined as the first utterance, and the other corresponding to another bandpass filter adjacent in the frequency direction to the bandpass filter corresponding to the demodulated signal of interest. If the utterance detected from the demodulated signal is the second utterance, and there is one or more second utterances that overlap at least part of the time with respect to the first utterance, these first utterances And a function for determining whether or not the second utterance is the same utterance;
A function for selecting one of the first utterance and the second utterance when it is determined that the first utterance and the second utterance are the same utterance; program.

Images