JP2004294659A - Speech recognition device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a speech recognition device capable of speech recognition through a simple operation wherein physical and mental burdens on a speaker are reduced. <P>SOLUTION: The speech recognition device is equipped with a sound signal output means 11 of inputting a sound including a speaker's speech and outputting a sound signal, a contact means 12 that the speaker touches when speaking, a speech section decision means 13 of deciding a speech section wherein the speaker speaks words according to a speech signal included in the sound signal and the contact state of the contact means 12, a speech recognition means 14 of recognizing the speech in the speech section, and a display means 15 of displaying a speech recognition result, thereby enabling speech recognition through easy operation. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a voice recognition device, and more particularly, to a voice recognition device that recognizes voice uttered by a speaker.
[0002]
[Prior art]
As a conventional speech recognition device, the one shown in FIG. 5 is known. The voice recognition device 1 shown in FIG. 5 includes a voice input unit 2 for inputting voice, a switch 3 operated by a speaker, a switch state detection unit 4 for detecting a state of the switch 3, and a state of the switch 3. The apparatus includes an input control unit 5 that controls the input of voice in response to the voice, a voice recognition unit 6 that recognizes voice, and a display unit 7 that displays a voice recognition result.
[0003]
In the conventional voice recognition device 1, first, the switch state detection unit 4 determines whether or not the speaker has turned on the switch 3 from off. When it is determined that the switch 3 has been turned on from off, the input control unit 5 notifies the voice input unit 2 of the start of voice input. Next, a voice is input by the voice input unit 2. Then, the input voice is recognized by the voice recognition unit 6, and the recognition result is displayed on the display unit 7.
[0004]
Subsequently, the switch state detection unit 4 determines whether or not the speaker has turned off the switch 3 from on. When it is determined that the switch 3 has been turned off from on, the input control unit 5 notifies the voice input unit 2 of the end of voice input with a delay of a predetermined time.
[0005]
As described above, in the conventional voice recognition device 1, the switch status detection unit 4 detects the status of the switch 3 and can recognize the voice input when the switch 3 is on (for example, And Patent Document 1).
[0006]
[Patent Document 1]
JP-A-2002-108390 (pages 4 to 5, FIG. 1)
[0007]
[Problems to be solved by the invention]
However, in such a conventional voice recognition device, since the speaker must keep pressing the switch continuously during the utterance, it is necessary to perform a complicated operation that places a large physical and mental burden on the speaker. There was a problem.
[0008]
The present invention has been made in order to solve such a problem, and provides a voice recognition device capable of performing voice recognition by a simple operation with reduced physical and mental burden on a speaker. It is.
[0009]
[Means for Solving the Problems]
The voice recognition device of the present invention includes: a sound signal output unit that inputs sound including a speaker's voice and outputs a sound signal; a contact unit that contacts the speaker when the speaker speaks; A voice section determination unit that determines a voice section in which the speaker is emitting the voice based on a voice signal included in the voice signal and a contact state of the contact unit; and a voice recognition unit that recognizes the voice in the voice section. It has a configuration characterized by being provided.
[0010]
With this configuration, the voice section determination unit determines the voice section in which the speaker is emitting a voice based on the voice signal included in the audio signal and the contact state of the contact unit, and the voice recognition unit uses the voice section determination unit. Since the voice of the determined voice section is recognized, the voice section is determined by the continuous or intermittent contact of the speaker during natural utterance, and the physical and mental burden on the speaker is reduced. Voice recognition can be performed with reduced simple operations.
[0011]
Further, in the voice recognition device of the present invention, the contact means detects a contact portion that contacts the speaker and whether the speaker is in contact with the contact portion, and the speaker detects the contact portion. A contact operation state signal indicating a contact operation state that is in contact with a contact state detection unit that outputs one of a contact stop state signal indicating a contact stop state in which the speaker has stopped contacting the contact unit. The audio section determination means includes: a power calculation unit that calculates the power of the audio signal; a power threshold setting unit that sets a power threshold of the audio signal according to a surrounding noise level; A point in time when the power of the audio signal exceeds the power threshold in the vicinity of the output time is set back to a predetermined time and is set as an audio section start point, and the audio is output in the vicinity of the point in time when the contact stop state signal is output. And a voice section determination unit that determines the voice section by setting a time point at which a predetermined time has elapsed from a time point at which the power of the signal falls below the power threshold value as a voice section end time point. I have.
[0012]
With this configuration, the voice section determination unit determines the voice section start time as the voice section start time when a predetermined time has elapsed from the time when the power of the voice signal exceeded the power threshold near the time when the contact operation state signal was output, and Since a voice section is determined by determining a voice section end time as a time point at which a predetermined time has elapsed from the time point at which the power of the voice signal falls below the power threshold near the time point when the signal is output, at the start of the voice section, It is possible to prevent a voice with low power, such as unvoiced consonants and unvoiced vowels, from being lost. Can be prevented.
[0013]
Further, in the voice recognition device of the present invention, the voice section determination means may include a voice input when the duration of the contact stop state following the contact operation state is equal to or less than a predetermined threshold value and the voice input before the contact stop state. Is included in the same voice section in the same voice section.
[0014]
With this configuration, it is possible to set a voice section corresponding to a voice input corresponding to the contact interval of each speaker.
[0015]
Also, the voice recognition device of the present invention, the voice section determination means includes a voice signal determination unit that determines whether the voice signal is included in the audio signal, the voice section determination unit, When the signal determination unit determines that the audio signal includes the audio signal, the configuration is such that the audio section is determined based on the contact state and the power of the audio signal. Have.
[0016]
With this configuration, the voice section determination unit determines the voice section based on the contact state and the power of the voice signal when the voice signal determination unit determines that the voice signal includes the voice signal. Even when the power of the acoustic signal greatly fluctuates due to ambient noise, it is possible to reliably determine the voice section.
[0017]
Further, in the voice recognition device according to the present invention, the voice section determination unit may determine that the power of the voice signal exceeds the power threshold near a time when it is determined that the voice signal is included in the voice signal. The audio section start time is defined as a point in time that is earlier than a predetermined time from the time when the power of the audio signal falls below the power threshold in the vicinity of the point in time when the contact stop state signal is output. The voice section is determined as the end point.
[0018]
With this configuration, the voice section determination unit determines that the voice section starts at a point in time when the power of the voice signal exceeds the power threshold and goes back a predetermined time near the time when the voice signal is determined to include the voice signal. The voice section is determined as the time point, and the time point when a predetermined time has elapsed from the time point when the power of the voice signal falls below the power threshold near the time point when the contact stop state signal is output is determined as the voice section end time point. At the beginning of the voice section, it is possible to prevent the lack of low-power voices such as unvoiced consonants and unvoiced vowels at the beginning of the voice section. It is possible to prevent a low-power voice such as a consonant or a unvoiced vowel from being lost.
[0019]
BEST MODE FOR CARRYING OUT THE INVENTION
Hereinafter, embodiments of the present invention will be described with reference to the drawings.
[0020]
First, the configuration of the speech recognition device according to the embodiment of the present invention will be described.
[0021]
As shown in FIG. 1, a voice recognition device 10 according to the present embodiment includes a sound signal output unit 11 that inputs a sound including a speaker's voice and outputs a sound signal, and a contact that contacts when the speaker speaks. A voice section determining means for determining a voice section in which a speaker is uttering a voice based on a voice signal included in an audio signal and a contact state of the contact means, and voice recognition for recognizing voice in the voice section. Means 14 and display means 15 for displaying the result of voice recognition.
[0022]
The acoustic signal output unit 11 includes a microphone 11a that collects sound and converts the sound into a sound signal, and an AD converter 11b that converts an analog signal into a digital signal.
[0023]
The contact means 12 detects a contact portion 12a with which the speaker is in contact and whether or not the speaker is in contact with the contact portion 12a, and indicates a contact operation state in which the speaker is in contact with the contact portion 12a. A contact state detection unit 12b that outputs either a state signal or a contact stop state signal indicating a contact stop state in which the speaker stops contacting the contact unit 12a. In the following description, the contact operation state and the contact stop state are referred to as a contact state and a non-contact state, respectively, and the contact operation state signal and the contact stop state signal are referred to as a contact signal and a non-contact signal, respectively.
[0024]
The contact portion 12a is configured by, for example, a keyboard, a switch, a piezoelectric element, a thermal element, or the like. The contact state detection unit 12b is configured by, for example, a CPU, a RAM, a ROM, and the like.
[0025]
The voice section determination means 13 includes, for example, a CPU, a RAM, a ROM, and the like, and includes a power calculation unit 13a that calculates the power of the audio signal, and an audio signal determination that determines whether the audio signal includes the audio signal. Unit 13b, a power threshold setting unit 13c for setting a power threshold of the audio signal according to the ambient noise level, a storage unit 13d for storing a predetermined threshold value and a variable, and the like, and a contact signal is output from the contact state detection unit 12b. A point in time near the time when the power of the audio signal exceeds the power threshold in the vicinity of the predetermined time is defined as a voice section start time, and in the vicinity of the time when the non-contact signal is output from the contact state detection unit 12b, And a voice section determination unit 13e that determines a voice section as a voice section end time when a predetermined time has elapsed since the power of the power falls below the power threshold. Eteiru.
[0026]
The power calculation unit 13a calculates the power of the audio signal output from the audio signal output unit 11, for example, every 20 msec. The audio signal determination unit 13b calculates at least one of an autocorrelation function, a zero-cross frequency, and a low-order cepstrum coefficient relating to the audio signal output from the audio signal output unit 11, and determines in advance a large amount of audio learned. The coefficient is used to determine whether or not the audio signal includes a voice signal (hereinafter referred to as voice determination). According to the determination result, either a signal representing voice or a signal representing non-voice is determined. For example, it outputs to the voice section determination unit 13e every 20 msec.
[0027]
The power threshold setting unit 13c dynamically sets the power threshold according to the surrounding noise level included in the audio signal output from the audio signal output unit 11. When the sound signal output from the sound signal output means 11 exceeds the power threshold, it is determined whether the sound signal is based on the surrounding noise level or the sound of the speaker. This can be performed based on the detection result.
[0028]
The voice recognizing means 14 includes, for example, a CPU, a RAM, a ROM, and the like, and recognizes the voice of the voice section determined by the voice section determining means 13. The voice recognition unit 14 includes a recognition vocabulary dictionary (not shown). The display means 15 includes a liquid crystal display, a CPU, a RAM, a ROM, and the like, and is configured to display a speech recognition result in characters, for example.
[0029]
Next, the operation of the speech recognition apparatus 10 according to the present embodiment will be described with reference to FIG.
[0030]
In FIG. 2, first, sound is collected by the microphone 11a, and the sound signal that has been AD-converted by the AD converter 11b is output to the power calculator 13a, the sound signal determiner 13b, and the sound section determiner 13e ( Step S21). Next, the power of the acoustic signal is calculated by the power calculation unit 13a, and the voice property is determined by the voice signal determination unit 13b (step S22).
[0031]
Subsequently, the contact state detection unit 12b detects whether or not the speaker has contacted the contact unit 12a, and if so, determines whether or not the contact has been made for the first time (step S23). Here, when the contact portion 12a is configured by a piezoelectric element, the pressure when the speaker touches the contact portion 12a is detected by the contact portion 12a, and the contact signal is transmitted from the contact state detection portion 12b to the voice section determination portion 13e. Is output. On the other hand, when the speaker is not touching the contact section 12a, a non-contact signal is output from the contact state detection section 12b to the voice section determination section 13e. Note that the above-mentioned first contact is a time threshold from the time when a certain contact starts, for example, a contact when there is no contact within a time period of 400 msec, or P ≧ Pth1 in step S24 to be described later and an audio signal. After the case where it is not determined, it means a contact when there is no contact within a period of time preceding the time when a certain contact has started by a time threshold.
[0032]
If it is determined in step S23 that the contact has been made for the first time, the voice section determination unit 13e determines whether the power value P calculated by the power calculation unit 13a is equal to or greater than the first power threshold Pth1 set by the power threshold setting unit 13c. The determination as to whether or not the voice signal is to be made and the determination of the voice of the acoustic signal output from the acoustic signal output means 11 are executed (step S24).
[0033]
In step S24, if P ≧ Pth1 and the voice is determined to be voice, the voice section start time is determined by the voice section determination unit 13e (step S25), while P ≧ Pth1 and the voice is voice. If not, the process returns to step S21.
[0034]
On the other hand, if it is not determined in step S23 that the contact has been made for the first time, the contact state detection unit 12b detects whether or not the speaker has contacted the contact unit 12a (step S26). In step S26, if it is detected that the speaker has touched the contact portion 12a, the process returns to step S21. If it is not detected that the speaker has touched the contact portion 12a, the voice section determination unit 13e stores the information. The time threshold Tth is read from the unit 13d, and it is determined whether the non-contact state continues for the time threshold Tth or more (step S27).
[0035]
If it is determined in step S27 that the non-contact state continues for the time threshold value Tth or more, the power value P calculated by the power calculation unit 13a is set by the voice section determination unit 13e by the power threshold setting unit 13c. It is determined whether or not the second power threshold Pth2 or less (Step S28). If it is not determined that the non-contact state has continued for the time threshold Tth or more, the process returns to Step S21.
[0036]
If it is determined in step S28 that P ≦ Pth2, the voice section determination unit 13e determines the end point of the voice section (step S29). If it is not determined that P ≦ Pth2, the process returns to step S21.
[0037]
Further, the voice recognition means 14 recognizes voice included from the voice section start time to the voice section end time determined by the voice section determination unit 13e (step S30). Then, the result of the voice recognition is displayed by, for example, characters by the display means 15 (step S31).
[0038]
Here, the process of determining a voice section will be described with a specific example. As the conditions for determining the voice section, the power of the acoustic signal, the determination result of the voiceness, and the contact state are shown in FIGS. 3A, 3B, and 3C, respectively. It will be described as having been performed. Further, the power threshold Pth1 and the power threshold Pth2 are 6 dB and 1 dB, respectively. The above-mentioned time threshold Tth is set to 400 msec, and the reference time when judging the start and end of the voice section based on the power threshold Pth1 and the power threshold Pth2 is set to 500 msec.
[0039]
First, the contents shown in FIGS. 3A to 3C will be described.
[0040]
In FIG. 3A, the horizontal axis represents time, the vertical axis represents the power of the audio signal, and a power curve 41 in which the power value of the audio signal output every 20 msec is plotted; a minimum power hold value 42; A point 43 and a second point 44 are shown.
[0041]
FIG. 3B shows the result of the voice determination performed by the voice signal determination unit 13b. The voice is determined to be voice from time t3 to time t4, and the voice is determined to be non-voice before time t3 and after time t4. It is shown that.
[0042]
FIG. 3C illustrates a contact signal indicating contact and a non-contact signal indicating non-contact output by the contact state detection unit 12b. The contact signal is output in the range from time t5 to t6, time t7 to t8, and time t9 to t10, and in other ranges, a non-contact signal is output. That is, FIGS. 3A and 3C show that the speaker uttered while touching the contact portion 12a three times.
[0043]
Next, determination of the voice section start time will be described.
[0044]
Due to the contact in the range of times t5 to t6 shown in FIG. 3C, it is determined that the contact has been made for the first time in step S23, and the process proceeds to step S24. As shown in FIG. 3A, the power value P of the acoustic signal starts to increase near the time t3, and at the first point 43 at the time t1, is 6 dB (Pth1) for the first time with respect to the minimum hold value 42 for the past 500 msec or more. That is all. At this time, as shown in FIG. 3B, the voice property is determined to be voice.
[0045]
Therefore, the judgment conditions in step S23 and step S24, that is, the detection result of the contact state detection unit 12b satisfies the three conditions of contact, P ≧ Pth1, and the vocality determination result satisfies the three conditions of the voice section determination. The unit 13e determines that the voice section has started, and as shown in FIG. 3D, the time t11 which is 600 msec earlier than the time t1 is determined as the voice section start time.
[0046]
Here, the point of time 600 msec before the time t1 is set as the voice section start time. In the above-described determination using the power and the voice property, for example, the power becomes low at the beginning of a word, the beginning of a sentence, etc. This is to prevent a situation in which the voice to be silenced cannot be detected and the head is cut off.
[0047]
Next, determination of the end point of the voice section will be described.
[0048]
Due to the contact in the range from time t9 to t10 shown in FIG. 3C, it is determined that the contact has occurred in step S26, and the process proceeds to step S27. As shown in FIG. 3 (c), after time t10, it is determined that the contact is non-contact continuously for the time threshold value Tth of 400 msec or more, so that the process proceeds from step S27 to step S28.
[0049]
The power value P of the acoustic signal starts to fall from near the time t9, and at the second point 44 at the time t2, becomes 1 dB (Pth2) or less for the first time with respect to the minimum hold value 42 for the past 500 msec or more.
[0050]
Therefore, the judgment condition in step S27 and step S28, that is, the detection result of the contact state detection unit 12b satisfies the two conditions of non-contact and P ≦ Pth2, so that the voice section is ended by the voice section determination unit 13e. Thus, as shown in FIG. 3D, time t12 at which 400 msec has elapsed from time t2 is determined as the end of the voice section.
[0051]
When contact is detected again within 400 msec of the time threshold Tth from the time when non-contact is detected by the contact state detection unit 12b, the contact and utterance of the speaker are intermittently performed by the voice section determination unit 13e. It is determined that the voice section is continued, and the voice section before the time of non-contact and the voice section after the time of contact again are the same voice section.
[0052]
For example, in FIG. 3C, if the time from time t6 to time t7 is 250 msec and the time from time t8 to time 9 is 200 msec, both times are equal to or less than the time threshold Tth of 400 msec. 8, and the sounds emitted at times t9 to t10 are sounds in the same section.
[0053]
Therefore, not only when the speaker continues to stroke the contact portion 12a during utterance, but also when the speaker strikes the contact portion 12a in synchronization with the syllables and the beats of the syllables, the speech segment determination unit 13e can accurately determine the speech segment. Can be determined.
[0054]
Note that the start point of the voice section may be determined based on the determination result of the voice signal determination unit 13b. That is, the time point when the power of the audio signal exceeds the power threshold value Pth1 by a predetermined time near the time point when the audio signal is determined to be audio by the audio signal determination unit 13b may be set as the audio section start time point. Good.
[0055]
Next, the processing of the voice recognition means 14 will be described with reference to FIG.
[0056]
FIGS. 4 (a) and 4 (b) show the same graphs as those of FIGS. 3 (a) and 3 (d), respectively, and FIG. 4 (c) shows a pre-stored vocabulary standard. This is a diagram schematically illustrating a section in which a pattern and a recognized result match well.
[0057]
In FIG. 4C, for example, if the first standard pattern 51 to the fourth standard pattern 54 are respectively “Tokyo”, “Yokohama”, “Chiba”, and “Sendai”, the first standard pattern A section that well matches 51 “Tokyo” is indicated by four arrows. Whether or not they match well is represented by a score representing the degree of coincidence between the standard pattern stored in advance and the input voice pattern, and is determined based on whether or not this score exceeds a predetermined threshold. Finally, the standard pattern having the maximum score near the voice section end time t12 is determined as the recognition result. FIG. 4C shows that, out of the second standard patterns 52 “Yokohama”, the second standard pattern 52 a represented by a thick arrow is a recognition result.
[0058]
In other words, the voice recognition means 14 performs a keyword spotting type operation of recognizing the input voice as needed, with the voice section start time of the voice section determined by the voice section determination means 13 as the start point and ending near the voice section end time. A recognition result of a vocabulary or a sentence to be a point is output.
[0059]
In general, a speaker often hits the contact portion 12a in accordance with a beat (mora) during his or her utterance, and often does not hit with a special beat such as sound generation, prompting sound, and long sound. By referring to the output from the unit 12b, it is possible to easily recognize a syllable including a special beat.
[0060]
Further, if the configuration is such that the speaker can arbitrarily set the above-mentioned time threshold Tth, it is possible to set the optimum time threshold Tth according to the utterance speed of each speaker. Further, an utterance speed learning means for learning the utterance speed of the speaker may be provided, and the time threshold value Tth may be set based on the learning result of the utterance speed learning means.
[0061]
In addition, when the number of words to be recognized is small and limited, and the number of mora and the number of syllables of the words are different, the configuration of the voice recognition device 10 is simplified, and only the number of times the speaker hits the contact portion 12a is used. May be referred to for voice recognition. For example, in the case of performing recognition by the number of moras, if there are seven hit inputs, words of 7 moras in the recognized vocabulary dictionary, for example, “route search” are regarded as the recognition result. In the case of performing recognition based on the number of syllables, if there are three hit inputs, a word of three syllables in the recognized vocabulary dictionary, for example, “computer” is used as the recognition result. Furthermore, recognition can also be made based on the number of kanji. If there are four hit inputs, words of four kanji in the recognized vocabulary dictionary, for example, "path search" are used as the recognition result.
[0062]
Further, by providing a switch for notifying a punctuation mark and a reading point to the speech recognition device 10 and setting the point at which this switch is hit as a break of a phrase or a speech input end timing, etc., the end of the speech can be detected more reliably. Can be.
[0063]
Further, in order to reduce the risk that the noise generated when the speaker hits or strokes the contact portion 12a and adversely affects the speech recognition, a configuration is provided in which the noise generated from the contact portion 12a is learned in advance, and the voice recognition is performed. In this case, if the influence is reduced by, for example, the spectral subtraction method, the recognition performance of the speech recognition unit can be further stabilized.
[0064]
For example, when the voice recognition device 10 of the present invention is applied to a stuffed toy, the emotions of the speaker, such as emotions and emotions, can be easily estimated based on the strength and pattern of the tapping of the contact portion 12a. Can also be improved. For example, it is presumed that the speed of tapping the contact portion 12a is high and the intensity is high, indicating anger, and that the speed of tapping of the contact portion 12a is low indicates sadness or loneliness. Communication with the stuffed toy can be facilitated.
[0065]
As described above, according to the voice recognition device 10 of the present embodiment, the voice section determination means 13 is based on the voice signal included in the sound signal output from the sound signal output means 11 and the contact state of the contact means 12. Therefore, the voice section is determined by the continuous or intermittent contact of the contact means 12 due to the natural movement of the uttering speaker, and the voice section of the speaker is determined. Speech recognition can be performed with simple operations that reduce physical and mental burdens.
[0066]
In addition, for example, when the voice recognition device 10 of the present invention is applied to a stuffed toy, actions such as hitting, rubbing, and stroking the stuffed animal do not impose a burden on the speaker, and it is possible to perform voice recognition by these actions. Since it is possible to facilitate communication between the speaker and the stuffed animal, it is possible to make the speaker feel closer to the stuffed animal and have an affinity for the stuffed animal rather than simply touching the stuffed animal. Therefore, especially when the speaker is an elderly person, a child, a lonely person, or the like, the speaker's heart can be healed and playfulness can be stimulated.
[0067]
【The invention's effect】
As described above, according to the present invention, it is possible to provide a voice recognition device capable of performing voice recognition by a simple operation with a reduced physical and mental burden on a speaker.
[Brief description of the drawings]
FIG. 1 is a block diagram of a speech recognition apparatus according to an embodiment of the present invention.
FIG. 2 is a flowchart of each step of the voice recognition device according to the embodiment of the present invention;
FIG. 3A is a diagram showing a power curve in which power values of acoustic signals are plotted.
(B) A diagram showing the determination result of voice characteristics
(C) A diagram showing a contact signal and a non-contact signal
(D) Diagram showing voice section
FIG. 4A is a diagram showing a power curve in which power values of acoustic signals are plotted.
(B) Diagram showing voice section
(C) A diagram schematically illustrating a section in which the standard pattern of the vocabulary matches the recognized result well.
FIG. 5 is a block diagram of a conventional voice recognition device.
[Explanation of symbols]
10 Speech recognition device
11 Acoustic signal output means
11a microphone
11b AD converter
12 contact means
12a Contact part
12b Contact state detector
13 Voice section determination means
13a Power calculator
13b Audio signal determination unit
13c Power threshold setting unit
13d storage unit
13e Voice section judgment unit
14 Voice recognition means
15 Display means
41 Power Curve
42 Minimum hold value
43 1st point
44 2nd point
51 1st standard pattern
52 Second Standard Pattern
52a Recognized second standard pattern
53 Third Standard Pattern
54 4th standard pattern

Claims (5)

Sound signal output means for inputting sound including a speaker's voice and outputting a sound signal; contact means for contacting the speaker when the speaker utters; a sound signal included in the sound signal and the contact A voice section determining means for determining a voice section in which the speaker is emitting the voice based on a contact state of the means; and voice recognition means for recognizing the voice in the voice section. Recognition device. The contact unit detects a contact portion that contacts the speaker and whether or not the speaker is in contact with the contact portion, and indicates a contact operation state in which the speaker is in contact with the contact portion. A contact state detection unit that outputs one of a contact operation state signal and a contact stop state signal indicating a contact stop state in which the speaker has stopped contacting the contact unit. A power calculation unit for calculating the power of the audio signal, a power threshold setting unit for setting a power threshold of the audio signal in accordance with an ambient noise level, and the audio signal near a time when the contact operation state signal is output A point in time when the power of the power exceeds the power threshold is a predetermined time, and the point in time is a voice section start point, and the power of the voice signal exceeds the power threshold in the vicinity of the point in time when the contact stop state signal is output. Speech recognition apparatus according to claim 1, characterized in that a determining speech segment determination unit the speech section by the time the predetermined time has elapsed from the time the speech section end time of turn. The voice section determination means may determine that the voice input when the duration of the contact stop state following the contact operation state is equal to or less than a predetermined threshold is the same as the voice input before the contact stop state. The speech recognition apparatus according to claim 2, wherein the speech section is included in the speech section. The audio section determination unit includes an audio signal determination unit that determines whether the audio signal includes the audio signal. The audio section determination unit uses the audio signal determination unit to convert the audio signal into the audio signal. The voice recognition according to claim 2 or 3, wherein when it is determined that a signal is included, the voice section is determined based on the contact state and the power of the voice signal. apparatus. The voice section determination unit determines a point in time when the power of the voice signal exceeds the power threshold by a predetermined time from a point in time near the time when it is determined that the voice signal is included in the audio signal. As the start time, the voice section is determined as the voice section end time when a predetermined time has elapsed from the time when the power of the voice signal falls below the power threshold near the time when the contact stop state signal is output. The speech recognition device according to claim 4, wherein

Images