WO2017154358A1 - Speech recognition device and speech recognition program - Google Patents

Abstract

This speech recognition device 10 comprises: a speech input unit 12 to which a user's speech is input; a speech storage unit 16 for storing an input speech; a speech recognition processing unit 21 for recognizing the input speech; a likelihood calculation processing unit 22 for calculating the likelihood of the recognition results of a speech; a validity determination processing unit 23 for determining that the recognition results of the speech are valid if the calculated likelihood is greater than or equal to a prescribed threshold; a similarity determination processing unit 24 for determining whether or not the stored previous speech and the present speech are identical on the basis of the degree of coincidence between the previous speech and the present speech; and a threshold adjustment processing unit 25 for decreasing the threshold by a prescribed adjustment value if the similarity determination processing unit 24 determines that the previous speech and the present speech are identical.

Description

Voice recognition device, voice recognition program Cross-reference of related applications

This application is based on Japanese Patent Application No. 2016-043348 filed on March 7, 2016, the contents of which are incorporated herein by reference.

This disclosure relates to a speech recognition apparatus and a speech recognition program.

For example, in a voice recognition device mounted on a vehicle, a correct voice recognition result may not be obtained due to ambient noise or a user's mistake. Therefore, a speech recognition device having a function of prompting the user to speak again when a correct speech recognition result cannot be obtained has been considered. For example, Patent Document 1 discloses a technique for determining whether or not the reliability of a user's voice, that is, the likelihood is equal to or higher than a predetermined threshold, and performing voice recognition when the reliability is higher than or equal to a predetermined threshold. Has been. Patent Document 1 discloses a technique for changing a threshold value according to the number of utterances of a user and the conversation time.

JP 2008-9153 A

According to the conventional technique described in Patent Document 1, it is possible to perform voice recognition based on a threshold value corresponding to the number of utterances of the user and the conversation time. However, in the conventional technique in which the threshold value is changed according to the “number of utterances” and the “time” of the dialogue, for example, even when the user utters different words between the previous time and this time, the “number of utterances” and the dialogue If the “time” satisfies the predetermined condition, the threshold value is changed. As a case where a correct speech recognition result cannot be obtained and the user is prompted to speak again, it is usually assumed that the user repeats the same words as the previous time. Therefore, from the viewpoint of obtaining a correct speech recognition result by re-speaking, it is sufficient to change the threshold when the user repeats the same word. Therefore, in the related art in which the threshold value is changed even when the user utters different words, the threshold value may be changed even though the threshold value need not be changed.

Therefore, the present disclosure relates to the reliability of speech uttered by a user, that is, a criterion for determination in a speech recognition apparatus and a speech recognition program that determine the effectiveness of speech recognition based on the magnitude relationship between likelihood and a predetermined threshold. It is an object of the present invention to provide a configuration that can appropriately change the threshold value.

In one aspect of the present disclosure, the speech recognition apparatus includes a speech input unit, a speech storage unit, a speech recognition processing unit, a likelihood calculation processing unit, a validity determination processing unit, an identical determination processing unit, and a threshold adjustment processing unit. A voice of the user is input to the voice input unit. The voice storage unit stores voice input to the voice input unit. The voice recognition processing unit recognizes the voice input to the voice input unit. The likelihood calculation processing unit calculates the likelihood of the speech recognition result by the speech recognition processing unit. The validity determination processing unit determines that the speech recognition result by the speech recognition processing unit is valid when the likelihood calculated by the likelihood calculation processing unit is equal to or greater than a predetermined threshold. The same determination processing unit determines whether or not the previous voice and the current voice are the same based on the degree of coincidence between the previous voice and the current voice stored in the voice storage unit. The threshold adjustment processing unit lowers the threshold by a predetermined adjustment value when the same determination processing unit determines that the previous speech and the current speech are the same.

In one aspect of the present disclosure, a speech recognition program is provided in a speech recognition apparatus that includes a speech input unit that receives user speech and a speech storage unit that stores speech input to the speech input unit. The likelihood calculation process, the validity determination process, the same determination process, and the threshold adjustment process are executed. The voice recognition process recognizes voice input to the voice input unit. In the likelihood calculation process, the likelihood of the speech recognition result by the speech recognition process is calculated. The validity determination process determines that the voice recognition result by the voice recognition process is valid when the likelihood calculated by the likelihood calculation process is equal to or greater than a predetermined threshold. The same determination process determines whether or not the previous voice and the current voice are the same based on the degree of coincidence between the previous voice and the current voice stored in the voice storage unit. The threshold adjustment process lowers the threshold by a predetermined adjustment value when it is determined by the same determination process that the previous voice and the current voice are the same.

According to the speech recognition apparatus and the speech recognition program according to the present disclosure, a threshold value that is a criterion for determining whether the speech recognition result is valid or invalid is an appropriate value only when the user repeatedly inputs the same word. Can be changed.

The above and other objects, features and advantages of the present disclosure will become more apparent from the following detailed description with reference to the accompanying drawings. The drawing
FIG. 1 is a diagram schematically illustrating a configuration example of a speech recognition apparatus according to the present embodiment. FIG. 2 is a diagram showing an example of a voice recognition screen. FIG. 3 is a flowchart illustrating an operation example of the speech recognition apparatus.

Hereinafter, an embodiment of a speech recognition apparatus will be described with reference to the drawings. A voice recognition device 10 illustrated in FIG. 1 is mounted on a vehicle, for example, and includes a control unit 11, a voice input unit 12, a voice output unit 13, a display output unit 14, an operation input unit 15, a storage unit 16, and the like. Prepare. The control unit 11 is configured mainly with a microcomputer (not shown), and controls the overall operation of the speech recognition apparatus 10.

The voice input unit 12 includes a microphone (not shown) and the like, and inputs the user's voice. The voice input unit 12 converts the input voice into voice data and outputs the voice data to the control unit 11. The control unit 11 stores the audio data input from the audio input unit 12 in the storage unit 16. The storage unit 16 is an example of an audio storage unit, and is configured of a storage medium such as a hard disk drive, for example. The storage unit 16 stores various types of data necessary for speech recognition processing such as a speech recognition dictionary database, in addition to speech speech data input from the speech input unit 12.

The voice output unit 13 includes a speaker (not shown) and outputs various information such as a voice recognition result by voice based on a voice output signal input from the control unit 11. The display output unit 14 includes a display panel (not shown), and displays various screens such as a voice recognition screen G shown in FIG. 2 based on a display output signal input from the control unit 11. The voice recognition screen G exemplifies commands that can be recognized, for example, and the user issues various voices, that is, commands for the voice recognition device 10 with reference to the commands shown on the voice recognition screen G. Can do. Note that the commands shown on the speech recognition screen G are merely examples, and the speech recognition apparatus 10 can recognize speech that is not shown on the speech recognition screen G. The operation input unit 15 includes an operation switch (not shown). When the operation switch is operated by the user, the operation content is input to the control unit 11.

The control unit 11 executes a speech recognition control program, which is an example of a speech recognition program, so that a speech recognition processing unit 21, a likelihood calculation processing unit 22, a validity determination processing unit 23, an identical determination processing unit 24, a threshold value The adjustment processing unit 25 is virtually realized by software. Note that these processing units may be realized by hardware, or may be realized by a combination of software and hardware.

The voice recognition processing unit 21 specifies the degree of coincidence between the two voice data by comparing the voice data of the voice input to the voice input unit 12 with the voice data stored in the dictionary database for voice recognition. . The voice recognition processing unit 21 recognizes that the input voice is the voice of the voice data of the collated dictionary database when the two voice data are completely or almost the same. It should be noted that the two audio data substantially coincide with each other means that the audio data coincide with each other to the extent that they can be regarded as being completely coincident, for example, when both audio data coincide with 80% to 90% or more.

The likelihood calculation processing unit 22 calculates the likelihood of the speech recognition result by the speech recognition processing unit 21. That is, the likelihood calculation processing unit 22 calculates a higher likelihood as the matching degree is higher according to the matching degree of the both voice data at the time of the voice recognition processing by the voice recognition processing unit 21, and lower likelihood as the matching degree is lower. Calculate the degree.

The validity determination processing unit 23 determines that the speech recognition result by the speech recognition processing unit 21 is valid when the likelihood calculated by the likelihood calculation processing unit 22 is equal to or greater than a predetermined threshold T. Is smaller, the speech recognition result by the speech recognition processing unit 21 is determined to be invalid. When the validity determination processing unit 23 determines that the voice recognition result by the voice recognition processing unit 21 is invalid, the determination reason, for example, information indicating that the user's voice is too low or too high, Invalid determination reason information such as information indicating that the user's voice is too fast or too slow is output.

The invalidity determination reason information may be output by, for example, a voice by the voice output unit 13, may be output by a screen display by the display output unit 14, or an auditory output and display output unit 14 by the voice output unit 13. It may be performed by both visual output by. Also, the loudness and speed of the voice can be specified by comparing the signal level of the voice data output from the voice input unit 12 with a threshold for judging the magnitude and a threshold for judging the speed, for example.

Based on the degree of coincidence between the voice data of the previous input voice stored in the storage unit 16 and the voice data of the current input voice, the same determination processing unit 24 determines whether the voice inputted last time and the voice inputted this time Are determined to be the same. That is, the same determination processing unit 24 specifies the degree of coincidence between the voice data of the previous input voice and the voice data of the current input voice. The identical determination processing unit 24 determines that the previously input sound and the sound input this time are the same when the two sound data are completely or approximately the same. It should be noted that the two audio data substantially coincide with each other means that the audio data coincide with each other to the extent that they can be regarded as being completely coincident, for example, when both audio data coincide with 80% to 90% or more.

The threshold adjustment processing unit 25 lowers the above threshold T by a predetermined adjustment value when the same determination processing unit 24 determines that the previous sound and the current sound are the same. The threshold adjustment processing unit 25 is configured to change the adjustment value according to the likelihood calculated by the likelihood calculation processing unit 22. In this case, the threshold adjustment processing unit 25 increases the adjustment value as the calculated likelihood is higher. The threshold adjustment processing unit 25 is configured to change the adjustment value according to the number of times that the likelihood calculated by the likelihood calculation processing unit 22 is equal to or greater than the threshold T in a plurality of speech recognition processes in the past. ing. In this case, the threshold adjustment processing unit 25 increases the adjustment value as the number of times that the calculated likelihood becomes equal to or greater than the threshold T increases.

Next, an operation example of the speech recognition apparatus 10 will be described. As illustrated in FIG. 3, when the voice recognition device G displays the voice recognition screen G, the voice recognition device 10 monitors whether or not voice is input (S1). Then, when voice is input (S1: YES), the voice recognition device 10 stores the voice data (S2). Then, the voice recognition device 10 monitors again whether or not voice is input in a state where the same voice recognition screen G as that at the time of step S1 is displayed (S3). When the voice is input again (S3: YES), the voice recognition device 10 stores the voice data (S4), and proceeds to step S5. Note that the voice recognition device 10 proceeds to step S1 when no voice is input again within a predetermined waiting time (S3: NO).

When the speech recognition apparatus 10 proceeds to step S5, the speech recognition processing unit 21 performs speech recognition processing. And the speech recognition apparatus 10 performs the likelihood calculation process by the likelihood calculation process part 22 (S6). And the speech recognition apparatus 10 performs the same determination process by the same determination process part 24 (S7). If the speech recognition apparatus 10 determines that the previously input speech and the current input speech are the same (S7: YES) by the same determination process, the speech recognition apparatus 10 decreases the threshold T (S8). Note that an example of threshold adjustment processing is configured by step S8. Further, the speech recognition apparatus 10 outputs that the threshold value T has been lowered, that is, that the judgment reference value for judging whether the speech recognition result is valid or invalid is adapted to the user's speech mode (S9), The process proceeds to step S10.

Note that the output indicating that the determination reference value is adapted may be output by voice by the voice output unit 13, may be output by screen display by the display output unit 14, or may be output by the voice output unit 13. You may perform both by an audio output and the visual output by the display output part 14. FIG. If the speech recognition apparatus 10 determines that the previously input speech and the current input speech are not the same (S7: NO) by the same determination processing, step S8 and S9 are not performed. The process proceeds to S10.

When the speech recognition apparatus 10 proceeds to step S10, the speech recognition apparatus 10 determines whether or not the likelihood calculated in step S6 is greater than or equal to a threshold value T. Step S10 is an example of a validity determination process. If the likelihood is greater than or equal to the threshold T (S10: YES), the speech recognition apparatus 10 determines that the speech recognition result in step S5 is valid (S11), and ends this control. On the other hand, when the likelihood is smaller than the threshold value T (S10: NO), the speech recognition apparatus 10 determines that the speech recognition result in step S5 is invalid (S12). Then, the speech recognition apparatus 10 outputs invalidity determination reason information (S13) and ends this control. An example of the validity determination process is configured by steps S10, S11, and S12.

According to the speech recognition apparatus 10, it is possible to determine that the previous speech and the current speech are the same when the user repeatedly inputs speech while the same speech recognition screen G is displayed. In this case, the threshold value T, which is a determination reference value for determining whether the speech recognition result is valid or invalid, is lowered by a predetermined adjustment value. Thereby, since it becomes easy to determine that the recognition result of the voice is valid, it is possible to avoid a situation where the user has to keep inputting the same voice over and over again. As described above, according to the speech recognition device 10, the determination reference value for determining whether the speech recognition result is valid or invalid can be changed to an appropriate value only when the user repeatedly inputs the same word. it can.

Further, according to the speech recognition apparatus 10, the adjustment value of the threshold T is changed according to the calculated likelihood. In this case, the speech recognition apparatus 10 increases the adjustment value as the calculated likelihood is higher. That is, by setting the threshold T to a smaller value as the likelihood of the speech recognition result is higher, the probability that the speech recognition result having a higher likelihood is determined to be effective can be increased.

Further, according to the speech recognition apparatus 10, the adjustment value of the threshold T is changed according to the number of times that the likelihood calculated in the plurality of speech recognition processes in the past becomes equal to or greater than the threshold T. In this case, the speech recognition apparatus 10 increases the adjustment value as the number of times that the calculated likelihood is equal to or greater than the threshold value T increases. That is, by increasing the number of times that the speech recognition result is determined to be valid during a plurality of speech recognition processes in the past, the threshold T is set to a smaller value. The probability that the speech recognition result is determined to be valid can be increased.

Also, according to the speech recognition apparatus 10, when it is determined that the speech recognition result is invalid, the determination reason is also output. Thereby, the user can improve the speech mode based on the output determination reason, and can increase the probability that the speech recognition result is determined to be valid.

Note that the present disclosure is not limited to the above-described embodiments, and can be applied to various embodiments without departing from the gist thereof.
For example, in step S8, the speech recognition apparatus 10 not only lowers the threshold T, but also performs processing such as reducing the processing speed of the speech recognition processing or increasing the time required for the speech recognition processing. It may be. That is, the speech recognition process can be performed more accurately by reducing the processing speed of the speech recognition process or increasing the processing time. When a correct speech recognition result cannot be obtained and the user repeatedly inputs the same word over and over, it is effective to take measures for performing such accurate speech recognition processing.

Further, the voice recognition device 10 may store at least voice data of the voice input last time. Therefore, the speech recognition apparatus 10 may be configured to erase past speech data in order from the oldest, and in this case, the limited storage capacity of the storage unit 16 can be effectively utilized. In addition, the speech recognition apparatus 10 may be configured to calculate the likelihood in a plurality of stages, for example, three stages of “high”, “medium”, and “low”, or to calculate the likelihood by, for example, continuous numerical values by percentage. It is good.

Further, the speech recognition apparatus 10 may be configured to decrease the adjustment value of the threshold T as the calculated likelihood is higher. Further, the speech recognition apparatus 10 may be configured to decrease the adjustment value of the threshold T as the number of times that the calculated likelihood is equal to or greater than the threshold T in a plurality of past speech recognition processes.
Further, the voice recognition program according to the present embodiment can be stored in, for example, a computer-readable storage medium.

Although the present disclosure has been described with reference to the embodiments, it is understood that the present disclosure is not limited to the embodiments and structures. The present disclosure includes various modifications and modifications within the equivalent range. In addition, various combinations and forms, as well as other combinations and forms including only one element, more or less, are within the scope and spirit of the present disclosure.

Claims (5)

1. A voice input unit (12) through which a user's voice is input;
   A voice storage unit (16) for storing voice input to the voice input unit;
   A speech recognition processing unit (21) for recognizing speech input to the speech input unit;
   A likelihood calculation processing unit (22) for calculating the likelihood of the speech recognition result by the speech recognition processing unit;
   A validity determination processing unit (23) that determines that the speech recognition result by the speech recognition processing unit is valid when the likelihood calculated by the likelihood calculation processing unit is equal to or greater than a predetermined threshold;
   An identical determination processing unit (24) for determining whether or not the previous voice and the current voice are the same based on the degree of coincidence between the previous voice and the current voice stored in the voice storage unit; ,
   A threshold adjustment processing unit (25) for lowering the threshold by a predetermined adjustment value when it is determined by the same determination processing unit that the previous voice and the current voice are the same;
   A speech recognition apparatus comprising:
2. The speech recognition apparatus according to claim 1, wherein the threshold adjustment processing unit changes the adjustment value according to the likelihood calculated by the likelihood calculation processing unit.
3. The voice recognition device according to claim 1 or 2, wherein the threshold adjustment processing unit changes the adjustment value according to the number of times the likelihood calculated by the likelihood calculation processing unit is equal to or greater than the threshold.
4. The validity determination processing unit determines that the speech recognition result by the speech recognition processing unit is invalid when the likelihood calculated by the likelihood calculation processing unit is smaller than a predetermined threshold, and the determination reason is The speech recognition apparatus according to any one of claims 1 to 3, which outputs the speech recognition apparatus.
5. A voice recognition device (10) comprising a voice input unit (12) for inputting a user's voice and a voice storage unit (16) for storing voice input to the voice input unit,
   A speech recognition process for recognizing speech input to the speech input unit;
   Likelihood calculation processing for calculating the likelihood of the speech recognition result by the speech recognition processing;
   When the likelihood calculated by the likelihood calculation process is equal to or greater than a predetermined threshold value, the validity determination process for determining that the voice recognition result by the voice recognition process is valid;
   The same determination process for determining whether the previous voice and the current voice are the same based on the degree of coincidence between the previous voice and the current voice stored in the voice storage unit;
   A threshold adjustment process for lowering the threshold by a predetermined adjustment value when it is determined by the same determination process that the previous voice and the current voice are the same;
   Voice recognition program that executes

Images