JPH09257970A - Time adjusting method and device through voice recognition - Google Patents

Abstract

PROBLEM TO BE SOLVED: To simply and accurately perform time adjustment operation in an apparatus with a buil-in timepiece by taking in the voice of time information service given by NTT, etc., through a telephone line, recognizing the voice by use of a voice recognition technique, and automatically adjusting the time with the result. SOLUTION: By use of continuous voice recognition means 5 performing voice recognition by a keyword spotting processing with a dinamic current neutral network, voice character data to the input voice obtained by analyzation of voice by an input signal analyzing unit 3 is compared with reference voice character data to respective words registered in a reference voice character data storing unit 51 by a word detection unit 52, and it is detected, in the form of numerical value showing the provability of existence of the registered word, that on which position on the time axis of input voice the registered word exists and that how extent of provability the registered word exists, for every registered words. Based on the numerical value showing the provability, the input voice is recognized, and according to the result the time of the timepiece is set by a specified timming.

Description

Detailed Description of the Invention

[0001]

TECHNICAL FIELD The present invention relates to home electric appliances, AVs.
(Audio and Visual) The present invention relates to a time setting method and apparatus using voice recognition, which uses a voice recognition technology to set the current time of a clock built in a device, a facsimile, a telephone with an answering machine, or the like.

[0002]

2. Description of the Related Art Recently, many home electric appliances and AV devices have a built-in clock.

Generally, the present time adjustment operation of this type of timepiece is performed not only when the device is purchased and started to be used, but also when the device is installed and the power is restored after a power failure or after the power plug is unplugged. Will be needed.

In order to set the correct time in this time adjustment, for example, in the case where a bidet is purchased and installed in a deck, for example, after connecting the VCR to the TV, the menu button on the attached remote controller is pushed. , "Current time setting" is selected from several menus displayed on the TV screen. For example, the current time is 8 am
If it was a little before the hour, set the display to 8:00 am with the number buttons on the remote control, and while listening to the NTT time service (No. 117) by phone, "I will inform you exactly at 8 am. , Pip, pip, pawn "
At the moment it becomes a pawn, the time is set by pressing the time setting button. This operation procedure may vary depending on the type of equipment, manufacturer, etc., but it seems to be more than the general operation procedure. By performing the time adjustment by such an operation procedure, it becomes possible to adjust the current time to a substantially accurate time in minutes.

As described above, recently, there are many devices that can be operated by a remote controller, and in such a device, it is common to use the remote controller to adjust the current time. In addition, in this type of equipment, it is extremely important to keep the time of the built-in clock always accurate in order to perform various functions of the equipment with high accuracy according to the procedure. There are also many.

[0006]

However, such a series of operations is a troublesome work even for a person who is accustomed to handling this type of equipment, and a person who is unfamiliar with handling the equipment is required. Even more so for me.
In particular, the elderly, who are said to be poor at handling equipment, are currently adjusting the time by repeating trial and error while viewing difficult specifications. Further, according to the method described above, it is difficult to set the time in seconds, and the time adjustment in minutes is limited.

On the other hand, recently, a video deck having a function of automatically adjusting the time from a time signal of a television has also been developed. This usually means that when it's not used for recording too often (eg, before 6am), the VCR automatically powers up and monitors certain channels to, for example, “pip” on a time signal at 6am. , Pip, pip, pawn, "pawn" has the function of adjusting the time.

However, in order to make use of such a function, at least the user has to set the current time in advance in advance, and the current time set by the user is set to a correct current time. Must be set within the range of. This is about 5 minutes at the maximum until the VCR power is turned on automatically and a specific channel is monitored, and the time is adjusted at a time signal of 6 am, pit, pit, pit, pit, and pawn, for example. This is because it is necessary to keep it within. Furthermore, if there is a power outage or the power plug is pulled out for any reason,
This means that the time adjustment function is not fulfilled.

In order to solve such a problem, the present invention takes in time information by voice and dial tone such as NTT's time service by telephone, performs voice recognition using a voice recognition technology, and outputs the recognition result. To realize a time adjustment method using voice recognition and a device therefor capable of easily and accurately adjusting the time in a device with a built-in clock by automatically adjusting the time. It is an object.

[0010]

A time adjustment method using voice recognition according to the present invention inputs a time guide including a voice when adjusting a current time of a clock built in a device, and inputs a voice guide of the time guide. In a time adjustment method using voice recognition, a part of the voice is recognized, and the time of the clock is set based on the recognition result at a predetermined timing at the time point of the time guidance. The means for performing is a voice recognition means by keyword spotting processing using the dynamic recurrent neural network method, voice feature data for the input voice obtained by analyzing the voice included in the time guidance input by the sound input means And the standard voice feature data for each registered word are compared, and to what part on the time axis of the input voice the registered word is and Whether present in probability of, for each registered word, detects a numerical value indicating the likelihood,
Based on the numerical value indicating the certainty, the input voice is recognized, and based on the recognition result, the time of the clock is set to the time of the time indicated by the time guidance at a predetermined timing. Characterize.

In such a time adjustment method, the time at which the built-in clock is set based on the recognition result is set to the time guidance voice for each time guidance voice corresponding to the passage of time. The voice recognition is performed on the other hand, and the time is the time when the certainty of the voice recognition result at each time is determined to be sufficient.

As for the time when the built-in clock is set based on the recognition result, the voice recognition is performed a preset number of times for each voice of the time guidance corresponding to the passage of time, and the time recognition is performed for each time. The time is determined based on the voice recognition result for the time guidance voice.

By adopting such a method, an accurate current time can be automatically set only by inputting time guidance including voice. As a result, even a user who is unfamiliar with handling this type of device can easily adjust the time. Especially, as a time guide including voice, NT
If you use the time service by "117" of T, you can call the "117" at any time and input the time guidance emitted from the handset, and the accurate current time in seconds will be automatically displayed. You can set the time. In addition, since the continuous speech recognition means by keyword spotting processing using the dynamic recurrent neural network method is used as the speech recognition technology, it is less susceptible to noise, and even if there is noise on the line, noise is output to word recognition output. Since there is no significant influence, it is possible to perform highly accurate voice recognition.

Further, the time adjusting device using the voice recognition of the present invention inputs the time guide including the voice when the current time of the clock built in the device is adjusted, and recognizes the voice part of the time guide. Then, the time of the timepiece is set based on the recognition result at a predetermined time at the time point of the time guidance provided by the time guidance. The sound input means for inputting time guidance, the voice feature data for the input voice obtained by analyzing the voice predicting the current time input to the sound input means, and the standard voice feature data for each registered word are provided. In comparison, the dynamics of detecting, with respect to each part of the registered word, the degree of certainty in which part on the time axis of the input voice the registered word is detected by a numerical value indicating the certainty. Voice recognition means by keyword spotting processing using the recurrent neural network method, and based on the recognition result by this voice recognition means, the time of the clock is set at a predetermined timing to the time of the time point that is guided by the time guidance. And a time setting means for setting.

In the time adjusting device using such voice recognition, the time when the built-in clock is set based on the recognition result is voice recognized for each voice of the time guidance corresponding to the passage of time. And the time at which the certainty of the speech recognition result at each time is determined to be sufficient.

As for the time when the built-in clock is set based on the recognition result, a predetermined number of times of voice recognition is performed for each time guidance voice corresponding to the passage of time, and the time recognition is performed for each time. The time is determined based on the voice recognition result for the time guidance voice.

By using such a device, an accurate current time can be automatically set only by inputting time guidance including voice. As a result, even a user who is unfamiliar with handling this type of device can easily adjust the time. Especially, as a time guide including voice, NTT
If you use the “117” time service, you can call the “117” at any time and have the time guide issued from the handset enter, and the current time will be displayed in seconds automatically. Can be set. In addition, since the continuous speech recognition means by keyword spotting processing using the dynamic recurrent neural network method is used as the speech recognition technology, it is less susceptible to noise, and even if there is noise on the line, noise is output to word recognition output. Since there is no significant influence, it is possible to perform highly accurate voice recognition.

[0018]

Embodiments of the present invention will be described below with reference to the drawings. In this embodiment,
Taking a video deck as an example of the device, the current time of the clock built into this deck is set by NTT "117
An example will be described in which the time is adjusted by using the "time" service.

(First Embodiment) FIG. 1 is a block diagram showing a configuration of a time adjustment processing unit for performing time adjustment according to an embodiment of the present invention. A sound input unit 1 for inputting a sound, The "117th" time guide signal input to the sound input unit 1 is passed through an amplifier, a low-pass filter, etc.
After an appropriate voice waveform, an A / D converter converts the input signal processing unit 2 into a digital signal, and the signal processed by the input signal processing unit 2 is subjected to frequency analysis at short time intervals to determine the frequency. An input signal analysis unit 3 that extracts a several-dimensional feature vector representing a feature and outputs a time series (feature vector sequence) of this feature vector, and is included in the time guidance based on the output from this input signal analysis unit 3. A voice that detects a dial tone and also detects the timing of the dial tone detection unit 4 (which will be described later) and a voice that recognizes each word (which will be described later) that constitutes the voice included in the time guidance. A recognition unit 5 (details of this voice recognition unit will be described later), a time display unit 6 for displaying the result of voice recognition as a time, and a built-in time corresponding to a correctly recognized voice. The time, the time at that point to be guided by the time the guide is composed of such as the time setting unit 7 for setting at a predetermined timing.

By the way, in the present embodiment, as described above, the time guide service of "117" of NTT is used. This "117" time guidance service is announced with a dial tone (first dial tone) emitted at 1-second intervals and a time guidance voice for notifying the current time superimposed on the first dial tone. A dial tone (second dial tone) that informs that the time has been reached, and further, a call that is output at an interval of 1 second before the second dial tone every hour, minute, thirty seconds, and exactly every hour, minute, and hour. Sound (third dial tone), and these dial tone and voice are continuously output with the passage of time.

That is, as an example of the time guidance,
"I will notify you at 7:20:50 am ... Pawn ...", "I will notify you just at 7:21 am ... Pippippi Pawn ...", "7:21 am
I will inform you about 10 minutes. "Pawn ...", the time is guided at intervals of 10 seconds. This time guide is the first dial tone that ticks the second every 1 second (This first dial tone notifies you of the above "7:20:50 am.-
・ "Pawn ..." is a portion indicated by "...", which is represented by a dial tone p1. A second dial tone every 10 seconds (the dial tone every 10 seconds is "7:00 am"). "Pawn" of "Pawn ..."
, Which is indicated by the dial tone p2), and the voice of the human announcement (this is the portion of "I will notify you at 7:20:50 am.", Which is indicated by the voice signal v1).
It also includes noise on the line (denoted by n). In addition, in the case of exactly what time, what minute and thirty seconds, and what time, what minute, and so on, the third dial tone "pippippi" always comes before "pawn" (this is represented by dial tone p3). The tone "pawn" p2
Is 880 Hz, and the beeping sound p3 is 44.
0 Hz. The dial tone p1 is emitted at intervals of 1 second, and the dial tone p3 that is emitted just before what hour and 30 minutes and what hour and minute is also emitted every second.

FIG. 2 (a) is a time guidance signal of "I will notify you at 7:21 am .... Pippipippon ...", and FIG. 2 (b) shows "7:21 am. The following is an example of a time guidance signal for "Pawn ...". Thus, FIG.
In the case of (a), a beep tone p1 is emitted at 1 second intervals after a "pawn" (a beep tone p2) indicating the previous time, and a beep tone p3 is issued after the beep tone p1 is emitted six times. "Pippippi" is emitted at 1 second intervals, and 1 second later, a beep sound p2 "Pawn" is emitted. Also, FIG.
In the case of, "Pawn" (diatone p
After 2), the dial tone p1 is emitted 9 times at intervals of 1 second, and 1 second later, the dial tone p2 of "pawn" is issued. It should be noted that the voice signal v1 by the human announcement is sent in a state of being superimposed on the dial tone p1. Further, (c) of the figure partially shows the feature vector obtained by performing the feature analysis of the time guidance signal shown in (b) of the same figure by the input signal analysis section 3, and stores the tone vector feature vector storage shown in FIG. The section 41 stores the respective feature vectors Vp1, VP2, Vp3 of the dial tones p1, p2, p3 shown in FIG. 2 (c). Note that Vv1 indicates the feature vector of the voice part. 2D shows the timing signal of the dial tone, particularly the timing signal of the dial tone p2.

By the way, in the present invention, the speech recognition unit 5 is a dynamic recurrent neural network (Dynamic Recurrent Neural Net) developed by the present applicant.
works: hereinafter referred to as DRNN) continuous speech recognition technology by keyword spotting processing is used. This DRNN voice recognition technology has already been applied for a patent (Japanese Patent Application Laid-Open Nos. 4-161075 and 6-409).
7, JP-A-6-119476, etc.). The outline of voice recognition using this DRNN will be described below.

The voice recognition unit 5 using the DRNN system is mainly composed of a standard voice feature data storage unit 51, a word detection unit 52, and a voice recognition processing unit 53. In this case, the standard voice characteristic data storage unit 51 stores standard voice characteristic data of words necessary for specifying the time. This standard speech feature data allows speech recognition by unspecified speakers to be recognized by the speech recognition technology based on DRNN, so that the speech to be recognized by a large number of speakers (for example, about 200 persons) is used as the recognition target word as described above. It can be voice feature data obtained by using it, but when the information source of the time guidance is limited, it may be voice feature data by a specific speaker.

Further, the word detecting section 52 compares the voice feature data for the input voice with the standard voice feature data for each registered word, and determines to what part of the input voice the registered word is on the time axis and how much. Is detected with a numerical value indicating the certainty for each registered word. The specific processing of the word detecting unit 62 will be described below with reference to FIG. .

If the input voice is "Notify me of 7:21:10 am", it is assumed that the voice signal as shown in FIG. 3 (a) is output. It should be noted that, here, for simplification of the description, the description will be made of the voice of the part “7:00 am”.

Of the input speech of "7:00 am", "am" and "7:00" are keywords, and their characteristic data are registered in advance in the standard speech characteristic data storage unit 51 as recognition candidate words. ing. And the word detection unit 5
2 includes a detection unit that detects each word corresponding to a word (a word necessary for specifying the time as described above) registered in the standard voice feature data storage unit 51 in advance. How certain is the part in the input voice,
It is to detect whether the corresponding word exists. That is, when the word "am" is present as the word 1 in the input voice, and the detection unit waiting for the signal "am" detects it, as shown in FIG. The signal rises in the section. Similarly, when the word "7 o'clock" is present in the input voice as the word 2, when the detection unit waiting for the signal "7 o'clock" detects it, the input is made as shown in FIG. 3 (c). Voice "7 o'clock"
The signal rises in the part.

In the figures (b) and (c), the numerical value such as 0.9 or 0.8 is a numerical value showing the certainty,
Below, it is called the degree of approximation. This approximation is 0.9 or 0.8
If the numerical value is high, it means that the certainty is high, and it is determined that “am” is a recognition candidate for word 1, and “7 o'clock” is similar for word 2. It is determined to be a recognition candidate. In other words, the registered word "am" has a degree of approximation of 0.9 in the time axis w1 part of the input voice.
It can be seen that the registered word "7 o'clock" exists in the part of the time axis w2 in the input voice with an approximation degree of 0.8.

Further, in the example of FIG. 3, the word "7 o'clock" is changed to the word 3 (this word 3 is the word "8 o'clock") as shown in FIG. The signal that waits for) stands up with a certain degree of certainty (the degree of approximation is about 0.6). Thus, on the same time as the input audio signal (in this case, on the time of w2)
In addition, when two or more registered words are present as recognition candidates, the word detection unit 62 outputs some recognition candidates from the first rank in descending order of similarity.

FIG. 3 (e) shows the output of the detection unit waiting for, for example, "10 seconds" as the word 4, and within the time shown in the figure, the voice "10 seconds" or Since there is no voice having a feature vector similar to that, almost no rise occurs.

The voice recognition processing section 63 is mainly composed of an arithmetic unit (CP
U) and a ROM storing a processing program, and understands the meaning of the input voice based on the recognition word output (degree of approximation) from the word detection unit 62 as described above. For example, detection data as shown in FIGS. 3B to 3D from the word detection unit 62 (this is called word lattice, and this word lattice is the registered word name, the degree of approximation, the start point s and the end point e of the word). Is included), first, based on the word lattice, one or more words as keywords in the input voice are determined. In this example, since the input voice is "7:00 am", "am" and "7:00" are detected. In this example, in order to simplify the explanation, the input voice is set to "7:00 am", but in reality, it is a voice such as "I will notify you at 7:21:10". In this case as well, as in the above, “am”, “7:00”, “21 minutes”, and “10 seconds” are the keywords, and “7:21:10 am” is selected from these keywords.
Understand what the continuous input voice says "I will tell you the seconds."

As described above as the voice recognition technique, D
By using the voice recognition technology by RNN, it is less susceptible to sudden noise and the like as compared with the voice recognition technology such as DP matching that has been conventionally used, and is influenced by the noise n as shown in FIG. Without this, highly accurate voice recognition can be performed. For example, in the voice recognition in DP matching, when comparing the feature vector obtained by the voice analysis with the feature vector stored in the standard voice feature data storage unit, each feature element Since feature vectors are associated with each other and compared, if feature vectors due to noise exist in the middle, it becomes difficult to associate feature vectors for each element. For this reason, in voice recognition by DP matching, it is necessary to separate noise and dial tone on the line from the voice when voice recognition is performed, and to minimize the influence of noise and dial tone.

On the other hand, the DRNN speech recognition technique used in the present invention has a great influence on the overall degree of approximation to the speech even if there is noise in the signal "AM". There is no. Therefore,
It is unlikely to be affected by noise that suddenly gets on the line or a dial tone other than voice, and there is no need to separate the dial tone from the voice. It should be noted that the fact that the voice recognition technology by the DRNN is not easily affected by sudden noise means that the present applicant has already applied for a patent.
In more detail.

FIG. 4 is a view showing the outer surface of the remote controller attached to the VCR.
Most of the functions of the VCR can be executed. In addition to the circuit section of the normal remote controller, the remote controller has a built-in time adjustment processing section based on voice recognition for adjusting the current time shown in FIG.

On the outer surface of the remote controller, the power button 31, the menu button 3 for performing various settings such as the current time setting and the program reservation setting, as in a normal remote controller.
2, a video / television switch button 33, number buttons 34 of "0" to "9", a button 35 for recording, reproducing, rewinding, fast-forwarding, etc. are provided. In addition to these buttons, various buttons for operating the VCR are provided, but these are not shown. In addition to these buttons, a sound input unit (hereinafter referred to as a microphone) 1 and a time display unit 7 in the time adjustment processing unit.
Further, a time data transfer button 36 for transferring the time data set on the remote controller to the VCR main body (the time displayed on the time display unit 7 will be described later) is provided.

The operation for setting the current time in such a configuration will be described using a specific example.

First, it is assumed that the current time is slightly before 7:20:50 am. At this point, the user is “117”
Call. Then, the microphone 1 of the remote controller shown in FIG. 4 is brought close to the handset of the telephone (this handset may be a cordless phone, its parent device, or a mobile phone) to hear the current time guidance emitted from the handset.

It is now assumed that the input start time is a short time before the "pawn" for notifying just 7:21 am is issued. The time information after the start of this input is "pawn" that informs you at 7:21 am, and then "7 am
We will inform you of the time of 21 minutes and 10 seconds. ···pawn··
・ 、 Next, "I will notify you at 7:21:20 am ... Pawn ...", and also "7:21 am
I will inform you about 30 minutes. ... Pippipippon
・ ・ ”Is displayed at 10-second intervals. In addition, as described above, this time guidance includes a beep sound p1 for notifying the second every 1 second and a beep sound p for notifying the time every 10 seconds.
2. For example, it is composed of a human-announced voice signal v1 such as "I will inform you of 7:20:50 am from now," and the noise n on the line is also included. Every 30 seconds and every 60 seconds (just what time and minute), "pippippi" before the "pawn"
The dial tone p3 is always input.

Here, the time guidance signal shown in FIG. 2 (b) (this time guidance signal indicates "7:21:10 am ... Pawn ...") is the remote control. When input to the microphone 1, the processing shown in the flowchart of FIG. 5 is performed. In FIG. 5, first, it is determined whether or not the level of the signal taken in from the microphone 1 is equal to or higher than a specified value (step s1), and if it is equal to or higher than the specified value, the characteristic analysis of the input signal is performed (step s).
2). Next, it is determined whether or not a dial tone is present (step s3). If a dial tone is present, detection of the dial tone and its timing are detected (step s4).

In the present invention, since the voice recognition by DRNN is used as the voice recognition technique, it is not necessary to specify the voice position when performing the voice recognition, and as described above, the voice feature vector is used. It is not necessary to separate the feature vector of the dial tone from. In other words, since the identification of this dial tone only requires the dial tone as the timing when setting the time, p1, p2, and
It is not necessary to detect all the beep sounds of p3, and it is sufficient to detect the beep sound p2 for notifying the time and know its timing.

Further, it is not necessary to specify the voice position because the recognition candidate word (in this case, the word specifying the time) exists in the input signal as described in the voice recognition by the DRNN. Since the detection signal is output, the voice between the previous dial tone p2 and the next dial tone p2 is input, and if the recognition candidate word exists in the voice, the position is not specified. Also, the recognition output corresponding to it can be taken out.

Therefore, in the dial tone detection and the timing detection processing thereof in step s4, it is sufficient to detect the dial tone p2 and know its timing. The detection of the dial tone p2 can be easily specified based on the feature vector analyzed by the input signal analysis unit 3, and its timing can be detected as shown in FIG. 2 (c). In this way, when the dial tone p2 is detected and the timing thereof is detected, a voice recognition process is then performed (step s5).

At this point, the voice "I will notify you at 7:21:10 am ... Pawn ..." is input through the microphone 1 and the recognition candidates shown in FIG. It is assumed that candidates are output. FIG.
(A) shows "am", "7:00" included in the input voice,
For each word "21 minutes" and "10 seconds", "am" is the first recognition candidate for "am"
Is selected, “7 o'clock” is selected as the first recognition candidate for “7 o'clock”, “11 minutes” is selected as the first recognition candidate for “21 minutes”, and the first recognition candidate for “10 seconds” is selected. "10 seconds" was selected as the first recognition candidate, and
As the second recognition candidate, “afternoon” is selected for “am”, “8:00” is selected for “7:00”, and “21 minutes” is selected for “21 minutes”. Is selected and “1
"20 seconds" is selected for "0 seconds", and "7 o'clock" is selected as the third recognition candidate for hours, minutes, and seconds.
"1 o'clock" is selected for, "20 minutes" is selected for "21 minutes", and "30 minutes" is selected for "10 seconds".
In this example, "second" is selected.

In this way, for each word, several recognition candidate words are selected in order from the first rank for each word. The output of the recognition candidate is based on the numerical value (approximation degree) representing the certainty obtained by the word detection unit 52. For example, the word 2 (7 1) with an approximation of 0.8
Word 3 (8 o'clock) selected as a candidate for rank
The recognition candidate is selected based on the degree of approximation, such that 0.6 is selected as the second recognition candidate.

Then, the content "7:11:10" composed of the first candidate word for recognition is displayed on the time display section 6 of the remote controller (step s6). In addition, the time display unit 6 displays "7:11:10 pm", and then immediately displays the time in one second increments. While listening to the ever-changing time guidance emitted from the handset, the user looks at the display content of the time display section 6 of the remote controller and determines whether or not the displayed content matches the content emitted from the handset ( In step s7), if they match, it is determined that the voice has been correctly recognized, the voice recognition operation is terminated at that time, and the time setting process of the clock in the remote controller is performed based on the recognition result at that time (step s8).

However, in this example, the content from the handset is the time guidance starting from "7:21:10", and the first recognition result for "7:21:10" is "AM". 7:11:10 and the time display unit 7
The display content of "is 7:11:10," in 1 second increments, "7:11:11, 12 seconds, 13 seconds, ...
. ”, The contents of the time displayed on the time display unit 6 and the time guidance emitted from the handset do not match (in this case, the time in minutes is recognized differently). Exist). In such a case, the recognition operation is continued until they match.

FIG. 6B shows the recognition result obtained by continuing the recognition operation. In other words, “7:21 am 10
I will inform you of the second. After "Pawn ...",
It is said that the sound "I will notify you at 7:21:20 am ... Pawn ..." was input through the microphone 1 and the recognition candidates as shown in FIG. 6 (b) were output as the recognition candidates. To do. In this case, "AM", "7"
For each of the words “hour”, “21 minutes”, and “20 seconds”, “am” is selected as the first recognition candidate for “am” and as the first recognition candidate for “7 o'clock”. "7 o'clock" was selected, "21 minutes" was selected as the first recognition candidate for "21 minutes", and "20 seconds" was selected as the first recognition candidate for "20 seconds". Become. In addition, "am" is the second candidate for recognition.
"Pm" is selected for, "8:00" is selected for "7:00", and "11 minutes" is selected for "21 minutes"
Is selected, and “10 seconds” is selected for “20 seconds”. In this way, even in the recognition operation performed continuously (called the second recognition processing operation),
For each word, several recognition candidate words are selected in order from the first rank for each word.

Also in the second recognition operation, immediately after the content "7:21:20 am" composed of the first-ranked recognition candidate word is displayed on the time display section 6 of the remote controller. The time is displayed every second for every second. While listening to the ever-changing time guide emitted from the handset, the user looks at the display content of the time display unit 6 of the remote controller and determines whether or not the display content matches the content emitted from the handset. In this case, the content from the handset is the time guidance from "7:21:20 am",
The recognition result of the first place for this is "7:21:20 am
Seconds ”and the display content of the time display section 6 is“ 7: 2 am
From "1 minute 20 seconds" in 1 second increments, "" 7:21:21 am
Seconds, 22 seconds, 23 seconds, etc. ”are displayed. Therefore, it is assumed that the time displayed on the time display unit 6 is consistent with the time guidance issued from the handset and that the time is recognized correctly. At that point, the voice recognition operation ends.

As a result, the clock in the remote controller is brought into a state where the time is correctly set to the current time, and thereafter, the clock operation is performed at the correct time. The time setting means 7 performs the process of adjusting the time of the clock in the remote controller based on the result of the voice recognition. This time setting process reads the time data corresponding to the recognition result of each word from the memory based on the result of the voice recognition, and gives a beep sound p2 for notifying the time, that is, "what time, what minute, what second" I will inform you ... The time is adjusted in synchronization with the "pawn" of "pawn ...".

When the time of the remote controller is adjusted in this manner, the user pushes the time data transfer button 36 provided on the remote controller to transfer the time data to the VCR main body side. This makes it possible to set the clock time of the VCR to the correct current time.

As described above, in the first embodiment, the time guidance emitted from the telephone handset is input through the microphone 1 of the remote controller, and the voice recognition technology is used to correctly recognize the time guidance announcement. The voice recognition operation is performed up to, and when the user determines that the voice is correctly recognized, the voice recognition process is ended. This allows
The clock in the remote control starts ticking from the time it is correctly recognized.

Therefore, the user does not have to go through the troublesome procedure for adjusting the time, which is conventionally done, and the user can select "11.
You can automatically set the correct current time in seconds by simply dialing "7" and letting the remote control enter the time guidance emitted from the handset and determining whether or not voice recognition was performed correctly. Further, since the voice recognition means based on DRNN is used as the voice recognition technology, even if there is noise on the line, the noise can be highly accurately recognized without significantly affecting the word recognition output.

(Second Embodiment) In the first embodiment described above, the user makes a judgment as to whether or not the voice recognition is correctly performed, and when the user is correctly recognized, the subsequent recognition is performed. Although the processing is not performed, the recognition processing is not limited to this, and the recognition processing is automatically performed a preset number of times.
The time may be determined by obtaining the most probable recognition result based on each recognition result.

FIG. 7 is a flow chart for explaining this processing, which will be briefly described below with reference to FIG. Note that the configuration of the time adjustment processing unit used in this second embodiment is the same as that in the first embodiment (see FIG. 1). Also, in the second embodiment, as in the first embodiment, the first time information entered is "7:21:10 am.-
・ Pawn ... ”

Here, a case where the number of times of recognition processing is set to 6 will be described. First, the first recognition process (the number of times of recognition is i and i = 1) is performed. This recognition process is performed at the time when “7:22:10 am It will be announced.·
··· Pawns ... ” First, as in the first embodiment, first, it is determined whether or not the level of the signal taken in from the microphone 1 is equal to or higher than a specified value (step s11). Analysis is performed (step s12). Next, it is determined whether a dial tone is present (step s13), and if a dial tone is present, detection of the dial tone and its timing are detected (step s14). In this case, as in the first embodiment, it is not always necessary to detect all the dial tones p1, p2, and p3, and the dial tone p2 notifying the time is set.
Should be detected and the timing should be known. In this way, when the dial tone p2 is detected and the timing thereof is detected, the voice recognition process is then performed (step s15).

In this voice recognition process, as in the first embodiment, voice recognition is performed using the voice recognition technique by DRNN, and recognition candidates are output.

This first recognition processing is performed at "7: 2 am
We will inform you about 1 minute and 10 seconds. ... Pawn ... ", and several recognition candidate characters are selected in order from the top for each word.
That is, "am", "7:00", "21 minutes", "10"
Each feature vector of the input speech word "second",
By comparing with the feature vector registered in the standard voice feature data storage unit, to which part and on what degree of certainty the registered word exists on the time axis of the input voice, for each registered word, Output with a numerical value (proximity) indicating the certainty.

Next, i = i + 1 is set (step s1
6), it is determined whether or not the number of times of recognition has reached a set value (i = 6) (step s17), and if it has not reached the set value,
The second recognition process is started, but this second recognition process is 10 seconds after the first recognition process, "7:21:20 am .... Pawn ..." For guidance.

This recognition process is also the same as the above step s11.
~ S15 are performed. The second recognition process is “7 am
We will inform you of the time of 22 minutes and 20 seconds. ···pawn··
・ "Is for the time guidance, so" am ",
The feature vectors of the input voice words “7 o'clock”, “21 minutes”, and “20 seconds” are compared with the feature vectors registered in the standard voice feature data storage unit, and the registered words are the input voice words. For each registered word, which part on the time axis exists with what degree of certainty is output as a numerical value (approximation degree) indicating the degree of certainty.

Next, i = i + 1 is set (step s1
6), it is determined whether or not the number of times of recognition has reached a set value (i = 6) (step s17), and if it has not reached the set value,
The third recognition process starts, but this third recognition process is called "7:21:30 am 10 seconds after the second time. Pippipippon ..." For time guidance.

In this way, recognition processing is performed 6 times until i = 6. Then, the recognition result with the highest degree of approximation is obtained based on the recognition results of six times, the current time data is obtained based on the recognition result, and the time of the clock built in the remote controller is set (step s18). ).

A process of obtaining the most probable recognition result based on the recognition results of six times and obtaining the current time data based on the recognition result will be described below using a specific example.

Here, it is assumed that the actual time when the first recognition process is performed (time emitted from the telephone) is "7:21:10". Therefore, the actual time when the second recognition process is performed is “10:21:20 am” since it is 10 seconds after that, and the actual time when the third recognition process is performed is Since the time is 10 seconds later, it is “7:21:30”, and the fourth recognition process is 10 seconds later, so the actual time is “7:21:40”. , The fifth recognition process is 10 seconds later, so the actual time is “7: 2 am
1 minute and 50 seconds ", and the sixth recognition process requires 10
Since it is a second later, the actual time is “just 7:22 am”.

Now, in the first recognition process, the first recognition candidates for each word are "am" and "7".
“Hour”, “11 minutes”, “10 seconds”, and in the second recognition process, the first-ranked recognition candidates for each word are “AM”, “7:00”, “21 minutes”, “10”. Second ”, and in the third recognition process, the first-ranked recognition candidates for each word are“ AM ”,“ 7:00 ”, and“ 21 ”.
Minutes ”and“ 40 seconds ”, and in the fourth recognition process, the first-ranked recognition candidates for each word are“ am ”,“ 7:00 ”,“ 21 minutes ”, and“ 40 seconds ”, Fifth
In the recognition process of the first time, the first for each word
Recognition candidates for rank are “AM”, “7:00”, “21 minutes”, “5”
0 second ", and in the sixth recognition process, the first-ranked recognition candidates for each word are" am "and" 7 ".
It is assumed that they are "hour", "22 minutes", and "just".

With respect to such a recognition result, the specification of the time in seconds will now be described with reference to FIG. In addition,
Here, since the time in seconds sent from the telephone is every 10 seconds, the current time in seconds that can be taken by the device (in this case, the VCR) is (1) What time, what minutes and 10 seconds, and then , 20 seconds, 3
Time changes such as 0 seconds.

(2) Time changes such as what hour, minute and 20 seconds, and then 30 seconds, 40 seconds, and so on.

(3) Time change starting from what hour and minute and 30 seconds, and then 40 seconds, 50 seconds, and so on.

(4) Time change such as starting at what hour and minute and 40 seconds, and then 50 seconds, 60 seconds (just), and so on.

(5) Time change such as what hour, minute, and 50 seconds, and then 60 seconds (just), 10 seconds, and so on.

(6) A time change such as 10 hours, 20 seconds, ...

There are six possible ways, and any one of them can be set as the second unit of the current time.

As shown in FIG. 8, the first recognition candidate per second obtained in the first recognition process is "10".
Second ”, the first-ranked recognition candidate in the second unit obtained in the second recognition process is“ 10 seconds ”, and the first-ranked recognition candidate in the second unit obtained in the third recognition process. The candidate is “40 seconds”, the first recognition candidate in the second unit obtained in the fourth recognition process is “40 seconds”, and the fifth recognition candidate is “40 seconds”.
The first-ranked recognition candidate obtained in the second recognition process in seconds is “50 seconds”, and the first-ranked recognition candidate obtained in the sixth recognition process in seconds is “just”. .

Each of the first-ranked recognition candidates obtained by the first to sixth recognition processes is made to correspond to the above-mentioned time changes of (1) to (6). In FIG. 8, the circles are attached to the times of the portions that match the times of the first-ranked recognition candidates obtained by the first to sixth recognition processes.

As described above, the time (second unit) of each of the first-ranked recognition candidates obtained by the first to sixth recognition processing is associated with the actual time change (second unit). Then, from the result, it is determined how the second unit of the current time is changing. In this case, since the time change shown in (1) is most frequently associated, the change in seconds in (1) is determined to be the change in seconds in the current time. That is, the second unit of the current time is the time of the third recognition process, which is what hour and minute and 10 seconds at the time of the first recognition process, and what time and minute and 20 seconds at the time of the second recognition process. Then, it can be determined that what hour and minute and thirty seconds are ... And exactly what hour and minute at the time of the sixth recognition process.

When the time in seconds is determined in this way, next, in the specification in minutes, attention is paid to the fact that the minutes always change among the six recognition processes. If it changes, for example, 7:58 to 7:59, 7:59
The minute at the current time is specified by paying attention to the fact that it continuously changes every minute, such as just from 8:00 to 8:00 and from 8:00 to 8:00.

For example, the first-ranked recognition candidate per minute obtained in the first recognition processing is “11 minutes”,
The first-ranked recognition candidate in the minute unit obtained in the second recognition process is “21 minutes”, and the first-ranked recognition candidate in the minute unit obtained in the third recognition process is “20 minutes”. , And the first-ranked recognition candidate in the minute unit obtained in the fourth recognition process is “21 minutes”, and the first-ranked recognition candidate in the minute unit obtained in the fifth recognition process. Is "21
It is assumed that “22 minutes” is the first-ranked recognition candidate in the minute unit obtained by the sixth recognition process.

When the minute of the current time is specified based on the first-ranked recognition candidates on the minute-by-minute basis obtained by the recognition processing from the first time to the sixth time, as described above, Focusing on the fact that the minute always changes in the recognition process and the continuous change every minute, the minute at the current time is specified based on the recognition candidates from the first time to the sixth time.
In this case, based on the information on the recognition result in units of minutes from the first to sixth times and the information that the minutes have changed in the sixth recognition process in the time setting in seconds, It is determined that the minute at the time of the first to fifth recognition processing is “21 minutes” and the minute at the time of the sixth recognition processing is “22 minutes”.

When the time in minutes is determined in this way, the time unit is specified next. However, since the time is a long time change of one hour, it can be obtained by six recognition processes. Judgment is made by integrating the results obtained.

For example, the first recognition candidate in the time unit obtained in the first recognition process is “7:00”,
The first-ranked recognition candidate in the time unit obtained in the second recognition process is “8 o'clock”, and the first-ranked recognition candidate in the time unit obtained in the third recognition process is “1 o'clock”. It is assumed that all the first-ranked recognition candidates in the time unit obtained by the fourth and subsequent recognition processes are “7 o'clock”.

Of the first-ranked recognition results in the first to sixth recognition processing, "7:00" is four times,
There is once each "8 o'clock" and "1 o'clock".
The current time is determined to be “7:00”.

As a result of the above processing, the current time is "7:21:10" at the time of the first recognition processing and the second time.
At the time of the second recognition processing, "7:21:20",
At the time of the third recognition processing, “7:21:30
Seconds, ..., At the time of the sixth recognition processing, “7
It is determined to be "just 22 minutes at the hour". Based on this, when the clock in the remote controller is set to the current time, a guide sound or the like is issued to inform the user that the time adjustment is completed. As a result, the user pushes the time data transfer button 36 provided on the remote controller to transfer the time data to the VCR main body side. This allows the clock on the VCR to be set to the correct current time.

In the second embodiment, the number of times the recognition processing is performed is 6 because the more reliable the recognition processing can be performed if the number of times the recognition processing is performed is increased. It is not preferable to spend a lot of time in terms of usability, and there are some limitations. On the other hand, if the number of times is extremely small, there is a problem in terms of recognition accuracy, and about 6 times is optimal in consideration of the recognition processing time and recognition accuracy. In addition, by performing recognition 6 times,
This is because the minute changes, which is advantageous for improving the recognition accuracy. However, it goes without saying that this is not limited to six times.

As described above, in the second embodiment, the time guidance emitted from the handset is input only through the microphone of the remote controller, and then the device side automatically announces the time guidance a predetermined number of times. Speech recognition processing is performed to determine the most probable time, and the current time data is obtained. Therefore, the user
Simply call "117" and let the remote control enter the time guidance sent from the handset without taking the troublesome procedure that was used to adjust the time. You can set the current time. Further, as in the first embodiment of the second embodiment, since the voice recognition means by DRNN is used as the voice recognition technology, even if there is noise in the line, the noise causes a large word recognition output. Highly accurate voice recognition can be performed without affecting.

(Third Embodiment) In the second embodiment, the number of times of recognition processing is set in advance to, for example, six times, and at the time when the recognition processing is performed the set number of times, Although the time is determined, in the third embodiment, the recognition process is terminated at the time when sufficient certainty is obtained without setting the number of times the recognition process is performed, and the certainty is determined. The present time data is obtained based on the time determined to be sufficient. This process is shown in the flowchart of FIG.

Hereinafter, description will be made with reference to FIG. The configuration of the time adjustment processing unit used in the third embodiment is the same as that in the first embodiment (see FIG. 1). Also in the third embodiment, as in the above case, the first time guidance entered is "I will notify you at 7:21:10 am ... Pawn ...
-"Signal.

First, the first recognition process is performed. This recognition process notifies the above-mentioned “7:21:10, which is input to the microphone 1 of the remote controller at that time. .. ” First, similarly to the first embodiment, first, the microphone 1
It is determined whether or not the level of the signal taken in from is equal to or higher than a specified value (step s21), and if it is equal to or higher than the specified value, characteristic analysis of the input signal is performed (step s22). Next, it is determined whether or not a dial tone is present (step s
23) If there is a dial tone, detection of the dial tone and its timing are detected (step s24). Note that in this case, as in the first embodiment, p1, p2, p3
It is not necessary to detect all the beep sounds, and it is sufficient to detect the beep sound p2 for notifying the time and to know its timing.
In this way, when the dial tone p2 is detected and the timing thereof is detected, next, the voice recognition process is performed (step s
25).

This voice recognition processing is registered in the standard voice feature data storage unit as well as the feature vectors of the input voice words of “AM”, “7:00”, “21 minutes”, and “10 seconds”, as described above. Comparing with the feature vector
For each registered word, the degree to which the registered word exists on the time axis of the input voice with certainty,
Output with a numerical value (proximity) indicating the certainty.

This first recognition process is performed at "7: 2 am
We will inform you about 1 minute and 10 seconds. ... Pawn ... ", and several recognition candidate characters are selected in order from the top for each word.

Next, it is judged whether or not the numerical value (approximation degree) indicating the certainty is sufficient for the obtained recognition candidate (step s26). Hereinafter, this processing will be described.

For example, for the voice "7 o'clock",
It is assumed that a recognition candidate such as “7:00” is selected as the first recognition candidate, “8 o'clock” is selected as the second recognition candidate, and “1 o'clock” is selected as the third recognition candidate. These several recognition candidates are selected by comparing each feature vector of the input voice word with the feature vector registered in the standard voice feature data storage unit to determine which of the registered words on the time axis of the input voice. For each registered word, the degree of certainty that exists in the portion is output as a numerical value (approximation degree) that indicates the certainty, but one of these recognition candidates is determined as the recognized voice. To do this, the following method is used.

For example, a threshold value is set in advance for the degree of approximation, and if the degree of approximation is greater than or equal to the threshold value, it is determined as the recognized voice. Specifically, set the threshold to 0.
If it is set to 8 and the degree of approximation is 0.8 or more, it is determined as the recognized voice. On the other hand, if none of the recognition candidates have the degree of approximation equal to or higher than the threshold value, the recognition process is repeated for the second time and the third time, and in the meantime, the recognition candidate whose value is equal to or higher than the threshold value is detected. If there is, it is determined as the recognized voice. Further, if there is no recognition candidate that is equal to or higher than the threshold value even after performing the recognition process a certain number of times, a recognition candidate having a stable large degree of approximation is determined as a recognition voice.

The determination as to whether or not such certainty is sufficient is made for each word, and when it is determined that the certainty of each word is sufficient, the current time data is obtained based on that and the remote controller The time of the clock is set (step s27).
In this way, when the clock in the remote controller is set to the current time, a guide sound or the like is issued to inform the user that the time adjustment has been completed. As a result, the user pushes the time data transfer button 36 provided on the remote controller to transfer the time data to the VCR main body side. This allows the clock on the VCR to be set to the correct current time.

As described above, in the third embodiment, the user simply inputs the time guidance emitted from the handset through the microphone of the remote controller, and then the device automatically announces the time guidance. When the recognition processing is performed and it is determined that the recognition is correctly performed, the recognition processing is ended and the current time data is obtained. Therefore, the user does not have to go through the troublesome procedure used to adjust the time, simply dials “117” and inputs the time guidance from the handset into the remote control. You can set the exact current time in seconds to. In addition, in the third embodiment, the first,
As in the second embodiment, since the voice recognition means based on DRNN is used as the voice recognition technique, even if there is noise on the line, the noise does not have a great influence on the word recognition output, and highly accurate voice recognition is possible. It can be performed.

In the first to third embodiments, the time guidance to be input is NTT sent from the telephone.
However, the present invention is not limited to this, and the same time broadcasting as the time service of NTT broadcast on shortwave radio can be performed in the same manner.
Furthermore, it is possible to realize even a single time guidance such as a TV or radio time signal.

Further, in the first to third embodiments, the example in which the time of the VCR is adjusted has been described, but the present invention is not limited to the VCR, and home electric appliances, electronic devices, and telephones with an answering machine function. It can be widely applied to devices in which a clock is built-in and the time of the clock needs to be adjusted by the user. Further, in each of the above-described embodiments, the "117" is dialed to obtain the time guidance, but the operation can be further simplified by incorporating the function of the phone dialer. In other words, by providing a time signal dial button on the remote control, raising the handset of the telephone, bringing the remote control closer to the telephone, and pressing the time signal dial button, the "117" automatically
It is also possible to connect to the telephone and to perform time recognition by performing the voice recognition processing as described above.

In addition, among the above-mentioned various devices, if the device does not have the time adjusting function by the remote controller, the device itself may be provided with the time adjusting function by the voice recognition of the present invention. Also, when the present invention is incorporated in a telephone having a built-in clock (such as a telephone with an answering machine known to oneself), the machine automatically selects "117" once a week, for example.
It is also possible to connect the line to and automatically adjust the time of the user's own clock to the current time.In addition, if the present invention is incorporated into a radio or TV with a built-in clock, the machine will automatically set the time. Broadcast (eg 7am, 7pm)
It is also possible to automatically recognize your own clock by recognizing the time of day, noon, etc.).
In this way, the built-in timepiece will always operate at the correct time.

The programs for performing the respective processes of the present invention described above can be stored in a storage medium such as a floppy disk, and the present invention also includes a storage medium in which those programs are stored. .

[0098]

According to the present invention, in a method for adjusting the current time of a timepiece in a device having a built-in timepiece, a time guidance including voice is input, and a voice portion for warning the current time is recognized by voice recognition. Based on the recognition result, the time of the built-in clock is set at a predetermined timing to the time at which the time guidance is given, so the user has conventionally performed the time adjustment. You can set the correct current time automatically by simply inputting the time guide consisting of voice and dial tone without going through the troublesome procedure. As a result, even a user who is unfamiliar with handling this type of device can easily adjust the time. In particular, if the time service provided by NTT's "117" is used as the time guide composed of the voice and dial tone, the time guide issued from the handset by calling "117" at any time. You can set the correct current time in seconds automatically by simply inputting. By the way, the device with a built-in clock is
Alternatively, it is necessary to adjust the time each time, for example, when the plug is unplugged after the power failure is restored, which is very troublesome and the setting of the accurate time is even more troublesome. For example, since it is possible to set the time accurately in seconds by a simple operation, it is possible to eliminate the feeling of resistance to the time setting of the clock, and to greatly improve the usability of this type of device.

Further, since the present invention uses the DRNN voice recognition means developed by the present applicant as a voice recognition technique, even if there is noise on the line, the noise has a great influence on the word recognition output. Without this, highly accurate voice recognition can be performed. Further, since the voice recognition technology by the DRNN enables the voice recognition by the unspecified speaker, the information source for notifying the time is not particularly limited, and a wide variety of information sources can be used. Furthermore, the voice recognition technology by the DRNN can be realized at an extremely low cost as a voice recognition technology that enables unspecified speaker voice recognition, so that the cost of a product incorporating the voice recognition technology is not significantly affected. , Can be widely applied to the equipment used on a daily basis.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a time adjustment processing unit according to an embodiment of the present invention.

FIG. 2 is a diagram for explaining a time guidance signal by NTT “117”.

FIG. 3 is a DRN used in an embodiment of the present invention.
The figure explaining the voice recognition by N.

FIG. 4 is a surface appearance diagram of a remote controller of a device used in the embodiment of the present invention.

FIG. 5 is a flowchart illustrating a process according to the first embodiment of this invention.

FIG. 6 is a diagram showing an example of a voice recognition result in a voice recognition unit.

FIG. 7 is a flowchart illustrating processing according to the second embodiment of this invention.

FIG. 8 is a diagram illustrating a specific example of time in seconds according to the second embodiment.

FIG. 9 is a flowchart illustrating processing according to the third embodiment of this invention.

[Explanation of symbols]

 1 sound input unit (microphone) 2 input signal processing unit 3 input signal analysis unit 4 dial tone detection unit 5 voice recognition unit 6 time display unit 7 time setting unit 31 power button 32 menu button 33 video / TV switch button 34 “0” ~ Numerical buttons of "9" 35 Buttons for recording, playing, rewinding, fast-forwarding 36 Time data transfer button 51 Standard voice feature data storage section 52 Word detection section 53 Voice recognition processing section

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁶ Identification number Internal reference number FI Technical display location // G06F 15/18 560 G06F 15/18 560G (72) Inventor Mitsuhiro Inazumi 3 Yamato, Suwa City, Nagano Prefecture 3-5, Seiko-Epson Co., Ltd. (72) Inventor, Osamu Urano, 3-3-5, Yamato, Suwa-shi, Nagano (72) In-house, Seiko-Epson, Inc. (72) 3-3.5, Yamato, Suwa, Nagano Prefecture No. Seiko Epson Corporation

Claims (6)

[Claims] 1. When adjusting the current time of a clock built in a device, a time guide including voice is input, the voice part of the time guide is voice-recognized, and the time of the clock is set based on the recognition result. In a time adjustment method using voice recognition, the time is set at a predetermined time at a current time indicated by time guidance, and the means for performing the voice recognition is a dynamic recurrent neural network (Dynamic Recurrent Neural Network).
The voice feature data for the input voice obtained by analyzing the voice included in the time guidance input by the voice inputting means by using the voice recognizing means by the keyword spotting process using the Networks method, and for each registered word. Compared with the standard voice feature data, the registration word is present in which part on the time axis of the input voice with a certain degree of certainty, for each registered word, it is detected by a numerical value indicating the certainty, Based on the numerical value indicating the certainty, the input voice is recognized, and based on the recognition result, the time of the clock is set to the time of the time indicated by the time guidance at a predetermined timing. Time adjustment method using featured voice recognition. 2. The time at which the built-in clock is set based on the recognition result is voice-recognized for each voice of the time guidance corresponding to the passage of time, and the voice recognition result for each time is obtained. 2. The time adjustment method using voice recognition according to claim 1, wherein the time is set to a time when it is determined that the certainty of is sufficient. 3. The time at which the built-in clock is set based on the recognition result is obtained by performing a preset number of times of voice recognition for each time guidance voice corresponding to the passage of time, and The time setting method using voice recognition according to claim 1, wherein the time is determined based on the voice recognition result for the time guidance voice. 4. When adjusting the current time of a clock built in the device, a time guide including voice is input, the voice part of the time guide is voice-recognized, and the time of the clock is detected based on the recognition result. A time input device for inputting time guidance including voice, in a time setting device for a timepiece using voice recognition, which sets time at a predetermined time to a time point to be guided by the time guidance, and a sound input unit for inputting the time guidance. Voice feature data for the input voice obtained by analyzing the voice that announces the current time input to the means,
By comparing with standard voice feature data for each registered word, it is shown for each registered word, the certainty of which part on the time axis of the input voice the certainity of the registered word is and with what degree of certainty. A voice recognition unit based on keyword spotting processing using a dynamic recurrent neural network method that detects numerical values, and the time of the clock based on the recognition result by the voice recognition unit is set to a predetermined time at the time point when the time guidance is provided. A time setting device using voice recognition, comprising: 5. The time at which the built-in clock is set based on the recognition result is voice-recognized for each time guidance voice corresponding to the passage of time, and the voice recognition result for each time is obtained. 5. The time adjustment device using voice recognition according to claim 4, wherein the time is a time at which the certainty of is determined to be sufficient. 6. The time when the built-in clock is set based on the recognition result is obtained by performing a preset number of times of voice recognition for each time guidance voice corresponding to the passage of time. 5. The time adjustment device using voice recognition according to claim 4, wherein the time determined based on the voice recognition result for the time guidance voice is used.