Keyword detection device, keyword detection method and computer program product

By introducing a delay unit and a keyword exclusion unit in the keyword detection device, combined with sound interval detection, the problem of false detection in the prior art is solved, more accurate keyword recognition is achieved, and the key to non-intention detection is reduced Word reaction.

CN112655043BActive Publication Date: 2024-11-12NIPPON TELEGRAPH & TELEPHONE CORP
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN201980057870.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Priority Date
2018-09-11
Filing Date
2019-08-28
Publication Date
2024-11-12
Estimated Expiration
2039-08-28

AI Technical Summary

Technical Problem

In the prior art, the keyword detection device is prone to misdetecting due to containing keywords or phonemes close to the keywords, resulting in the pronunciation of keywords that are not intended to be mistaken for target sounds.

Method used

A keyword detection device is designed, including a keyword detection unit, a sound detection unit, a delay unit and a keyword exclusion unit in the sentence. Through delay processing and sound interval detection, false detection results are eliminated, and only keywords to be detected are ensured. Only when pronunciation is recognized.

Benefits of technology

It effectively suppresses false detection of keyword pronunciations that are not intended to be detected, and reduces unnecessary keyword responses.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN112655043B_ABST
    Figure CN112655043B_ABST
Patent Text Reader

Abstract

Suppressing the erroneous detection of a keyword that is spoken without the intention of detecting the keyword. A keyword detection unit (11) generates a keyword detection result indicating the result of detecting the pronunciation of a specific keyword from an input voice. A voice detection unit (12) generates a voice interval detection result indicating the result of detecting a voice interval from the input voice. A delay unit (13) assigns a delay at least longer than the utterance time of the keyword to the voice interval detection result. A sentence keyword exclusion unit (14) updates the keyword detection result to a result indicating that the keyword is not detected when the keyword detection result indicates that the keyword is detected and the voice interval detection result indicates that the voice interval is detected.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to a technique for detecting pronunciation of a keyword. Background Art

[0002] For example, some devices that can be controlled by voice, such as smart speakers or car systems, are equipped with a function called keyword wake-up, which starts voice recognition when a keyword that serves as a trigger is pronounced. Such a function requires technology that uses a voice signal as input and detects the pronunciation of the keyword.

[0003] Figure 1 This is a conventional structure disclosed in Non-Patent Document 1. In the conventional technology, when keyword detection section 91 detects the pronunciation of a keyword from an inputted audio signal, target sound output section 99 turns on a switch to output the audio signal as a target sound for audio recognition or the like.

[0004] Prior art literature

[0005] Non-patent literature

[0006] Non-patent document 1: Sensory, Inc., "TrulyHandsfreeTM", [online], [retrieved on August 17, 2011], Internet<URL:http: / / www.sensory.co.jp / product / thf.htm> Summary of the invention

[0007] Problems to be solved by the invention

[0008] However, in the prior art, even if the utterance is not intended to detect a keyword, if the utterance contains a keyword or a phoneme close to the keyword, the keyword or the phoneme close to the keyword may be reacted to, resulting in erroneous detection of the keyword. For example, the following case can be cited: when the keyword is "Hello ABC", the keyword is said to the user even if there is no intention to detect the keyword, such as saying "The keyword is 'Hello ABC'".

[0009] In view of the above-mentioned technical problems, an object of the present invention is to suppress erroneous detection of a keyword that is spoken without the intention of detecting the keyword.

[0010] Means for solving problems

[0011] In order to solve the above-mentioned problems, a keyword detection device of one embodiment of the present invention includes: a keyword detection unit, which generates a keyword detection result indicating the result of detecting the pronunciation of a specific keyword from an input sound; a sound detection unit, which generates a sound interval detection result indicating the result of detecting a sound interval from the input sound; a delay unit, which assigns a delay that is at least longer than the utterance time of the keyword to the sound interval detection result; and a keyword exclusion unit in a sentence, which updates the keyword detection result to a result indicating that the keyword is not detected when the keyword detection result indicates that the keyword is detected and the sound interval detection result indicates that the sound interval is detected.

[0012] Effects of the Invention

[0013] According to the present invention, it is possible to suppress erroneous detection of a keyword that is spoken without intending to detect the keyword. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 This is a diagram illustrating a functional structure of a conventional keyword detection device.

[0015] Figure 2 This is a diagram illustrating the functional configuration of the keyword detection device according to the first embodiment.

[0016] Figure 3 It is a diagram illustrating the processing procedure of the keyword detection method according to the first embodiment.

[0017] FIG. 4 is a diagram for explaining the principle of the first embodiment.

[0018] Figure 5 This is a diagram illustrating the functional configuration of the keyword detection device according to the second embodiment.

[0019] Figure 6 It is a diagram illustrating the processing procedure of the keyword detection method according to the second embodiment.

[0020] FIG. 7 is a diagram for explaining the principle of the second embodiment.

[0021] Figure 8 It is a diagram illustrating the functional structure of the keyword detection device according to the third embodiment. DETAILED DESCRIPTION

[0022] Hereinafter, the embodiments of the present invention will be described in detail. In addition, in the drawings, the same numbers are attached to the components having the same functions, and the repeated description is omitted.

[0023] [First embodiment]

[0024] The keyword detection device 1 of the first embodiment receives a user's voice (hereinafter referred to as "input voice") as input, and if a keyword is detected from the input voice, outputs a target voice to be used for voice recognition or the like. Figure 2 As shown in the figure, the keyword detection device 1 comprises a keyword detection unit 11, a sound detection unit 12, a delay unit 13, a sentence keyword elimination unit 14 and a target sound output unit 19. Figure 3 The processing of each step shown implements the keyword detection method S1 of the first embodiment.

[0025] The keyword detection device 1 is, for example, a special device that is constructed by reading a special program into a well-known or special-purpose computer having a central processing unit (CPU), a main storage device (RAM), etc. The keyword detection device 1 executes each process based on the control of the central processing unit. The data input to the keyword detection device 1 or the data obtained in each process is, for example, stored in the main storage device, and the data stored in the main storage device is read out to the central processing unit as needed and used for other processes. Each processing unit of the keyword detection device 1 may also be at least partially composed of hardware such as an integrated circuit.

[0026] Below, reference Figure 3 , the keyword detection method performed by the keyword detection device of the first embodiment is described.

[0027] In step S11, the keyword detection unit 11 detects the pronunciation of a predetermined specific keyword from the input sound. The keyword detection is performed by, for example, using a neural network learned in advance to determine whether a pattern of a power spectrum obtained in a short period of time is similar to a pattern of a keyword recorded in advance. The keyword detection unit 11 outputs a keyword detection result indicating that a keyword is detected ("keyword detected") or that a keyword is not detected ("keyword not detected") to the sentence keyword exclusion unit 14.

[0028] In step S12, the sound detection unit 12 detects a sound interval from the input sound. The detection of the sound interval is performed, for example, as follows. First, the stationary noise level N(t) is calculated based on the long-term average of the input sound. Next, the threshold is set by multiplying the stationary noise level N(t) by a specific constant α. And, the interval in which the short-time average level P(t) is greater than the threshold is detected as a sound interval. In addition, the sound interval can also be detected by adding whether the shape of the spectrum or cepstrum is consistent with the characteristics of the sound as a judgment factor. The sound detection unit 12 outputs the sound interval detection result indicating that the sound interval is detected ("sound detected") or the sound interval is not detected ("no sound detected") to the delay unit 13.

[0029] In the calculation of the short-time average level P(t), the root mean square power obtained by applying a rectangular window of the average keyword speech time T or multiplying it by an exponential window is calculated. If the power at discrete time t is set to P(t) and the input signal is set to x(t), it becomes:

[0030] [Number 1]

[0031]

[0032] P(t)=αP(t-1)+(1-α)x(t) 2

[0033] Here, α is the forgetting coefficient, and a value of 0<α<1 is preset. α is set so that the time constant becomes the average keyword speech time T (sample). That is, α=1-1 / T. Alternatively, the absolute value average power obtained by applying a rectangular window of the keyword speech time T or the absolute value average power obtained by multiplying by an exponential window can be calculated as follows.

[0034] [Number 2]

[0035]

[0036] P(t)=αP(t-1)+(1-α)|x(t)|

[0037] In step S13, the delay unit 13 delays the sound interval detection result output by the sound detection unit 12 by a time equal to the sum of the detection delay time of the keyword detection, the average utterance time of the keyword, and the margin time. Figure 4A ) and the key words appear in the sentence ( Figure 4B) in each time. The margin time is set to be in the range of several hundreds of milliseconds to several seconds. The delay unit 13 outputs the delayed sound section detection result to the sentence keyword exclusion unit 14.

[0038] In step S14, the sentence keyword exclusion unit 14 excludes the detection results related to the keyword in the sentence from the keyword detection results output by the keyword detection unit 11, and outputs the results to the target sound output unit 19. Specifically, when the keyword detection result output by the keyword detection unit 11 is "keyword detected", and when the sound section detection result output by the delay unit 13 is "sound detected", it is determined that the keyword is a keyword in the sentence, and the keyword detection result is updated to "keyword not detected" and output. When the keyword detection result output by the keyword detection unit 11 is "keyword detected", and when the sound section detection result output by the delay unit 13 is "sound not detected", it is determined that the utterance is intended to detect the keyword, and "keyword detected" is output as it is. At this time, it is also possible to set a method in which, when the sound section detection result is "sound detected", instead of always updating the keyword detection result to "keyword not detected", a threshold value for the likelihood of keyword detection is set, and the keyword detection result is updated to "keyword not detected" only when the likelihood of keyword detection is lower than the threshold value.

[0039] In step S19, the target sound output unit 19 turns on the switch and outputs the input sound as the target sound when the keyword detection result output by the sentence keyword elimination unit 14 is "keyword detected". When the keyword detection result output by the sentence keyword elimination unit 14 is "keyword not detected", the switch is turned off and the output is stopped.

[0040] With such a configuration, according to the first embodiment, it is possible to exclude detection of keywords that appear in sentences spoken by the user, and to reduce erroneous detection of keywords that are spoken without the intention of detecting the keywords.

[0041] [Second Embodiment]

[0042] The keyword detection device 2 of the second embodiment takes the user's voice as input, similarly to the first embodiment, and outputs a target sound for voice recognition, etc., if a keyword is detected from the input voice. Figure 5 As shown in the figure, in addition to the keyword detection unit 11, the sound detection unit 12, the delay unit 13, the sentence keyword exclusion unit 14 and the target sound output unit 19 of the first embodiment, a buffer unit 21 is also provided. Figure 6 The processing of each step shown implements the keyword detection method S2 of the second embodiment.

[0043] Below, reference Figure 6 , the keyword detection method performed by the keyword detection device of the second embodiment is described with a focus on the differences from the keyword detection method of the first embodiment.

[0044] In step S21, the buffer unit 21 holds the sound interval detection result output by the delay unit 13 in a first-in-first-out manner for a certain period of time. Fig. 7A ) and the key words appear in the sentence ( Figure 7B The retention time (FIFO length) is set to several hundred milliseconds to several seconds.

[0045] In step S14, when the keyword detection result output by the keyword detection unit 11 is "keyword detected", and at least one of the sound segment detection results held in the buffer unit 21 is "sound detected", the sentence keyword exclusion unit 14 determines that the keyword is a keyword in the sentence, updates the keyword detection result to "keyword not detected", and outputs it. When the keyword detection result output by the keyword detection unit 11 is "keyword detected", and all the sound segment detection results held in the buffer unit 21 are "sound not detected", it is determined that the utterance is intended to detect the keyword, and "keyword detected" is output as it is.

[0046] With this configuration, according to the second embodiment, the presence or absence of sound is determined in the entire time interval stored in the buffer unit 21. This can prevent erroneous detection of a keyword appearing in a sentence by accidentally determining whether or not a keyword is in the sentence in a pause interval between utterances.

[0047] [Third Embodiment]

[0048] The keyword detection device 3 of the third embodiment receives a multi-channel audio signal as input, and outputs the audio signal of the channel in which the keyword is detected as a target sound to be used for audio recognition or the like. Figure 8 As shown, in addition to the group of the keyword detection unit 11, delay unit 13, sentence keyword exclusion unit 14 and target sound output unit 19 of the first embodiment corresponding to the number of channels M (≥2) of the input sound, it also has a multi-input sound detection unit 32 with M-channel input and output.

[0049] The multi-input sound detection unit 32 takes a multi-channel sound signal as an input, and outputs a sound interval detection result of detecting a sound interval from the sound signal of channel i to the delay unit 13-i for each integer i greater than 1 and less than M. The multi-input sound detection unit 32 can more accurately detect the sound interval by exchanging sound level information between channels. The method of detecting the sound interval of multi-channel input can use the method described in the following reference 1, etc.

[0050] [Reference 1] Japanese Patent Application Publication No. 2017-187688

[0051] With such a configuration, according to the third embodiment, when a multi-channel audio signal is input, the audio section can be detected with high accuracy, resulting in an improvement in the accuracy of keyword detection.

[0052] The above describes the embodiments of the present invention, but the specific structure is not limited to these embodiments, and even if the design is appropriately changed within the scope of the purpose of the present invention, it is obviously included in the present invention. The various processes described in the embodiments are not only executed in a time series according to the order of the description, but can also be executed in parallel or individually according to the processing capacity of the device executing the process or as needed.

[0053] [Program, recording medium]

[0054] When the various processing functions in the various devices described in the above embodiments are implemented by a computer, the processing contents of the functions that each device should have are described by a program. Then, by executing the program on a computer, the various processing functions in the various devices described above are implemented on the computer.

[0055] The program describing the processing contents may be recorded in advance on a computer-readable recording medium, which may be any medium such as a magnetic recording device, an optical disk, a magneto-optical recording medium, or a semiconductor memory.

[0056] In addition, the program can be circulated by, for example, selling, transferring, or lending a portable recording medium such as a DVD or CD-ROM on which the program is recorded. Furthermore, the program can also be circulated by storing the program in advance in a storage device of a server computer and forwarding the program from the server computer to other computers via a network.

[0057] For example, a computer that executes such a program first temporarily stores the program recorded in a portable recording medium or the program forwarded from a server computer in its own storage device. In addition, when executing the process, the computer reads the program stored in its own storage device and executes the process according to the read program. In addition, as another execution method of the program, the computer can also directly read the program from the portable recording medium and execute the process according to the program, and then each time the program is forwarded from the server computer to the computer, the process according to the acquired program can be executed one by one. In addition, it can also be configured as follows: without forwarding the program from the server computer to the computer, the so-called ASP (Application Service Provider) type service that realizes the processing function only through its execution instructions and result acquisition is used to execute the above-mentioned process. In addition, in the program in this method, it is set to include information that is used for processing by the electronic computer and is equivalent to the program (although it is not a direct instruction to the computer, it has the nature of data that specifies the processing of the computer, etc.).

[0058] In addition, in this embodiment, the present device is configured by causing a specific program to be executed on a computer, but at least a part of the processing contents may be realized in hardware.

[0059] Description of labels:

[0060] 1, 2, 3, 9 keyword detection device

[0061] 11. 91 keyword detection unit

[0062] 12 Sound detection units

[0063] 13 Delay Units

[0064] 14 Keyword exclusion units in sentences

[0065] 19, 99 target sound output unit

[0066] 21 buffer units

[0067] 32 multi-input sound detection units

Claims

1. A keyword detection device, comprising: A keyword detection unit that generates a keyword detection result indicating a result of detecting an utterance of a specific keyword from an input voice; a voice detection unit that generates a voice section detection result indicating a result of detecting a voice section from the input voice in parallel with the keyword detection, the voice section being calculated based on an average keyword utterance time; a delay unit for providing a delay to the sound section detection result which is at least longer than a time obtained by adding the average keyword utterance time and a detection delay time of the keyword detection; as well as The sentence keyword excluding unit updates the keyword detection result to indicate that the keyword is not detected when the keyword detection result indicates that the keyword is detected and the delayed voice segment detection result outputted from the delay unit indicates that the voice segment is detected.

2. The keyword detection device according to claim 1, further comprising: The buffer unit holds the delayed sound interval detection result output from the delay unit for an amount corresponding to a specific time. The sentence keyword excluding unit updates the keyword detection result to indicate that the keyword is not detected when the keyword detection result indicates that the keyword is detected and at least one of the voice segment detection results stored in the buffer indicates that a voice segment is detected.

3. The keyword detection device according to claim 1, The input sound is a sound signal containing multiple channels. The sound detection unit generates the sound section detection result corresponding to each channel included in the input sound, The keyword detecting device includes the keyword detecting unit, the delay unit, and the sentence keyword excluding unit corresponding to the number of channels included in the input sound.

4. A keyword detection method, The keyword detection unit generates a keyword detection result indicating a result of detecting a pronunciation of a specific keyword from an input voice. The voice detecting unit generates a voice section detection result indicating a result of detecting a voice section from the input voice in parallel with the keyword detection, the voice section being calculated based on an average keyword utterance time, The delay unit provides a delay to the sound section detection result that is at least longer than a time obtained by adding the average keyword utterance time and the detection delay time of the keyword detection. The sentence keyword excluding unit updates the keyword detection result to indicate that the keyword is not detected when the keyword detection result indicates that the keyword is detected and the delayed voice segment detection result output from the delay unit indicates that the voice segment is detected.

5. A computer program product, comprising a computer program for causing a computer to function as the keyword detection device according to any one of claims 1 to 3.