TW201717192A - Electronic apparatus and voice trigger method therefor - Google Patents

Abstract

A voice trigger method for an electronic apparatus is provided. The voice trigger method includes the following steps. Receive a current voice signal. Extract a current voice feature from the current voice signal. Determine whether a previous voice feature of a previous voice signal is stored. When the previous voice feature is stored, at least one of adjusting a confidence threshold and determining whether to wake up the electronic apparatus is performed according to whether the previous voice feature is corresponding to a False Acceptance situation or a False Rejection situation and the similarity between the current voice feature and the previous voice feature.

Description

Electronic device and method for waking up by voice recognition

The present invention relates to an electronic device and a method for awakening thereof, and more particularly to an electronic device and a method for awakening by voice recognition.

In recent years, due to advances in technology, users can control electronic devices through voice. For example, users can wake up electronic devices through voice. However, due to various factors, it often causes False Acceptance (that is, the electronic device is awakened in unexpected situations), or causes false rejection (False Rejection) (that is, the user). I want to wake up the electronic device through voice, but I can't wake up the electronic device successfully.) For example, noisy ambient sounds may wake up electronic devices in unexpected situations. Or, the user is talking to someone, but the voice of the speech wakes up the electronic device unexpectedly. Or, the electronic device cannot be successfully woken up due to the user's accent problem. In general, to improve the above problems, you must adjust the On-line adaptation keyword group speech model or pre-train the keyword group speech model of a specific user, but online adjustment The complexity of the whole or pre-trained keyword group speech model is high and does not meet cost considerations. If an error occurs in adjusting or pre-training the keyword group speech model, the keyword group speech model may be unusable. Therefore, how to effectively improve the situation of false reception and false rejection to increase the probability of successfully waking up electronic devices is one of the efforts of the industry.

The present invention relates to an electronic device and a method for awakening through voice recognition, which can adjust the accuracy of waking up an electronic device through voice recognition.

According to an aspect of the present invention, a method for waking up an electronic device through voice recognition is provided. The method includes the following steps. Receive a current voice signal. Capture the current sound characteristics of one of the current sound signals. It is determined whether there is a previous sound feature stored in one of the previous audio signals. When the previous sound feature is stored, at least one confidence threshold is adjusted according to a previous sound feature corresponding to a false acceptance or a false rejection and a similarity between the current sound feature and the previous sound feature. Decide whether to wake up one of the electronic devices.

According to another aspect of the present invention, an electronic device is proposed. The electronic device includes a storage device, a sound receiving device, and a processor. The sound receiving device is configured to receive a current sound signal. The processor is configured to capture a current sound characteristic of one of the current audio signals, and determine whether the storage device has a previous sound feature stored in one of the previous audio signals. When the storage device stores the previous sound feature, the processor corresponds to a false acceptance condition or a false rejection condition and the sound according to the previous sound feature. One degree of similarity between the feature and the previous sound feature, at least adjusting a confidence threshold and deciding whether to wake up one of the electronic devices.

In order to better understand the above and other aspects of the present invention, the preferred embodiments are described below, and in conjunction with the drawings, the detailed description is as follows:

S101~S110, S201~S214, S301~S314, S401~S411, S501~S515‧‧‧ process steps

100‧‧‧Electronic devices

101‧‧‧Storage device

102‧‧‧Sound receiver

103‧‧‧ processor

104‧‧‧User interface

1021‧‧‧Microphone

1022‧‧‧Voice Activity Detection Circuit

1023‧‧‧ Analog Digital Converter

FIG. 1 is a block diagram of an electronic device according to an embodiment of the invention.

FIG. 2 is a block diagram of an electronic device according to another embodiment of the present invention.

FIG. 3 is a flow chart of a method for waking up an electronic device through voice recognition according to an embodiment of the invention.

Figure 4 is a schematic diagram showing an example of a keyword group speech model.

Figure 5 shows a schematic diagram of the Viterbi algorithm.

6A and 6B are flowcharts showing a method for waking up an electronic device through voice recognition according to another embodiment of the present invention.

Figure 7 shows a schematic diagram of a dynamic time warping algorithm.

8A and 8B are flowcharts showing a method for waking up an electronic device through voice recognition according to another embodiment of the present invention.

FIG. 9 is a flow chart of a method for waking up an electronic device through voice recognition according to another embodiment of the present invention.

10A and 10B are flowcharts showing a method for activating a specific function of an electronic device through voice recognition according to another embodiment of the present invention.

Please refer to FIG. 1 . FIG. 1 is a block diagram of an electronic device 100 according to an embodiment of the invention. The electronic device 100 includes a storage device 101, a sound receiving device 102, a processor 103, and optionally a user interface 104. The storage device 101 is used to store data, and the storage device 101 is, for example, a memory. The sound receiving device 102 is configured to receive sound and convert the analog sound signal into a digital sound signal. The storage device 101 and the sound receiving device 102 are coupled to the processor 103. The processor 103 is configured to receive the digital audio signal output by the sound receiving device 102, capture the sound characteristics of the digital audio signal, and access the storage device 101 and manage The data stored in the storage device 101. The user interface 104 is configured to receive input from a user and provide output information. The user interface 104 is, for example, a touch panel.

Please refer to FIG. 2. FIG. 2 is a block diagram of an electronic device 100 according to another embodiment of the present invention. In this embodiment, the sound receiving device 102 can be implemented, for example, by a microphone 1021, a voice activity detection (VAD) circuit 1022, and an analog to digital converter (1023). The microphone 1021 is for receiving sound. Analog/digital converter 1023 is used to convert analog audio signals into digital audio signals. The voice activity detecting circuit 1022 is configured to detect the sound and send a signal to the processor 103 when the sound is detected. For example, when the voice activity detecting circuit 1022 detects the sound, a signal (for example, an interrupt) is sent to the processor 103 to wake up the processor 103 or notify the processor 103 to cause the processor 103 to process the analog/digital converter. 1023 output digital audio signal.

Please refer to FIGS. 1 and 3 . FIG. 3 is a flow chart of a method for waking up the electronic device 100 through voice recognition according to an embodiment of the invention. In this embodiment, the processor 103 can determine whether the current sound feature of the current audio signal is a false reception or a false rejection. When the current sound feature is a false reception or a false rejection, the current sound feature is stored for subsequent use.

In step S101, the sound receiving device 102 receives a current sound signal. The current sound signal is obtained, for example, based on the sound the user is currently speaking. In step S102, the processor 103 captures the current sound characteristics of the current audio signal.

In step S103, the processor 103 compares the current sound feature with a keyword group speech model through a matching algorithm to obtain a confidence score. That is, the processor 103 compares the degree of similarity between the current sound feature and the keyword group of the keyword group speech model to obtain a confidence score to determine whether the current sound feature can wake up the electronic device 100 based on the confidence score. When the current sound feature is highly similar to the keyword group of the keyword group speech model, the confidence score is high, and the content representing the user's speech is the same as or very similar to the keyword group of the keyword group voice model, so the user may want to wake up the electronic through the voice. Device 100. When the current sound feature is less similar to the keyword group of the keyword group voice model, the confidence score is low, and the content representing the user's speech is different from the keyword group of the keyword group voice model, so the user does not want to wake up the electronic device through voice. 100.

In an embodiment, the matching algorithm may be a Viterbi algorithm. Please refer to Figures 4 and 5, and Figure 4 shows the schematic of the keyword group speech model. Fig. 5 is a schematic diagram showing an example of the Viterbi algorithm. It is assumed that the keyword group speech model includes six states, which are sil, State 0, State 1, State 2, State 3, and sil, wherein the arrows on each state represent the transition of states, and the transition of each state includes a transfer probability ( Not marked). In Fig. 5, the horizontal axis is the current sound frame fr0~fr12, the vertical axis is the state s0~s5, and the states s0~s5 correspond to the six states sil, State 0, State 1, and State of the keyword group speech model, respectively. 2. State 3 and sil. The processor 103 can find an optimal path according to the Viterbi algorithm such that the score of the entire path is the largest, and the sum of the scores of each of the shortest paths is the confidence score. In another embodiment, the matching algorithm may be any algorithm as long as the degree of similarity between the current sound feature and the keyword group in the keyword group speech model can be calculated.

After the confidence score is obtained, proceeding to step S104, the processor 103 determines whether the confidence score is greater than or equal to the confidence threshold. The confidence threshold represents the ease with which the current sound feature can wake up the electronic device 100. When the confidence threshold is lower, it means that the current sound feature and the keyword group in the keyword group voice model are not too high to wake up the electronic device 100, so the electronic device 100 is more easily awakened by the current sound feature. When the confidence threshold is higher, it indicates that the current sound feature is similar to the keyword group in the keyword group voice model to wake up the electronic device 100, so the electronic device 100 is less likely to be awakened by the current sound feature.

When the confidence score is greater than or equal to the confidence threshold, the process proceeds to step S105; when the confidence score is less than the confidence threshold, the process proceeds to step S106. At step S105, the processor 103 wakes up the electronic device 100. When the electronic device 100 is awakened Thereafter, the user can operate the electronic device 100 in various modes of operation. For example, the user can operate the electronic device 100 through voice. In another embodiment, the user operates the electronic device 100 via the button or the user interface 104 provided by the electronic device 100, for example, by operating the electronic device 100 by touching the touch screen. At step S106, the processor 103 does not wake up the electronic device 100.

When the processor 103 determines that the confidence score is greater than or equal to the confidence threshold and wakes up the electronic device 100 (that is, after steps S104 and S105). Next, proceeding to step S107, the processor 103 determines whether the current sound feature corresponds to a false acceptance. The case of misrecognition is that the electronic device 100 is woken up in an unexpected situation. The manner of determining may include whether the user turns off the awake electronic device 100 within a certain time after the current sound feature wakes up the electronic device 100. It is assumed that when the current sound feature wakes up the electronic device 100, the user turns off the awake electronic device 100 within a certain time. In this case, the representative user does not want to wake up the electronic device 100, but the electronic device 100 is awakened by the current sound feature, so that it can be judged as a false acceptance. Next, proceeding to step S108, the processor 103 stores the current sound feature to the storage device 101 and records the current sound feature as corresponding to the false acceptance.

When the processor 103 determines that the confidence score is less than the confidence threshold value without waking up the electronic device 100 (that is, after steps S104 and S106). Next, proceeding to step S109, the processor 103 determines whether the current sound feature corresponds to a false rejection. The case of false rejection is that the user wants to wake up the electronic device 100 through voice, but cannot successfully wake up the electronic device 100. The manner of determining may include that the current sound feature does not wake up the electronic device 100, and the current sound feature is compared with the keyword group voice model. The confidence score obtained is less than the confidence threshold and the difference is within a predetermined range. In this case, since the confidence score is less than the confidence threshold and the difference is within the predetermined range (that is, the confidence score is very close to the confidence threshold), the user wants to wake up the electronic device 100, but the electronic device 100 is not At present, the sound feature is awakened, so it can be judged as a false rejection. Next, proceeding to step S110, the processor 103 stores the current sound feature to the storage device 101 and records the current sound feature as corresponding to the false rejection.

Please refer to FIGS. 1 , 3 , 6A and 6B . FIG. 6A and FIG. 6B are flowcharts illustrating a method for waking up the electronic device 100 through voice recognition according to another embodiment of the present invention. In this embodiment, the processor 103 further determines whether the storage device 101 stores a previous sound feature of a previous audio signal, and determines whether the previous sound feature is based when the storage device 101 stores the previous sound feature. And the current sound feature wakes up the electronic device 100.

After the sound receiving device 102 receives a current sound signal, and the processor 103 captures the current sound characteristic of one of the current sound signals (that is, after steps S201 and S202), the process proceeds to step S211.

In step S211, the processor 103 determines whether the previous sound feature of one of the previous audio signals is stored in the storage device 101. When the storage device 101 stores the previous sound feature, the process proceeds to step S212; when the storage device 101 does not store the previous sound feature, the process proceeds to step S203.

At step S212, the processor 103 calculates a similarity between the current sound feature and the previous sound feature. That is, since the storage device 101 stores the previous sound feature, the processor 103 determines whether to wake up the electronic device 100 based on the previous sound feature and the current sound feature. In an embodiment, the processor 103 can calculate the similarity between the previous sound feature and the current sound feature through a Dynamic Time Warping (DTW) algorithm. Please refer to FIG. 7 , which shows a schematic diagram of a dynamic time warping algorithm. In Fig. 7, P = P ₁ , ..., P _s , ..., P _k , P _s = (i _s , j _s ), where P is a warping function. The processor 103 can calculate the shortest distance between the current sound feature vector and the previous sound feature vector through a dynamic time warping algorithm, and the shortest distance is the similarity. In another embodiment, the processor 103 can calculate the similarity between the previous sound feature and the current sound feature through any algorithm, as long as the similarity between the two sound features can be calculated. limit.

After calculating the similarity, proceeding to step S213, the processor 103 determines whether the similarity between the current sound feature and the previous sound feature is greater than or equal to a similarity threshold. When the similarity is greater than or equal to the similarity threshold, it indicates that the similarity between the current sound feature and the previous sound feature is high. That is, the content currently spoken by the user is highly similar to the content of the previous speech; when the similarity is less than the similarity threshold, the similarity between the current sound feature and the previous sound feature is low. That is, the content currently spoken by the user is less similar to the content of the previous conversation. When the similarity is greater than or equal to the similarity threshold, the process proceeds to step S214; and when the similarity is less than the similarity threshold, the process proceeds to step S203.

In step S214, the processor 103 determines whether to wake up the electronic device 100 based on the case where the previous sound feature corresponds to the erroneous acceptance or the erroneous rejection. Also That is to say, since the similarity between the current sound feature and the previous sound feature is high, the processor 103 determines whether to wake up the electronic device 100 based on the previous sound feature.

The current sound feature corresponds to the case of false acceptance, and the current sound feature and the previous sound feature are greater than or equal to the similarity threshold (that is, the similarity between the current sound feature and the previous sound feature is high). The processor 103 does not wake up the electronic device 100. Since the previous sound feature corresponds to the case of misacceptance, it indicates that the user's previous speech does not wake up the electronic device 100, but wakes up the electronic device 100. Therefore, if the current sound feature is corresponding to the case of the false acceptance and the content currently spoken by the user is similar to the content of the previous conversation, the processor 103 can determine that the user does not currently want to wake up the electronic device 100, so the electronic device is not woken up. 100.

The current sound feature corresponds to the case of false rejection, and the current sound feature and the previous sound feature are greater than or equal to the similarity threshold (that is, the similarity between the current sound feature and the previous sound feature is high). The processor 103 wakes up the electronic device 100. Since the previous sound feature corresponds to the case of false rejection, it means that the content that the user spoke last time actually wants to wake up the electronic device 100, but does not successfully wake up the electronic device 100. Therefore, if the current sound feature corresponds to the case of false rejection and the content currently spoken by the user is similar to the content of the previous conversation, the processor 103 can determine that the user currently wants to wake up the electronic device 100, so that the electronic device 100 is woken up.

In step S203, since the storage device 101 does not store the previous sound feature or the similarity between the current sound feature and the previous sound feature is less than similar The threshold value is not determined by the processor 103. Instead of waking up the electronic device 100 based on the previous sound feature and the current sound feature, the processor 103 determines whether to wake up the electronic device 100 based on the current sound feature and the keyword group voice model. The process of the processor 103 determining whether to wake up the electronic device 100 based on the current sound feature and the keyword group voice model is as follows. In step S203, the processor 103 compares the current sound feature with the keyword group voice model through a matching algorithm to obtain a confidence score. Next, proceeding to step S204, the processor 103 determines whether the confidence score is greater than or equal to the confidence threshold, to decide to wake up the electronic device 100 (step S205) or not to wake up the electronic device 100 (step S206), and determine that the current sound feature is falsely accepted. Case (step S207) or case of erroneous rejection (step S210). When the current sound feature is a case of false acceptance or a case of false rejection, the current sound feature is stored (step S208 or step S210).

In this way, when the storage device 101 stores the previous sound feature, the processor 103 can determine whether to wake up the electronic device by the similarity between the current sound feature and the previous sound feature and the corresponding situation of the previous sound feature. The device 100 (such as steps S211 to S214) does not need to compare the current sound feature with the keyword group voice model (step S203), so that the amount of calculation can be reduced, and the efficiency and accuracy of the voice wake-up electronic device 100 can be improved.

Please refer to FIGS. 1 , 6A , 6B , 8A and 8B . FIG. 8A and FIG. 8B are flowcharts illustrating a method for waking up the electronic device 100 through voice recognition according to another embodiment of the present invention. In this embodiment, the processor 103 can adjust the confidence threshold according to the case where the previous sound feature corresponds to the error acceptance or the case of the false rejection.

Steps S301, S302, S311, S312, and S313 are similar to the 6A Steps S201, S202, S211, S212, and S213 of the figure. The difference is that, in step S313, the processor 103 determines that the similarity between the current sound feature and the previous sound feature is greater than or equal to the similarity threshold, and proceeds to step S314, the processor 103 correspondingly accepts the error according to the previous sound feature. In the case of a situation or a false rejection, adjust the threshold of confidence.

In detail, in step S314, the current sound feature corresponds to the case of misacceptance, and the current sound feature and the previous sound feature are greater than or equal to the similarity threshold (that is, the current sound feature and the previous sound feature). When the similarity between the two is high, the processor 103 will increase the confidence threshold. The reason is that the content of the previous voice is corresponding to the case of misacceptance, indicating that the content of the user's previous speech does not wake up the electronic device 100, but wakes up the electronic device 100. In this case, it is very likely that because the confidence threshold is too low, the sound feature is likely to wake up the electronic device 100, so the confidence threshold is raised to reduce the occurrence of false reception.

The current sound feature corresponds to the case of false rejection, and the current sound feature and the previous sound feature are greater than or equal to the similarity threshold (that is, the similarity between the current sound feature and the previous sound feature is high). The processor 103 will lower the confidence threshold. The reason is that since the previous sound feature corresponds to the case of false rejection, it means that the content of the previous time the user actually wants to wake up the electronic device 100, but does not successfully wake up the electronic device 100. In this case, it is very likely that because the confidence threshold is too high, the sound feature does not easily wake up the electronic device 100, so the confidence threshold is lowered to reduce the occurrence of false rejection.

After step S314 is performed, the process proceeds to step S303. At The processor 103 determines whether to wake up the electronic device 100 according to the current sound feature and the keyword group voice model. The flow is as follows. In step S303, the processor 103 compares the current sound feature with the keyword group voice model through the matching algorithm to obtain a confidence score. Next, proceeding to step S304, the processor 103 determines whether the confidence score is greater than or equal to the confidence threshold, to decide to wake up the electronic device 100 (step S305) or not to wake up the electronic device 100 (step S306), and determine that the current sound feature is falsely accepted. Case (step S307) or case of false rejection (step S310). When the current sound feature is a case of false acceptance or a case of false rejection, the current sound feature is stored (step S308 or step S310).

In this embodiment, when the storage device 101 stores the previous sound feature, the processor 103 can adjust the confidence threshold by the similarity between the current sound feature and the previous sound feature and the corresponding situation of the previous sound feature. . When the confidence threshold is raised, the false reception can be reduced; when the confidence threshold is lowered, the false rejection can be reduced. In this way, by adjusting the confidence threshold, the situation of false reception and false rejection can be effectively improved to increase the probability and accuracy of successfully waking up the electronic device 100 under the expected situation.

Please refer to FIG. 1 , FIG. 3 and FIG. 9 . FIG. 9 is a flow chart of a method for waking up the electronic device 100 through voice recognition according to another embodiment of the present invention. In this embodiment, the processor 103 can adjust the confidence threshold based on the user interaction. Steps S401 to S410 are similar to steps S101 to S110 of FIG. The difference is that the processor 103 determines that the current sound feature corresponds to the error acceptance, and stores the current sound feature to the storage device 101 (steps S407 and S408), or The processor 103 determines that the current sound feature corresponds to the erroneous rejection, and stores the current sound feature to the storage device 101 (steps S409 and S410), and proceeds to step S411. In step S411, the processor 103 can adjust the confidence threshold according to the user interaction. For example, the processor 103 can determine whether the electronic device 100 is turned off by the user immediately after being awake (that is, the case of misacceptance), or can determine whether the user wants to wake up the electronic device through the same sound content every time. Device 100, but none of them succeeded (ie, in the case of false rejection). If the above situation occurs continuously, it is likely that the confidence threshold is too high or too low. Therefore, the processor 103 can determine whether to adjust the confidence threshold according to whether the number of consecutive occurrences of the above situation is excessive.

Specifically, the processor 103 determines whether the number of consecutive occurrences of the false acceptance or the number of consecutive occurrences of the false rejection is greater than a threshold value to determine whether the confidence threshold needs to be adjusted. The threshold value can be defined by the designer. When the number of consecutive erroneous acceptances is greater than the threshold value, it is likely that the confidence threshold is too low and the sound feature is likely to wake up the electronic device 100, and the processor 103 raises the confidence threshold to reduce the occurrence of erroneous reception. When the number of consecutive false rejections is greater than the threshold value, it is likely that the confidence threshold is too high and the sound feature does not easily wake up the electronic device 100, and the processor 103 lowers the confidence threshold to reduce the occurrence of false rejection. .

In this way, the processor 103 can adjust the confidence threshold according to the user interaction to reduce the occurrence of false acceptance or false rejection.

Please refer to Figures 1 and 10A and 10B, Figures 10A and 10B A flowchart of a method of initiating a specific function of an electronic device 100 through voice recognition according to another embodiment of the present invention. In this embodiment, after the electronic device 100 is woken up, the processor 103 can further activate the specific function of the electronic device 100 through voice recognition. That is, after the electronic device 100 is woken up, the sound receiving device 102 further receives a current audio signal. Next, the processor 103 further determines whether the previous sound feature of one of the previous audio signals is stored in the storage device 101. When the storage device 101 stores a sound feature before the previous audio signal, the processor 103 determines whether to activate the specific function of the electronic device 100 according to the previous sound feature. When a sound feature preceding the previous audio signal is not stored in the storage device 101, the processor 103 determines whether to activate the specific function of the electronic device 100 based on the current sound feature.

In step S501, the electronic device 100 is woken up. In one embodiment, the user wakes up the electronic device 100 by pressing a button, or wakes up the electronic device 100 by touching the touch screen. In another embodiment, the method for the user to wake up the electronic device 100 through voice and wake up the electronic device 100 through voice is as described above, and will not be described here.

In step S502, the sound receiving device 102 receives a current sound signal. The user can speak a particular voice command to activate a particular function of the electronic device 100. For example, a specific voice command may include at least one of "Pairing", "Check battery", and "Am I connected". At present, the sound signal is the sound signal corresponding to the specific voice command spoken by the user. In step S503, the processor 103 captures one of the current sound characteristics of the current audio signal.

Next, proceeding to step S504, the processor 103 determines whether the previous sound feature of one of the previous audio signals is stored in the storage device 101. When the storage device 101 stores the previous sound feature, the process proceeds to step S505; when the storage device 101 does not store the previous sound feature, the process proceeds to step S508.

At step S505, the processor 103 calculates a similarity between the current sound feature and the previous sound feature. After calculating the similarity, proceeding to step S506, the processor 103 determines whether the similarity between the current sound feature and the previous sound feature is greater than or equal to a similarity threshold. When the similarity is greater than or equal to the similarity threshold, the process proceeds to step S507; when the similarity is less than the similarity threshold, the process proceeds to step S508.

In step S507, the processor 103 determines whether to activate the specific function of the electronic device 100 based on the case where the previous sound feature corresponds to the erroneous acceptance or the erroneous rejection. When the current sound feature corresponds to the case of false acceptance, and the current sound feature and the previous sound feature are greater than or equal to the similarity threshold, the processor 103 does not activate the specific function of the electronic device 100. Since the previous sound feature corresponds to the case of misacceptance, it indicates that the content of the user's previous speech is not to activate the specific function of the electronic device 100, but activates the specific function of the electronic device 100. Therefore, if the current sound feature corresponds to the case of the false acceptance and the content currently spoken by the user is similar to the content of the previous conversation, the processor 103 can determine that the user does not currently want to activate the specific function of the electronic device 100, so The particular function of the electronic device 100 is activated. The current sound feature corresponds to the case of false rejection, and the current sound feature and the previous sound feature are greater than or equal to the similarity threshold, then The processor 103 activates a particular function of the electronic device 100. Since the previous sound feature corresponds to the case of false rejection, it indicates that the user's previous speech actually wants to activate the specific function of the electronic device 100, but does not successfully activate the specific function of the electronic device 100. Therefore, if the current sound feature corresponds to the case of false rejection and the content currently spoken by the user is similar to the content of the previous conversation, the processor 103 can determine that the user currently wants to activate the specific function of the electronic device 100, thereby starting the electronic device. 100 specific features. For example, when the specific voice command corresponding to the current sound feature is “Pairing”, the previous sound feature corresponds to the case of false rejection, and the current sound feature is greater than or equal to the previous sound feature. When the similarity threshold is reached, the processor 103 starts a function of confirming whether the wireless communication pairing of the electronic device 100 with other electronic devices is successful according to the current sound feature.

In step S508, the processor 103 compares the current sound feature with the keyword group speech model through a matching algorithm to obtain a confidence score. That is, when the storage device 101 does not store the previous sound feature or the similarity between the current sound feature and the previous sound feature is low, the processor 103 does not need to decide whether to activate the electronic device 100 according to the previous sound feature. Specific features. The processor 103 compares the current sound characteristics with the keyword group speech model to determine whether to initiate a particular function of the electronic device 100 based on the current sound characteristics.

After the confidence score is obtained in step S508, proceeding to step S509, the processor 103 determines whether the confidence score is greater than or equal to the confidence threshold. When the confidence score is greater than or equal to the confidence threshold, the process proceeds to step S513; when the confidence score is less than the confidence threshold, the process proceeds to step S510.

At step S513, the processor 103 activates a particular function of the electronic device 100 based on the current sound characteristics. For example, the processor 103 activates at least a function of confirming whether the wireless communication pairing of the electronic device 100 and other electronic devices is successful according to the current sound feature, starts a function of checking the battery power of the electronic device 100, or starts checking the network of the electronic device 100. One of the features of the connection. More specifically, if the specific voice command corresponding to the current voice feature received by the voice receiving device 102 is “Pairing”, the processor 103 starts to confirm the wireless of the electronic device 100 and other electronic devices according to the current voice feature. The function of communication pairing success. Assuming that the specific voice command corresponding to the current voice feature received by the voice receiving device 102 is "Check Battery", the processor 103 activates the function of checking the battery power of the electronic device 100 based on the current voice feature. Assuming that the specific voice command corresponding to the current voice feature received by the voice receiving device 102 is "Am I connected", the processor 103 starts to check whether the network of the electronic device 100 is connected according to the current voice feature. Features.

Next, proceeding to step S514, the processor 103 determines whether the current sound feature corresponds to a false acceptance. The case of misrecognition is that the processor 103 initiates a particular function of the electronic device 100 in an unexpected situation. The manner of determining may include whether the user turns off the specific function within a specific time after the processor 103 activates the specific function of the electronic device 100 through the current sound feature. If the user turns off the specific function within a certain time, it means that the user does not want to activate the specific function, but the specific function is activated by the current sound feature, so it can be judged as a false acceptance. Next, proceeding to step S515, the processor 103 stores the current sound feature to the storage device 101 and record the current sound characteristics as corresponding to the case of false acceptance.

At step S510, the processor 103 does not initiate a particular function of the electronic device 100 based on the current sound characteristics. Next, proceeding to step S511, the processor 103 determines whether the current sound feature corresponds to a false rejection. The case of false rejection is that the user wants to activate the specific function of the electronic device 100 through the current sound, but cannot successfully start the specific function. The manner of determining may include that the current sound feature does not activate the specific function of the electronic device 100, and the current sound feature is compared with the keyword group voice model to obtain a confidence score that is less than the confidence threshold and the difference is within a predetermined range. In this case, since the confidence score is less than the confidence threshold and the difference is within the predetermined range (that is, the confidence score is very close to the confidence threshold), the user wants to activate the specific function of the electronic device 100 through the current sound, but The specific function of the electronic device 100 is not activated by the current sound feature, so it can be judged as a false rejection. Next, proceeding to step S512, the processor 103 stores the current sound feature to the storage device 101 and records the current sound feature as corresponding to the false rejection.

In this embodiment, speech recognition can be applied to activate a particular function of the electronic device 100. After the electronic device 100 is woken up, the user can activate a specific function of the electronic device 100 through voice recognition, and the processor 103 can activate a specific function of the electronic device 100 according to the current sound feature, and can record that the current sound feature is incorrectly accepted. The specific function is activated by the situation, or it is falsely rejected without starting the specific function. Therefore, when the electronic device 100 receives the next sound of the user, if the previous sound feature has been stored, the processor 103 can calculate the similarity between the sound feature of the next sound and the stored sound feature, and the stored sound. The situation corresponding to the levy determines whether to activate the specific function of the electronic device 100, and does not need to compare the keyword group voice model, thereby improving the efficiency of starting the specific function of the electronic device 100 through the voice.

The method for waking up an electronic device through voice recognition according to the above embodiment of the present invention, when the storage device stores the previous sound feature, may be based on the previous voice feature corresponding to the situation of false acceptance or false rejection, and the current sound feature and The similarity between the previous sound features determines whether or not the electronic device is woken up, thereby improving the efficiency of waking up the electronic device. In addition, the present invention can also adjust the confidence threshold according to the situation that the previous sound feature corresponds to the error acceptance or the false rejection, and the similarity between the current sound feature and the previous sound feature, so as to reduce the false acceptance or error. The probability of rejection is not required to adjust the keyword group speech model or the pre-trained keyword group speech model to improve the situation of false acceptance or false rejection. Therefore, the adjustment complexity is low, and the error acceptance or error can be effectively reduced. The situation of rejection is to increase the probability and accuracy of successfully waking up the electronic device.

In conclusion, the present invention has been disclosed in the above preferred embodiments, and is not intended to limit the present invention. A person skilled in the art can make various changes and modifications without departing from the spirit and scope of the invention. Therefore, the scope of the invention is defined by the scope of the appended claims.

S101, S102, S103, S104, S105, S106, S107, S108, S109, S110‧‧‧ process steps

Claims (21)

A method for waking up an electronic device through voice recognition, comprising: receiving a current voice signal; performing a voice recognition wake-up algorithm; receiving and determining a user feedback; and adjusting the voice recognition wake-up algorithm. A method for waking up an electronic device through voice recognition, comprising: receiving a current audio signal; capturing a current sound characteristic of one of the current audio signals; determining whether a previous sound characteristic of one of the previous audio signals is stored; wherein when storing When there is the previous sound feature, according to the previous sound feature, a case of False Acceptance or a case of false rejection (False Rejection) and one of the current sound feature and the previous sound feature are Similarity, at least adjust a confidence threshold and decide whether to wake up one of the electronic devices. The method of claim 2, wherein when the previous sound characteristic is the case of the false acceptance and the similarity is greater than or equal to a similarity threshold, the confidence threshold is raised or the electronic is not woken up. Device. The method of claim 2, wherein when the previous sound feature is the case of the false rejection and the similarity is greater than or equal to a similarity threshold, the confidence threshold is lowered or the electronic device is awakened. . The method of claim 2, wherein the similarity is calculated by a Dynamic Time Warping (DTW) algorithm. The method of claim 2, wherein when the previous sound feature of the previous audio signal is not stored, the method further comprises: comparing the current sound feature with a keyword group voice through a matching algorithm Modeling to obtain a confidence score; determining whether the confidence score is greater than or equal to the confidence threshold; when the confidence score is greater than or equal to the confidence threshold, waking up the electronic device; and when the confidence score is less than the confidence threshold, The electronic device is not woken up. The method of claim 6, wherein after waking up the electronic device, the method further comprises: determining whether the current sound feature is a false acceptance; wherein when the current sound feature is the false acceptance, The current sound feature is stored and the confidence threshold is adjusted based on a user interaction. The method of claim 6, wherein after the electronic device is not woken up, the method further comprises: determining whether the current sound feature is a false rejection; wherein when the current sound feature is the false rejection And storing the current sound feature and adjusting the confidence threshold based on a user interaction. The method of claim 8, wherein the step of determining whether the current sound feature is the erroneous rejection comprises: determining whether the difference between the confidence score and the confidence threshold is within a predetermined range ; When the difference between the confidence score and the confidence threshold is within the predetermined range, it is determined that the current sound feature is the false rejection. The method of claim 6, wherein the matching algorithm is a Viterbi algorithm. The method of claim 7 or 8, wherein the user interaction comprises a number of consecutive false acceptances or a number of consecutive false rejections, in the step of adjusting the confidence threshold based on the user interaction, including : determining whether the number of consecutive false acceptances is greater than a threshold value or determining whether the number of consecutive false rejections is greater than the threshold value; and when the number of consecutive false acceptances is greater than the threshold threshold, increasing the confidence threshold; And when the number of consecutive false rejections is greater than the threshold of the number of times, the confidence threshold is lowered. An electronic device comprising: a storage device; a sound receiving device for receiving a current sound signal; and a processor for capturing a current sound characteristic of the current sound signal and determining whether the storage device has a storage device a sound feature of a previous sound signal, when the storage device stores the previous sound feature, the processor corresponds to a false acceptance or a false rejection according to the previous sound feature and the current A similarity between the sound feature and the previous sound feature, at least adjusting a confidence threshold and determining whether to wake up one of the electronic devices. The electronic device of claim 12, wherein the processor is configured to increase the confidence threshold when the previous sound characteristic is the false acceptance and the similarity is greater than or equal to a similarity threshold Or not wake up the electronic device. The electronic device of claim 12, wherein the processor is configured to lower the confidence threshold when the previous sound feature is the false rejection and the similarity is greater than or equal to a similarity threshold Or wake up the electronic device. The electronic device of claim 12, wherein the processor is configured to calculate the similarity through a Dynamic Time Warping (DTW) algorithm. The electronic device of claim 12, wherein the processor is configured to compare the current sound feature with a keyword group speech model by a matching algorithm to obtain a confidence score, and determine whether the confidence score is greater than or equal to the The confidence threshold is devaluated; wherein when the confidence score is greater than or equal to the confidence threshold, the processor wakes up the electronic device, and when the confidence score is less than the confidence threshold, the processor does not wake up the electronic device. The electronic device of claim 16, wherein after the processor wakes up the electronic device, the processor is further configured to determine whether the current sound feature is a false acceptance, and when the current sound feature is In the case of the mis-acceptance, the processor stores the current sound feature to the storage device and adjusts the confidence threshold based on a user interaction. The electronic device of claim 16, wherein the processing is After the device does not wake up the electronic device, the processor is further configured to determine whether the current sound feature is a false rejection, and when the current sound feature is the false rejection, the processor stores the current sound feature. The confidence threshold is adjusted to the storage device and based on a user interaction. The electronic device of claim 18, wherein the processor determines whether the difference between the confidence score and the confidence threshold is within a predetermined range, when the difference between the confidence score and the confidence threshold is Within the predetermined range, the processor determines that the current sound feature is the false rejection. The electronic device of claim 16, wherein the matching algorithm is a Viterbi algorithm. The electronic device of claim 17 or 18, wherein the user interaction includes a number of consecutive false acceptances or a number of consecutive false rejections, and the processor is further configured to determine whether the number of consecutive false acceptances is greater than a threshold value or a judgment of whether the number of consecutive false rejections is greater than the threshold value, and when the number of consecutive false acceptances is greater than the threshold value, the confidence threshold is raised, and when the number of consecutive false rejections is greater than the threshold When the threshold is exceeded, the confidence threshold is lowered.