JP3118023B2 - Voice section detection method and voice recognition device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice section detection method and a voice recognition apparatus using the voice section detection method, and more particularly to a voice section detection technique of a voice input device in a high noise environment. Speech recognition device below, for example, in a car,
The present invention can be applied to voice recognition devices in factories and homes, and can be applied to noise removal in other voice input devices such as voice synthesis devices and communication devices.

[0002]

2. Description of the Related Art In putting a speech recognition apparatus into practical use, countermeasures against noise from surrounding noise are important issues. In particular, since it is not easy to accurately detect a voice section from a voice on which noise is superimposed, the recognition rate is significantly reduced in such an environment. Japanese Patent Publication No. Sho 63-29754 discloses, for example, a speech section detection method suitable for speech in which noise is superimposed. This means that when two thresholds are used and a section above the high threshold is longer than a certain time, a section above the low threshold is used as a voice section. In such an environment, section detection becomes difficult. Japanese Patent Application Laid-Open No. 58-130395 is an example of a method that has an effect on the time non-stationary noise.
In this method, a voice section is determined by comparing a difference between powers of two inputs of a voice microphone and a noise microphone with a threshold. However, the noise component superimposed on the voice in the voice section is not removed.

Further, as a method of removing a noise component by using two inputs, there is a spectrum subtraction method which has been widely used in the past, but this method has a drawback that it cannot cope with noise having a strong time non-stationary state. is there. Japanese Patent Application Laid-Open No.
196599, JP-A-63-262695, JP-A-1-115798, JP-A-1-239
No. 596. Each of the inventions described in these publications is a kind of an adaptive noise canceling method using two input means of a voice input and a noise input, and can be expressed as follows. X is the spectrum of the audio input
(i) Assuming that the spectrum of the noise input is N (i) (i: represents each frequency band), the ratio k (i) of the noise obtained by the two inputs for each frequency band is k (i) = X (i) / N (i), and the estimated value S (i) of the voice spectrum in the voice section is calculated as S (i) = X (i) −k (i) · N (i). Things. According to this method, assuming that there is only one noise source for a certain band i, the ratio k (i) of noise obtained from two inputs even if the noise level changes.
Does not change, the noise component included in X (i) in the voice section can be estimated by k (i) · N (i). Therefore, the present invention is also effective for noise that is not stationary for some time. However, in this method, if the calculation of the ratio k (i) is performed when the noise is small, the error increases, and if a relatively large noise component is included in the next voice section, it is not possible to estimate a proper voice spectrum. Therefore, there is a drawback that proper noise removal cannot be performed unless noise always exists in all bands.

[0004]

SUMMARY OF THE INVENTION The present invention has been made in view of the above circumstances, and, first, provides a voice section detection method capable of performing appropriate voice section detection even in a time non-stationary noise environment. Secondly, it is an object of the present invention to provide a noise elimination device capable of appropriately performing noise elimination even in a time non-stationary noise environment. It is an object of the present invention to provide a speech recognition device that can obtain a good recognition rate even under a steady noise environment.

[0005]

To achieve the above object, the present invention provides (1)
A first input unit for inputting voice, a first feature amount extraction unit for obtaining a feature amount of a plurality of elements of the first input signal obtained by the first input unit, and a noise; Input means for the second input means and the second input means obtained by the second input means.
A second feature value extracting means for calculating a feature value composed of a plurality of elements of the input signal, a coefficient calculating means for calculating a coefficient for each element from the first feature value and the second feature value, ,
A noise component removing unit that estimates a feature amount of a voice by removing a noise component from the first feature amount using the second feature amount and the coefficient, and at least the noise component estimated by the noise component removing unit; A voice section detecting means for detecting a voice section for each of the above-mentioned elements by using a feature amount of the voice; or (2) a first input means for inputting a voice; A first feature extracting means for obtaining a feature comprising a plurality of elements of the first input signal obtained by the input means, a second input means for inputting noise, and a second
A second feature value extracting means for obtaining a feature value comprising a plurality of elements of the second input signal obtained by the first input means;
Coefficient calculating means for calculating a coefficient for each element from the characteristic amount of the second characteristic amount and the second characteristic amount, and removing a noise component from the first characteristic amount by using the second characteristic amount and the coefficient Noise component removing means for estimating a speech feature amount by performing
A first voice section detecting means for obtaining a voice volume using the voice feature amount estimated by the noise component removing means and detecting a first voice section from the voice volume; In a section obtained by adding a predetermined section before and after the first speech section detected by the speech section detection means, at least the feature amount of the speech estimated by the noise component removal means is used for each of the above elements. And a second voice section detecting means for detecting a second voice section, or
(3) first input means for inputting voice, first feature amount extracting means for obtaining a feature amount comprising a plurality of elements of the first input signal obtained by the first input means, and noise Input means for inputting the first characteristic, second characteristic amount extracting means for obtaining a characteristic amount comprising a plurality of elements of the second input signal obtained by the second input means, and the first characteristic The quantity and the second
Coefficient calculating means for calculating a coefficient for each element from the characteristic amount of the first characteristic amount, and the first characteristic value using the second characteristic amount and the coefficient.
A noise component removing unit for estimating a speech feature amount by removing a noise component from the feature amount of the first speech section and a first speech section using at least the magnitude of the first signal and the magnitude of the second signal. A first voice section detecting means for detecting the noise component, and a section in which predetermined sections are added before and after the first voice section detected by the first voice section detecting means, respectively. And a second voice section detecting means for detecting a second voice section for each of the elements using the feature amount of the voice estimated in the above.
Alternatively, (4) the speech section detection method according to (1), wherein the noise component removing unit holds a coefficient for removing the noise component for each of the elements, and The coefficient is updated based on information on a voice section for each corresponding element in the voice section detection means,
Alternatively, (5) the voice section detection method according to the above (2) or (3), wherein the noise component removing means holds a coefficient for removing a noise component for each of the elements, The coefficient for each element is updated based on the information on the voice section for each corresponding element in the second voice section detection means, or (6) the voice section detection method described in (1) above. Further, the voice section detecting means holds a threshold value for detecting a voice section of each element, and the threshold value of each element corresponds to a voice of each corresponding element in the voice section detecting means. (7) the voice section detection method according to the above (2) or (3), wherein the second voice section detection means further comprises: For detecting the second speech section of Hold, and the threshold value for each element is updated based on the information about the corresponding voice section for each second element, or (8) the above (1) or (2) or (3) Voice section detection method,
The coefficient calculating means is configured to determine whether the first characteristic amount or the second characteristic amount
If each element of the feature amount is large, the value of the corresponding coefficient is set to the value of the ratio of the corresponding element of the first feature amount to the corresponding element of the second feature amount, or close to the value. If each element of the first feature amount or the second feature amount is small, the value of the corresponding coefficient is a predetermined value or a value close to the value, or (9) In the voice section detection method according to (8), the coefficient calculating means may further include: each element X (i) of the first feature quantity; and a corresponding element N of the second feature quantity. For (i), the value of the corresponding coefficient k (i) is k (i) = [X (i) + C] / [N (i) + C] (C = constant> 0), or (10) The voice section detection method according to the above (8), wherein the coefficient calculating means further comprises: each element X (i) of the first feature amount; For the corresponding element N (i) of the quantity, the value of the corresponding coefficient k (i) is given by: k (i) = [X (i) + C ₁ ] / [N (i) + C ₂ ] (C ₁ , C ₂ = constant> 0) or (11) the above (1) to (10)
Wherein the first input means for inputting voice and the second input means for inputting noise are both microphones, or (12) The above (1) to (10)
Wherein the first input means for inputting a voice is a microphone, and the second input means for inputting noise is located near the microphone. A speaker placed in the speaker, and an audio signal reproduced from the speaker as an input, or (13) obtained using the voice section detection method according to any one of (1) to (12). An input pattern generation unit that creates an input pattern of the input voice from the feature amount of the input voice, a standard pattern memory that stores a standard pattern of the voice registered in advance, and performs a recognition process using the input pattern and the standard pattern. And a recognition unit for performing the recognition. Hereinafter, a description will be given based on an example of the present invention.

FIG. 1 is a block diagram for explaining an embodiment of the present invention. In FIG. 1, reference numeral 1 denotes a first microphone, 2 denotes a second microphone, 10 denotes a first feature amount extraction unit, and 20 denotes a first feature amount extraction unit.
Is a second feature amount extraction unit, 30 is a first voice section detection unit,
40 is a coefficient calculator, 50 is a noise component remover, and 60 is a second
Is a voice section detection unit. The first microphone 1 is a microphone for inputting voice (main input). The microphone 1 is placed near the mouth of the speaker, and the main input signal obtained here includes voice and ambient noise. included. The second microphone 2 is a microphone for inputting ambient noise (reference input). The microphone 2 is placed at a position away from the mouth of the speaker, and the reference input signal obtained here includes the surrounding input. It contains only noise and almost no voice.

The first feature quantity extraction unit 10 includes a microphone amplifier 11, a band-pass filter 12 for 15 channels, a rectifier 13, a low-pass filter 14, a multiplexer 15,
A main input signal spectrum X (i) (i =
1, 2, ..., 15). Second feature amount extraction unit 20
Consists of a microphone amplifier 21, a 15-channel band-pass filter 22, a rectifier 23, a low-pass filter 24, a multiplexer 25, and an A / D converter 26, and the spectrum N of the reference input signal obtained by the microphone 2 every fixed frame time. Find (i). The microphone amplifier 11 and the microphone amplifier 21 previously adjust the gain of each input so that the main input signal and the reference input signal have substantially the same level with respect to noise from a distant sound source. The characteristics of the other parts of the first and second feature quantity extraction units are the same. The feature amounts obtained by the first and second feature amount extraction units may be other known feature amounts.

[0008] The first voice section detector 30 determines the approximate voice section by determining whether the difference between the power ΣX (i) of the main input signal and the power ΣN (i) of the reference input signal exceeds a threshold value T _pwr. To detect. The power of each input signal may be obtained by another means, or another method may be used as a general method for detecting a voice section. The threshold value T _pwr is a power X _pwr = ΣX (i) of the main input signal and a power N _pwr = ΣN of the reference input signal in a plurality of frames before the frame in a section other than the approximate voice section.
It is calculated from the average value of (i) ( _{denoted as} Av · X _pwr and Av · N _pwr , respectively) and updated sequentially. That is, T _pwr = a _pwr (Av · X _pwr −Av · N _pwr ) + b _pwr (1) (a _pwr , b _pwr : constant, a _pwr , b _pwr > 0) A power ΣS (i) of the voice is obtained from a feature amount S (i) of the voice from which a noise component is removed by a noise component removing unit 50 described later, and a threshold T _pwr is obtained.
It may be compared with.

[0009] The coefficient calculation unit 40 provides, for each channel i,
The coefficient k (i) is calculated from the spectrum X (i) of the main input signal and the spectrum N (i) of the reference input signal as follows. k (i) = [X (i) + C ₁₎ / (N (i) + C _2]: coefficient (C _1, C _{2 constant, C 1, C 2> 0} ) (2) Equation (2) k ( i) may be calculated from the average value of the spectrum X (i) of the main input signal and the spectrum N (i) of the reference input signal in a plurality of frames before that frame. A number smaller than the number of frames for _{obtaining pwr} and Ti is _preferable .

As is apparent from the equation (2), X
When the values of (i) and N (i) are large, k (i) approaches the value used in the prior art, X (i) / N (i), and conversely,
If the values of X (i) and N (i) are small, k (i) is C ₁ / C
₂ (Constant). Therefore, when the values of X (i) and N (i) are small, the error of k (i) can be reduced by setting the value of C ₁ / C ₂ to an appropriate value according to the system. Here, the values of C ₁ and C ₂ are such that the values of X (i) and N (i) are 8 bits (0 to 2).
When expressed by 55), about 8 to 32 is good. The value of C ₁ / C ₂ is X (i), N which is experimentally measured in advance for noise from a distant sound source or noise from a fixed sound source.
If (i) is at substantially the same level, 1 is sufficient, but if not, X (i) / N (i) is used. Further, C ₁ , C ₂
May be different depending on the channel.
Further, in a system in which sound is mixed into the microphone 2 to some extent, if the value of C ₁ / C ₂ is set to 1 (C ₁ = C ₂ ), a part of the sound component is removed as a noise component. 1
It is better to make the value smaller (C ₁ <C ₂ ). The calculation of Expression (2) is performed in a section where no voice is input, and is sequentially updated. The value of the coefficient k (i) obtained here may be subjected to a smoothing process on the time axis. Further, the relationship between the coefficient k (i) and the spectra X (i), N (i) is not limited to the relationship of Expression (2), and may be an expression using a hyperbolic function or an exponential function, A similar effect can be provided by using a technique.

The noise component removing section 50 provides, for each channel i, a characteristic of a voice obtained by removing a noise component from the spectrum X (i) of the main input signal, the spectrum N (i) of the reference input signal, and the coefficient k (i). The quantity S (i) is calculated as follows. S (i) = X (i) −k (i) · N (i) (3) In the case of a non-voice section, S (i) = 0 may be set. In equation (3), it is also possible to further add or subtract to adjust an error or the like and obtain a more accurate voice feature amount S (i).

The second speech section detection section 60 is a section in which predetermined sections are added before and after the general speech section detected by the first speech section detection section 30, respectively. Depending on whether or not the estimated voice spectrum S (i) exceeds the threshold value Ti, for each channel i,
A voice section for each band is detected. Therefore, S (i) is set when the voice section is for each band, and 0 (zero) otherwise.
Is output to a subsequent speech recognition device or the like.

FIG. 2 shows the relationship between a general voice section (A), a section (B) in which predetermined sections are added before and after the voice section (A), and a voice section (C) for each band for each channel i. FIG. The threshold value Ti is expressed as Ti = a (Av.X (i) -Av.N (i)) + b (4) (a, b: constant, a, b> 0) From the average value (Av.X (i) and Av.N (i), respectively) of the spectrum X (i) of the main input signal and the spectrum N (i) of the reference input signal in a plurality of frames before that frame in a section other than the section. Calculated and updated sequentially.

In this embodiment, when the apparatus is used in an environment where the noise level is not so large, the first voice section detection section 30 directly calculates the second voice without using the rough voice section. It is also possible to determine the voice section of the band by the section detecting section 60 (claim 1). Further, the calculation of the threshold value Ti in the second voice section detection unit 60 and the coefficient calculation unit 4
The calculation of the coefficient k (i) at 0 is not per channel i,
The processing may be performed for each group by grouping some channels.

FIG. 3 is a block diagram showing one embodiment of the present invention for removing noise from a speaker placed near a microphone for inputting voice. In the drawing, the embodiment shown in FIG. 1 is used. The parts having the same functions as in FIG. 1 are denoted by the same reference numerals as in FIG. Thus, this embodiment is the same as the invention of FIG. 1 except that an acoustic signal to be sent to the speaker 2s is input instead of the input from the microphone 2 in FIG.
The microphone amplifiers 11 and 21 are adjusted so that the main input signal and the reference input signal have substantially the same level with respect to the acoustic signal.

FIG. 4 is a block diagram showing an example of a speech recognition device using the noise elimination device of the present invention. The input pattern generation section 70 is input from the speech spectrum obtained by the above-described noise elimination device. A voice input pattern is created, the standard pattern memory 80 stores a voice standard pattern registered in advance, and the recognition unit 90 performs recognition processing on the input pattern and the standard pattern. Although the configurations and operations of the input pattern generation unit 70, the standard pattern memory 80, and the recognition unit 90 are based on the known BTSP speech recognition system, other systems may be used.

[0017]

According to the first aspect of the present invention, a voice signal is input for each band using a characteristic value of a main input signal for inputting voice and a characteristic value obtained from a reference input signal for inputting noise. Since section detection is performed, even if a certain band is a voice section, other bands where no voice component is present are not regarded as a voice section, so that a feature amount of the voice can be accurately extracted, and a time non-stationary high-noise environment is used. In the speech recognition below, a good recognition rate is obtained. According to the second aspect of the present invention, the loudness of the voice is obtained by using the feature amount of the voice estimated by the noise component removing means, a rough voice section is detected from the loudness of the voice, and the rough voice is detected. Since a voice section is detected for each band in a section where a predetermined section is added before and after the section, noise in a section that is somewhat distant from the general voice section is not mistaken for voice, and it is more accurate. Voice section detection. According to the third aspect of the present invention, the loudness of the voice is obtained by using the power of the main input signal and the power of the reference input signal, and the rough voice section is detected from the loudness of the voice. Since the voice section detection is performed for each band in a section where a predetermined section is added before and after the section, even if the noise is as loud as the voice, it will not be mistaken for the voice and it will be more accurate Voice section detection. According to the fourth and fifth aspects of the present invention, since the coefficient for each band for removing the noise component is updated based on the information of the voice section for each band, it is assumed that the noise environment changes during the voice section. Also, in a band that does not include a sound component, the coefficient can be updated to cope with a change in the noise environment, and more accurate voice section detection can be performed. According to the sixth and seventh aspects of the present invention, since the threshold for detecting the voice section for each band is updated based on the information of the voice section for each band, the noise environment changes during the voice section. Even in the case where the voice component is not included, the threshold value can be updated in the band where the voice component is not included, to cope with a change in the noise environment, and more accurate voice section detection can be performed. Claim 8
According to the invention described in the above, as is apparent from the equation (2),
Under non-stationary noise, the coefficient k (i) is determined when the surrounding noise level is low, and when a relatively large noise component is included in the next speech section, the coefficient k (i) is determined. Is close to a predetermined constant and the coefficient k
When the error of (i) is small and the noise level is large, the coefficient k (i) becomes close to the ratio between the main input and the reference input. The spectrum can be estimated. According to the ninth aspect of the present invention, in a system in which the main input and the reference input are substantially at the same level and the reference input does not include voice, the coefficient k (i) when the ambient noise level is small is 1 , The error of the coefficient k (i) is small, and when the noise level is large, the coefficient k (i) is close to the ratio of the main input and the reference input, and further, when the noise level is small to large. Since the coefficient k (i) changes continuously, an appropriate noise component can be removed at any noise level, that is, an appropriate voice spectrum can be estimated. According to the tenth aspect of the present invention, in a system in which the main input and the reference input are not substantially at the same level, or where the reference input is mixed with voice to some extent, the coefficient k ( i) approaches a constant suitable for the system,
There is an effect similar to the effect of the ninth aspect of the present invention. According to the eleventh aspect, when two microphones are used as the main input and the reference input, the above-described effect is obtained. According to the twelfth aspect, the above-described effect is obtained when an audio signal sent to a microphone is used as a main input and an audio signal sent to a speaker is used as a reference input. According to the thirteenth aspect, a good recognition rate can be obtained in voice recognition in a noise environment where time is not stationary.

[Brief description of the drawings]

FIG. 1 is a configuration diagram for explaining an embodiment of the present invention.

FIG. 2 is a time chart for explaining the operation of FIG. 1;

FIG. 3 is a configuration diagram for explaining an embodiment of a voice section detection method according to the present invention.

FIG. 4 is a configuration diagram for explaining an embodiment of a speech recognition device according to the present invention.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... 1st microphone, 2 ... 2nd microphone, 10 ... 1st feature-value extraction part, 20 ... 2nd feature-value extraction part, 30 ... 1st audio | voice area detection part, 40 ... Coefficient calculation part, 50: noise component removing unit, 60: second voice section detecting unit, 70: input pattern generating unit, 80: standard pattern memory, 90: recognition unit.

Continuation of the front page (51) Int.Cl. ⁷ Identification symbol FI // G10L 101: 023 (56) References JP-A-1-200294 (JP, A) JP-A-60-193000 (JP, A) JP JP-A-62-244100 (JP, A) JP-A-62-123499 (JP, A) JP-A-62-129897 (JP, A) JP-A-58-142400 (JP, A) JP-A-3-147000 (JP) JP-A-63-262695 (JP, A) JP-A-55-7749 (JP, A) JP-A-2-272499 (JP, A) JP-A-58-196599 (JP, A) 1-255897 (JP, A) JP-A-2-238493 (JP, A) JP-A-60-101598 (JP, A) JP-A-2-205898 (JP, A) Patent 2897230 (JP, B2) Patent 2807457 (JP, B2) Patent 2856934 (JP, B2) Patent 2539027 (JP, B2) JP-B 7-113834 (JP, B2) JP-B 3-76475 (JP, B2) JP-B 4-49952 (JP, B B2) Tokiko Hei 22-22398 (JP, B2) Jiko 62-37279 ( P, Y2) (58) investigated the field (Int.Cl. ^7, DB name) G10L 15/00 - 17/00 JICST file (JOIS)

Claims (13)

(57) [Claims] A first input unit for inputting a voice; a first feature extraction unit for obtaining a feature comprising a plurality of elements of the first input signal obtained by the first input; ,
A second input unit for inputting noise, a second feature extracting unit for obtaining a feature comprising a plurality of elements of the second input signal obtained by the second input, and the first feature extracting unit; Coefficient calculating means for calculating a coefficient for each element from the characteristic amount and the second characteristic amount; and removing a noise component from the first characteristic amount using the second characteristic amount and the coefficient. Noise component removing means for estimating a feature amount of the speech by using the feature amount of the speech estimated by the noise component removing means at least, and a voice section detecting means for detecting a voice section for each of the elements. A voice section detection method characterized by the following. 2. A first input means for inputting a voice, a first feature quantity extracting means for obtaining a feature quantity comprising a plurality of elements of the first input signal obtained by the first input means, ,
A second input unit for inputting noise, a second feature extracting unit for obtaining a feature comprising a plurality of elements of the second input signal obtained by the second input, and the first feature extracting unit; Coefficient calculating means for calculating a coefficient for each element from the characteristic amount and the second characteristic amount; and removing a noise component from the first characteristic amount using the second characteristic amount and the coefficient. And a noise component estimating means for estimating a feature amount of the voice, and a loudness of the voice is obtained by using the feature amount of the voice estimated by the noise component removing means, and a first voice section is determined from the loudness of the voice. A first voice section detecting means to be detected, and a section in which predetermined sections are added before and after the first voice section detected by the first voice section detecting means, respectively, and at least the noise component removing means Using the estimated speech features, VAD method, characterized in that formed by and a second voice activity detection means for detecting a second audio segment. 3. A first input means for inputting a voice, and a first feature quantity extracting means for obtaining a feature quantity comprising a plurality of elements of the first input signal obtained by the first input means. ,
A second input unit for inputting noise, a second feature extracting unit for obtaining a feature comprising a plurality of elements of the second input signal obtained by the second input, and the first feature extracting unit; Coefficient calculating means for calculating a coefficient for each element from the characteristic amount and the second characteristic amount; and removing a noise component from the first characteristic amount using the second characteristic amount and the coefficient. Noise component removing means for estimating a speech feature amount, and a first speech section detection for detecting a first speech section using at least the magnitude of the first signal and the magnitude of the second signal. Means, and a first speech section detected by the first speech section detection means.
In a section in which a predetermined section is added before and after each of the voice sections, a second voice section is detected for each element by using at least the feature amount of the voice estimated by the noise component removing means. 2. A voice section detection method comprising: a second voice section detection means. 4. The voice section detection method according to claim 1, wherein said noise component removing means holds a coefficient for removing a noise component for each of said elements, and a coefficient for each of said elements. Is a voice section detection method which is updated based on information on a voice section of each corresponding element in the voice section detection means. 5. The voice section detection method according to claim 2, wherein said noise component removing means holds a coefficient for removing a noise component for each of said elements. A voice section detection method, wherein the coefficient for each element is updated based on information on a voice section for each corresponding element in the second voice section detection means. 6. The voice section detection method according to claim 1, wherein said voice section detection means holds a threshold for detecting a voice section for each of said elements, and a threshold for each of said elements. Is a voice section detection method which is updated based on information on a voice section of each corresponding element in the voice section detection means. 7. The voice section detection method according to claim 2, wherein said second voice section detection means further comprises:
A threshold value for detecting a second voice section for each of the elements is held, and the threshold value for each of the elements is updated based on information on a corresponding voice section for each of the second elements. Voice section detection method. 8. The voice section detection method according to claim 1, wherein the coefficient calculating means is configured to determine whether or not each of the first feature amount and the second feature amount is large.
The value of the corresponding coefficient is the value of the ratio between the corresponding element of the first feature and the corresponding element of the second feature, or a value close to the value, and the first feature or the second (2) When each element of the feature amount is small, the value of the corresponding coefficient is set to a predetermined value or a value close to the value. 9. The voice section detection method according to claim 8, wherein said coefficient calculating means further comprises: each element X (i) of said first characteristic amount; and a corresponding element of said second characteristic amount. N
For (i), the value of the corresponding coefficient k (i) is k (i) = [X (i) + C] / [N (i) + C] (C = constant> 0) Voice section detection method. 10. The voice section detection method according to claim 8, wherein said coefficient calculation means further comprises: each element X (i) of said first characteristic amount; and a corresponding element of said second characteristic amount. N
For (i), the value of the corresponding coefficient k (i) is given by: k (i) = [X (i) + C ₁ ] / [N (i) + C ₂ ] (C ₁ , C ₂ = constant> 0 ). 11. The voice section detection method according to claim 1, further comprising: first input means for inputting voice and second input means for inputting noise. Is a voice section detection method, wherein each is a microphone. 12. The voice section detection method according to claim 1, wherein said first input means for inputting a voice is a microphone, and said first input means for inputting a noise. 2. An audio section detection method, wherein the input means is a speaker placed near the microphone, and receives an audio signal reproduced from the speaker. 13. The method according to claim 1, wherein
An input pattern generation unit that creates an input pattern of the voice input from the feature amount of the voice obtained using the voice section detection method described in, a standard pattern memory that stores a pre-registered voice standard pattern, A speech recognition apparatus comprising: a recognition unit that performs a recognition process on the input pattern and the standard pattern.