JP7542826B2 - Voice recognition program and voice recognition device - Google Patents

Description

The present invention relates to a voice recognition program and a voice recognition device.

Patent Document 1 discloses a system 200 that recognizes input speech and uses the recognition results to provide route guidance, drive a vehicle, and the like. The system 200 is provided with a speech input device 21 that inputs speech uttered by a user, a speech interval detection unit 11 that detects speech intervals from the beginning to the end that constitute a continuous speech input from the speech input device 21, and a speech recognition unit 12 that recognizes speech within the speech interval detected by the speech interval detection unit 11.

When a first voice (first utterance) is input to the voice input device 21 first and then a second voice (second utterance) is input to the voice input device 21, first the voice section corresponding to the first voice is detected by the voice section detection unit 11, and voice recognition of the first voice by the voice recognition unit 12 is started from the start of the first voice section. Then, after voice recognition of the first voice section up to the end of the first voice section is completed, detection of a second voice section corresponding to the second voice and voice recognition from the start of the second voice section are started. This makes it possible to distinguish between the first voice and the second voice and perform voice recognition.

International Publication No. 2019/058453 (e.g., paragraphs 0013-0039, Figures 1 and 5)

When the first voice and the second voice are spoken consecutively, the interval between the first voice and the second voice is short. In such a case, it is necessary to perform voice recognition of the first voice section, detection of the second voice section, and start of voice recognition of the second voice section in a short time. Therefore, if voice recognition of the first voice section takes time, there is a risk that the start of detection of the second voice section of the second voice input from the voice input device 21 thereafter will be delayed, and the start of the detected second voice section will be detected later than the actual start of the second voice. This causes a problem in that voice recognition of the content spoken at the start of the second voice will be missed, and there is a risk that the second voice will not be accurately recognized.

The present invention has been made to solve the above-mentioned problems, and aims to provide a speech recognition program and a speech recognition device that can accurately recognize each of the first and second utterances even when they are input consecutively.

In order to achieve this object, the voice recognition program of the present invention is a program for causing a computer having a storage unit to execute a voice recognition process, and includes a voice storage step for causing the storage unit to function as a voice storage means for storing voice, and storing input voice in the voice storage means, a start time acquisition step for acquiring a start time of an utterance by voice stored in the voice storage means, an end time acquisition step for acquiring an end time of an utterance by voice stored in the voice storage means, an interval acquisition step for acquiring an utterance interval that is a time difference between the end time of a first utterance acquired in the end time acquisition step and the start time acquired in the start time acquisition step, which is a start time of a second utterance input after the first utterance, a retroactive time acquisition step for acquiring a retroactive time that is a time going back from the start time of the second utterance acquired in the start time acquisition step based on the utterance interval acquired in the interval acquisition step, and a voice recognition step for starting voice recognition of the second utterance from a time going back from the start time of the second utterance acquired in the start time acquisition step by the retroactive time acquired in the retroactive time acquisition step, in the voice stored in the voice storage means.

The voice recognition device of the present invention also includes voice input means for inputting voice, voice storage means for storing the voice input by the voice input means, start time acquisition means for acquiring the start time of an utterance by the voice stored by the voice storage means, end time acquisition means for acquiring the end time of an utterance by the voice stored in the voice storage means, interval acquisition means for acquiring an utterance interval which is the time difference between the end time of a first utterance acquired by the end time acquisition means and the start time acquired by the start time acquisition means which is the start time of a second utterance input after the first utterance, retroactive time acquisition means for acquiring a retroactive time which is the time going back from the start time of the second utterance acquired by the start time acquisition means based on the utterance interval acquired by the interval acquisition means, and voice recognition means for starting voice recognition of the second utterance from a time going back from the start time of the second utterance acquired by the start time acquisition means by the retroactive time acquisition means, in the voice stored in the voice storage means.

According to the speech recognition program of claim 1, the input speech is stored in a speech storage means, the end time of the first utterance and the start time of the second utterance stored in the speech storage means are acquired, and a retroactive time based on the speech interval, which is the time difference between them, is acquired. Then, speech recognition of the second utterance is started from the time that is the retroactive time from the start time of the second utterance in the speech of the speech storage means. This makes it possible to reliably start speech recognition from the start of the second utterance stored in the speech storage means, and has the effect of accurately recognizing each of the first utterance and the second utterance even if they are input consecutively. In addition, since the retroactive time is set according to the speech interval between the first utterance and the second utterance, speech recognition can be started from the start of the second utterance, and the effect of minimizing the influence of the first utterance on the speech recognition of the second utterance is also achieved.

The speech recognition program of claim 2 has the following effect in addition to the effect of the speech recognition program of claim 1. When the speech interval is less than or equal to the first predetermined time, that is, when the speech interval between the first utterance and the second utterance is short and they are consecutive, the retroactive time is set to the first retroactive time that is equal to or greater than the first predetermined time in such a case, so that speech recognition can be reliably started from the start of the second utterance that follows the first utterance.

According to the voice recognition program of claim 3, in addition to the effects of the voice recognition program of claim 2 , the following effects are achieved. When the speech interval is equal to or longer than the second predetermined time longer than the first predetermined time , i.e., when the speech interval between the first utterance and the second utterance is long, a second retroactive time equal to or shorter than the second predetermined time is acquired as the retroactive time, so that it is possible to prevent the time lag from the start of speech recognition of the second utterance to the actual start of the second utterance from increasing. This has the effect of preventing erroneous speech recognition of ambient sounds until the start of the second utterance, and reducing the processing time of the computer for speech recognition of the second utterance, thereby reducing the processing load of the computer.

The speech recognition program according to claim 4 has the following effect in addition to the effect of the speech recognition program according to claim 3. Since the first retroactive time is set to a time equal to or greater than the first predetermined time and equal to or less than the second predetermined time, it is possible to prevent the time at which speech recognition of the second utterance is started from going back to the start time of the first utterance. This has the effect of preventing the entire first utterance from being speech-recognized along with the second utterance.

The speech recognition program according to claim 5 has the following effect in addition to the effect of the speech recognition program according to claim 3 or 4. The second retroactive time is set to a time equal to or greater than the first predetermined time and equal to or less than the second predetermined time. This makes it possible to prevent the time at which speech recognition of the second utterance starts from going back to the start time of the first utterance. This has the effect of preventing the entire first utterance from being speech-recognized along with the second utterance.

According to the voice recognition program of claim 6, in addition to the effects of the voice recognition program of any one of claims 1 to 5, the following effect is achieved: When the speech interval is between the first and second predetermined times, the speech interval is set as the look-back time. This makes it possible to easily obtain the look-back time and reliably start voice recognition from the start of the second utterance, since the start of the speech recognition of the second utterance coincides with the end time of the first utterance.

The voice recognition device described in claim 7 achieves the same effect as the voice recognition program described in claim 1.

FIG. 2 is an external view of a mobile terminal. 1 is a diagram showing the volume of a voice and the start and end times of a user's speech. (a) is a diagram showing the timing for starting voice recognition when the speech interval is less than or equal to a first predetermined time, (b) is a diagram showing the timing for starting voice recognition when the speech interval is more than or equal to a second predetermined time, and (c) is a diagram showing the timing for starting voice recognition when the speech interval is between the first and second predetermined times. FIG. 2 is a block diagram showing the electrical configuration of the mobile terminal. 1A is a flowchart of a voice process, and FIG. 1B is a flowchart of a recording process. 13 is a flowchart of a voice recognition process.

A preferred embodiment of the present invention will now be described with reference to the accompanying drawings. First, the configuration of a mobile terminal 1 in this embodiment will be described with reference to FIG. 1. FIG. 1 is an external view of the mobile terminal 1. The mobile terminal 1 is an information processing device (computer) that performs voice recognition of speech uttered by a user H.

The mobile terminal 1 is configured to be able to input voice V, and judges whether the utterance is made by user H based on the volume of the input voice V, and executes voice recognition for each utterance. Note that a known method is used for voice recognition, such as converting voice V into a character string and replacing the converted character string with a corresponding word or phrase. First, referring to FIG. 2, a method for judging the start and end of user H's utterance from voice V input to the mobile terminal 1 will be described.

Figure 2 is a diagram that shows a schematic representation of the volume of voice V and the start time StT and end time EdT of user H's speech. In Figure 2, the horizontal axis represents time and the vertical axis represents the volume (dB) of voice V, with the maximum volume being "0 dB" and the minimum volume being "-120 dB." The range of volume is not limited to 0 dB to -120 dB, and may be any other range.

In the mobile terminal 1 of this embodiment, it is determined whether the user H is speaking based on the volume of the input voice V. Specifically, a start judgment value St_A, which is a volume threshold for determining whether speech has started, and an end judgment value Ed_A, which is a volume threshold for determining whether speech has ended, are set. A volume greater than the end judgment value Ed_A is set for the start judgment value St_A, and examples of the start judgment value St_A and the end judgment value Ed_A are "-25 dB" and "-30 dB", respectively.

When the volume of the input voice V goes from being lower than the start judgment value St_A to being equal to or higher than the start judgment value St_A, it is determined that user H has started speaking, and the time at which this occurs is designated as the start time StT. On the other hand, if the volume falls below the end judgment value Ed_A after the start time StT, it is determined that user H has ended speaking, and that time is designated as the end time EdT. In other words, it is determined that user H has been speaking between the start time StT and the end time EdT.

By setting the start judgment value St_A at a volume greater than the end judgment value Ed_A, it is possible to clearly distinguish between the surrounding environmental sounds and the start of speech, and to prevent the surrounding environmental sounds from being mistaken for user H's pronunciation. On the other hand, by setting the end judgment value Ed_A at a volume less than the start judgment value St_A, in a situation in which it is determined that user H is speaking, even if the volume of the speech temporarily drops below the start judgment value St_A, it can be determined that the speech is continuing. As a result, the start and end of user H's speech can be appropriately obtained.

Based on the start time StT and end time EdT of user H's utterance thus acquired, voice recognition of that utterance is performed. In this embodiment, when user H speaks continuously, the timing to start voice recognition of the subsequent utterance is set according to the speech interval ΔT, which is the time difference between the previous utterance and the subsequent utterance. The timing to start voice recognition will be described with reference to Figure 3.

Figure 3(a) is a diagram showing the timing for starting voice recognition when the speech interval ΔT is equal to or less than the first predetermined time x1, Figure 3(b) is a diagram showing the timing for starting voice recognition when the speech interval ΔT is equal to or more than the second predetermined time x2, and Figure 3(c) is a diagram showing the timing for starting voice recognition when the speech interval ΔT is between the first predetermined time x1 and the second predetermined time x2.

In Figures 3(a) to (c), the first utterance V1 is "Good morning" uttered by user H, and the second utterance V2 is "It's a sunny day today" uttered by user H immediately after the first utterance. The time difference between the end time EdT of the first utterance V1 and the start time StT of the second utterance V2 is the speech interval ΔT between the first utterance V1 and the second utterance V2, and the retroactive time T is calculated depending on the length of the speech interval ΔT.

Here, the voice V input to the mobile terminal 1 is stored in the voice buffer 11b described later in FIG. 4, regardless of whether the user H has spoken. Voice recognition of the second utterance V2 starts at the recognition start time StR, which is the time going back the retroactive time T from the start time StT of the second utterance V2 in the voice V in the voice buffer 11b.

First, referring to FIG. 3(a), the retroactive time T when the second utterance V2 starts immediately after the first utterance V1 will be described. FIG. 3(a) shows the case where the second utterance V2 starts immediately after the first utterance V1, i.e., the above-mentioned utterance interval ΔT is equal to or shorter than the first predetermined time x1. An example of the first predetermined time x1 is "0.1 seconds."

In this way, when the speech interval ΔT is short and equal to or less than the first predetermined time x1, and the first utterance V1 and the second utterance V2 are consecutive, the first retroactive time Tx1, which is equal to or greater than the first predetermined time x1, is set as the retroactive time T. An example of the first retroactive time Tx1 is "0.5 seconds." This allows the recognition start time StR of the second utterance to be set to a timing earlier than the start time StT of the second utterance, so that speech recognition can be reliably started from the start of the second utterance.

In this case, an utterance near the end time EdT of the first utterance (for example, the "su" in "Good morning") may be included in the recognition start time StR of the second utterance. In such a case, the recognition results before the start time StT of the second utterance may be excluded or removed from the results of the speech recognition that began at the recognition start time StR of the second utterance, thereby outputting only the results of the speech recognition from the start time StT of the second utterance.

Next, referring to FIG. 3(b), a case where the speech interval ΔT between the first utterance V1 and the second utterance V2 is long will be described. FIG. 3(b) shows a case where the speech interval ΔT is equal to or longer than the second predetermined time x2. An example of the second predetermined time x2 is "3 seconds". In this way, when the speech interval ΔT between the first utterance V1 and the second utterance V2 is equal to or longer than the second predetermined time x2, a second retroactive time Tx2 equal to or shorter than the second predetermined time x2 is set as the retroactive time T. An example of the first retroactive time Tx1 is "2 seconds".

This makes it possible to prevent the time lag from increasing between when speech recognition of the second utterance V2 starts and when the second utterance V2 actually starts. This makes it possible to prevent erroneous speech recognition of ambient sounds until the second utterance V2 starts, and also reduces the processing time of the mobile terminal 1 (specifically, the CPU 10 described later in FIG. 4) for speech recognition of the second utterance V2, thereby reducing the processing load of the mobile terminal 1.

Next, referring to FIG. 3(c), a case will be described in which the speech interval ΔT between the first utterance V1 and the second utterance V2 is between the first predetermined time x1 and the second predetermined time x2. In such a case, the speech interval ΔT is set as the retroactive time T. As a result, the recognition start time StR of the second utterance V2 becomes the end time EdT of the first utterance V1, so that the retroactive time T can be easily obtained and speech recognition can be reliably started from the start of the second utterance V2.

Here, the first retroactive time Tx1 and the second retroactive time Tx2 are set to a time equal to or greater than the first predetermined time x1 and equal to or less than the second predetermined time x2. This prevents the recognition start time StR of the second utterance from going back to the start time StT of the first utterance V1, thereby preventing the entire first utterance V1 from being voice recognized together with the second utterance V2.

As described above, by starting speech recognition from the recognition start time StR, which is calculated by going back from the start time StT of the second utterance V2 by a retroactive time T corresponding to the speech interval ΔT between the immediately preceding first utterance V1 and second utterance V2, speech recognition of the second utterance can be reliably started from the start of the second utterance V2. This ensures that speech recognition can be started reliably from the start of the second utterance even when the first utterance and the second utterance are input consecutively, thereby enabling accurate speech recognition of the first utterance and the second utterance.

In addition, the volume of the voice V when user H starts the second utterance may be low, and user H may actually be speaking at the time determined to be the start time StT of second utterance V2. Even in such a case, by starting voice recognition from a time that is backdated by the retroactive time T from the start time StT of second utterance V2, it is possible to reduce the loss of voice recognition of the voice V that user H was actually speaking before the start time StT of second utterance V2 was determined.

The first predetermined time x1 is not limited to 0.1 seconds, and may be 0.1 seconds or more or 0.1 seconds or less as long as it is equal to or less than the second predetermined time x2. The second predetermined time x2 is not limited to 3 seconds, and may be 3 seconds or more or 3 seconds or less as long as it is equal to or more than the first predetermined time x1. The first retroactive time Tx1 is not limited to 0.5 seconds, and may be 0.5 seconds or more or 0.5 seconds or less as long as it is equal to or more than the first predetermined time x1 and equal to or less than the second predetermined time x2. Similarly, the second retroactive time Tx2 is not limited to 2 seconds, and may be 2 seconds or more or 2 seconds or less as long as it is equal to or more than the first predetermined time x1 and equal to or less than the second predetermined time x2. Furthermore, the first retroactive time Tx1 is shorter than the second retroactive time Tx2, but is not limited to this. The first retroactive time Tx1 and the second retroactive time Tx2 may be the same time, or the first retroactive time Tx1 may be longer than the second retroactive time Tx2.

Next, the electrical configuration of the mobile terminal 1 will be described with reference to FIG. 4. FIG. 4 is a block diagram showing the electrical configuration of the mobile terminal 1. As shown in FIG. 4, the mobile terminal 1 has a CPU 10, a flash ROM 11, and a RAM 12, which are each connected to an input/output port 14 via a bus line 13. The input/output port 14 is further connected to a microphone 15 for inputting voice V, an LCD 16 for displaying the results of voice recognition, and a touch panel 17 for inputting instructions from the user H.

The CPU 10 is a calculation device that controls each part connected by the bus line 13. The flash ROM 11 is a rewritable non-volatile memory that stores a voice recognition program 11a and a voice buffer 11b in which voice V is stored. When the voice recognition program 11a is executed by the CPU 10, the voice processing of FIG. 5 is executed. The RAM 12 is a memory for rewritably storing various work data, flags, etc. when the CPU 10 executes the voice recognition program 11a, and is provided with a retroactive time memory 12a in which the retroactive time T described above is stored.

Next, the processing executed by the CPU 10 of the mobile terminal 1 will be described with reference to Figures 5 and 6. Figure 5(a) is a flowchart of the voice processing. The voice processing is a process that is executed when an instruction to execute the voice recognition program 11a is input from the user H via the touch panel 17 or the like.

The audio process first clears the contents of the audio buffer 11b (S1), and sets the audio acquisition time and the above-mentioned start time StT and end time EdT to 0 (S2). The audio acquisition time is the time when the sampling period of the audio V (e.g., 1/44100 seconds) is one unit time, and is used as time information for acquiring the audio V stored in the audio buffer 11b at the above-mentioned sampling period intervals in order from 0 seconds, i.e., the time when storage of the audio V in the audio buffer 11b began.

After the process of S2, the current volume and previous volume are set to -120 dB, which is the minimum volume value (S3). The current volume stores the volume for determining the start time StT and end time EdT of the speech, and the previous volume stores the previous volume of the current volume.

After the process of S3, the recording process starts (S4). The recording process is executed every sampling period of the sound V, and the sound V input from the microphone 15 is stored in the sound buffer 11b every sampling period. The process of S4 starts the periodic execution of the recording process. Here, the recording process will be described with reference to FIG. 5(b).

Figure 5 (b) is a flowchart of the recording process. As described above, the recording process is an interrupt process that is executed for each sampling period of the audio V. The recording process adds the audio V acquired from the microphone 15 to the audio buffer 11b (S20) and ends. As a result, the audio V acquired for each sampling period is stored in the audio buffer 11b.

Returning to FIG. 5(a), after the process of S4, the volume of the voice V at the time of voice acquisition is obtained from the voice buffer 11b and set as the current volume (S5). After the process of S5, the voice recognition process (S6) is executed. Here, the voice recognition process will be explained with reference to FIG. 6.

Figure 6 is a flowchart of the voice recognition process. The voice recognition process first checks whether the previous volume was smaller than the start judgment value St_A described above in Figure 2, and whether the current volume at the time of voice acquisition is equal to or greater than the start judgment value St_A (S30). That is, it checks whether the voice V in the voice buffer 11b has changed from a state smaller than the start judgment value St_A to equal to or greater than the start judgment value St_A, at the start time StT when speech began.

In the process of S30, if the previous volume is smaller than the start judgment value St_A and the current volume at the voice acquisition time is equal to or greater than the start judgment value St_A (S30: Yes), the voice acquisition time is set to the start time StT (S31). After the process of S31, the speech interval ΔT is calculated by subtracting the end time EdT described above in FIG. 3, which is set in the processes of S39 and S40 described below, from the start time StT (S32). After the process of S32, the calculated speech interval ΔT is confirmed (S33).

In the process of S33, if the speech interval ΔT is less than or equal to the first predetermined time x1 (ΔT≦x1), a first retroactive time Tx1 is set in the retroactive time memory 12a (S34); if the speech interval ΔT is greater than or equal to the second predetermined time x2 (ΔT≧x2), a second retroactive time Tx2 is set in the retroactive time memory 12a (S35); if the speech interval ΔT is between the first predetermined time x1 and the second predetermined time x2 (x1<ΔT<x2), the speech interval ΔT is set in the retroactive time memory 12a (S36).

After the processes of S34 to S36, speech recognition is performed from a time that is the retroactive time T in the retroactive time memory 12a back from the start time StT set in the process of S31 in the speech buffer 11b (i.e., the recognition start time StR) (S37). As a result, the retroactive time T corresponding to the speech interval ΔT described above in Figures 3(a) to (c) is set in the retroactive time memory 12a, and speech recognition is started from the retroactive time T from the start time StT. The results of the speech recognition performed in the process of S37 may be displayed on the LCD 16, or may be transmitted to an information processing device such as another mobile terminal 1 via a communication device (not shown).

In the process of S30, if the previous volume is equal to or greater than the start judgment value St_A, or if the current volume is smaller than the start judgment value St_A (S30: No), the processes of S31 to S37 are skipped. After the processes of S30 and S37, it is confirmed whether the previous volume is greater than the end judgment value Ed_A and the current volume is equal to or less than the end judgment value Ed_A (S38).

In the process of S38, if the previous volume is greater than the end judgment value Ed_A and the current volume is equal to or less than the end judgment value Ed_A (S38: Yes), the voice acquisition time is the timing of the end time EdT when the utterance described above in FIG. 2 ends, so the voice acquisition time is set to the end time EdT (S39). On the other hand, if the previous volume is equal to or less than the end judgment value Ed_A or the current volume is greater than the end judgment value Ed_A (S38: No), the process of S39 is skipped. After the processes of S38 and S39, the voice recognition process ends.

Returning to FIG. 5(a), after the voice recognition process of S6, the sampling period is added to the voice acquisition time, and the voice acquisition time is advanced to the next timing for acquiring the volume from the voice buffer 11b (S7). After the process of S7, it is confirmed whether an instruction to end the voice processing has been acquired from the user H via the touch panel 17 (S8). If an instruction to end the voice processing has not been acquired in the process of S8 (S8: No), the processes from S5 onwards are repeated, and if an instruction to end the voice processing has been acquired (S8: Yes), the voice processing is terminated.

The present invention has been described above based on the embodiments, but the present invention is in no way limited to the above-mentioned embodiments, and it can be easily imagined that various improvements and modifications are possible within the scope of the invention without departing from its spirit.

In the above embodiment, the retroactive time T is set to the first retroactive time Tx1 when the speech interval ΔT is equal to or less than the first predetermined time x1, the second retroactive time Tx2 when the speech interval ΔT is equal to or greater than the second predetermined time x2, and the speech interval ΔT when the speech interval ΔT is between the first predetermined time x1 and the second predetermined time x2, but this is not limited to the above. Regardless of the speech interval ΔT, the retroactive time T may be set to the speech interval ΔT, or a time obtained by multiplying the speech interval ΔT by a predetermined coefficient (e.g., 0.8), or a time obtained by adding a predetermined time (e.g., 0.5 seconds) to the speech interval ΔT. Furthermore, regardless of the speech interval ΔT, the retroactive time T may be set to the first retroactive time Tx1 or the second retroactive time Tx2.

In the above embodiment, the start time StT and end time EdT of the speech are determined based on the volume of the voice V, but this is not limited to the above. For example, the start time StT of the speech may be determined as the time when a frequency band of human voice (e.g., 100 Hz to 1000 Hz) begins to be observed in the voice V, and the end time EdT may be determined as the time when the frequency band is no longer observed in the voice V.

In the above embodiment, the voice V is input from the microphone 15, but this is not limited to the above. For example, the voice V may be voice data stored in advance in the flash ROM 11, or voice data transmitted from another mobile terminal 1 or the like via a communication device (not shown) may be used as the voice V.

In the above embodiment, the audio acquisition time is set to one unit time of the sampling period, and the volume is acquired from the audio buffer 11b at sampling period intervals, but this is not limited to this. For example, one unit time of the audio acquisition time may be one second, and the volume may be acquired from the audio buffer 11b at one second intervals.

In the above embodiment, a mobile terminal 1 incorporating the voice recognition program 11a is exemplified, but the present invention is not limited to this, and the voice recognition program 11a may be executed by another information processing device (computer) such as a personal computer or a tablet terminal. In addition, the voice recognition program 11a may be stored in a ROM or an IC chip, and the present invention may be applied to a dedicated device that executes only the voice recognition program 11a.

1. Mobile terminal (computer)
11 Flash ROM (storage unit)
11b Audio buffer (audio storage means)
11a Voice recognition program V Voice V1 First utterance V2 Second utterance S20 Voice storage step StT Start time EdT End time ΔT Speech interval x1 First predetermined time x2 Second predetermined time T Retrospective time Tx1 First retrospective time Tx2 Second retrospective time S31 Start time acquisition step, start time acquisition means S39 End time acquisition step, end time acquisition means S32 Interval acquisition step, interval acquisition means S34 ~S36 Retrospective time acquisition step, retrospective time acquisition means S37 Voice recognition step, voice recognition means

Claims (7)

A speech recognition program for causing a computer having a storage unit to execute a speech recognition process,
The storage unit functions as a voice storage means for storing voice,
a voice storage step of storing the input voice in the voice storage means;
a start time acquisition step of acquiring a start time of an utterance by voice stored in the voice storage means;
an end time acquisition step of acquiring an end time of a voice utterance stored in the voice storage means;
an interval acquiring step of acquiring an utterance interval which is a time difference between an end time of the first utterance acquired in the end time acquiring step and a start time of a second utterance which is acquired in the start time acquiring step and is input after the first utterance;
a retroactive time acquisition step of acquiring a retroactive time that is a time going back from the start time of the second utterance acquired in the start time acquisition step, based on the utterance interval acquired in the interval acquisition step;
a voice recognition step of starting voice recognition of the second utterance from a time going back from the start time of the second utterance acquired in the start time acquisition step by the retroactive time acquired in the retroactive time acquisition step, for the voice stored in the voice storage means. The speech recognition program according to claim 1, characterized in that, in the retroactive time acquisition step, if the speech interval acquired in the interval acquisition step is equal to or shorter than a first predetermined time, a first retroactive time that is equal to or longer than the first predetermined time is acquired as the retroactive time. 3. The speech recognition program according to claim 2, wherein the retroactive time acquisition step acquires, when the speech interval acquired in the interval acquisition step is equal to or greater than a second predetermined time that is longer than the first predetermined time, a second retroactive time that is equal to or less than the second predetermined time as the retroactive time. The speech recognition program according to claim 3, characterized in that the first retroactive time is equal to or greater than the first predetermined time and equal to or less than the second predetermined time. The speech recognition program according to claim 3 or 4, characterized in that the second retroactive time is equal to or greater than the first predetermined time and equal to or less than the second predetermined time. 6. The speech recognition program according to claim 1, wherein the retroactive time acquisition step acquires the speech interval as the retroactive time when the speech interval acquired in the interval acquisition step is between a first predetermined time and a second predetermined time. A voice input means for inputting voice;
a voice storage means for storing the voice inputted by the voice input means;
a start time acquisition means for acquiring a start time of an utterance by the voice stored in the voice storage means;
an end time acquisition means for acquiring an end time of an utterance by voice stored in the voice storage means;
an interval acquiring means for acquiring an utterance interval, which is a time difference between an end time of a first utterance acquired by the end time acquiring means and a start time of a second utterance acquired by the start time acquiring means and input after the first utterance;
a retroactive time acquisition means for acquiring a retroactive time that is a time going back from the start time of the second utterance acquired by the start time acquisition means, based on the utterance interval acquired by the interval acquisition means;
a voice recognition means for starting voice recognition of the second utterance from a time going back from the start time of the second utterance acquired by the start time acquisition means by the retroactive time acquired by the retroactive time acquisition means, in the voice stored in the voice storage means.

Images