JP2001142477A - Voiced sound generator and voice recognition device using it - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a voiced sound generator which has a simple constitution and is saved in the trouble of preliminarily learning the voice of a user and can generate a voiced sound from whispering. SOLUTION: Whispering voice data outputted from a whispering voice input part 10 and delayed voice data outputted from a delay device 12 are given to an adder 11, and the adder 11 adds these data to output the additional result. Whispering voice data newly outputted from the whispering voice input part 10 is added to voice data circulating along a closed loop, which returns from the adder 11 to the adder 11 through the delay device 12, each time when circulating once, and thus voiced sound data is generated.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voiced sound forming apparatus and a voice recognition apparatus using the same, and more particularly, to a whispered voice, a voice that has been devoiced during a conversation (hereinafter referred to as a voiceless voice, etc.). , Whispering) and a voiced sound forming device that forms a voiced sound that can be heard as normal voice, and a voice recognition device that uses the voiced sound forming device to improve the whispering voice recognition rate.

[0002]

2. Description of the Related Art In recent years, mobile phones have been rapidly spreading. One of the social problems caused by this is that a person who uses a mobile phone in a train has a loud voice, and the surrounding people are troubled. Also, with the advancement of voice recognition technology, it has become possible to input commands and character information by voice, that is, voice, in addition to keyboard input, in man-machine interfaces such as word processors and personal computers. However,
In an environment where a large number of personal computers are installed, such as in an office, there is a problem that the use of each other's input sound is noisy, and thus the use of the personal computers does not progress.

In order to solve such a problem, Japanese Patent Application Laid-Open No. 7-219567 proposes a device for converting a whisper into a normal voice. Hereinafter, JP-A-7-
A conventional voice conversion device described in Japanese Patent Publication No. 219567 will be described. FIG. 8 is a block diagram illustrating a configuration of the above-described conventional voice conversion apparatus. In FIG. 8, a conventional voice conversion device includes an input voice analysis unit 511 and a whisper voice analysis unit 51.
2, a normal speech analysis unit 513, a mapping function estimation unit 514, a spectrum conversion unit 516, and a speech synthesis unit 516. The operation of the conventional speech converter configured as described above will be described below.

[0004] First, for learning, a whisper (a specific phoneme whispered by the user) and a normal voice normally uttering the whisper (the specific phoneme) are respectively provided by a whisper voice analyzer 512 and a normal voice. Analysis unit 5
13 is input. The whispering voice analyzer 512 and the normal voice analyzer 513 each perform voice analysis on the input voice for each frame, extract spectrum information of the input voice, and use the obtained spectrum information as a mapping function estimator. 514. Mapping function estimator 514
Are whispering voice analyzer 512 and normal voice analyzer 5
13 estimates a mapping function that correlates a pair of spectrum information groups (that is, spectrum information of a whispering voice and spectrum information of a normal voice paired with the whispering voice), and maps the mapping function and the paired spectrum information. The information group is stored as a learning result. The spectrum information can be obtained by a known speech processing technique, for example, by performing a linear prediction analysis (LPC analysis) on the input speech for each frame.

[0005] As described above, first, the whisper and the normal voice that is paired with the whisper are respectively sent to the whisper voice analysis unit 5
12 and a normal speech analysis unit 513 to extract a pair of spectrum information groups, and then input the pair of spectrum information groups to a mapping function estimation unit 514 to estimate a mapping function that associates them with each other. I do. The obtained pair of spectral information groups and the mapping function that associates them,
It is stored in the mapping function estimating unit 514. By learning a large number of pairs of whispering voices and normal voices that are paired with each other in advance, the mapping function estimating unit 514 includes:
A database of learning results regarding a mapping that associates a whisper with a normal voice is constructed.

[0006] When the learning is completed as described above, the whispering voice can be converted into the normal voice.
That is, when a whisper to be converted is input to the input voice analysis unit 511, the whisper is subjected to the same voice analysis as performed by the whisper voice analysis unit 512, and spectrum information is extracted. The spectrum information extracted by the input voice analysis unit 511 is provided to a spectrum conversion unit 515. The spectrum conversion unit 515 maps the spectrum information provided from the input speech analysis unit 511 using the mapping function specified by the mapping function estimation unit 514 (that is, the database relating to the mapping stored in the mapping function estimation unit 514). , The whispering spectrum information given from the input speech analysis unit 511 is
It is converted to the spectrum information of normal voice that is paired with the whisper.) The spectrum information of the normal voice thus obtained is input to the voice synthesizer 516. Voice synthesis unit 51
Reference numeral 6 synthesizes a speech having a spectrum equivalent to a normal speech based on the input spectrum information, and outputs the resultant synthesized speech.

Next, a conventional speech recognition apparatus will be described. FIG. 9 is a block diagram illustrating an example of a configuration of a conventional voice recognition device. Such a speech recognition device is described in, for example, Japanese Patent Application Laid-Open No. 60-53998. In FIG. 9, a conventional speech recognition apparatus includes an input speech analysis unit 523, a similarity calculation unit 524, a phoneme coefficient memory 52
5, a determination unit 526, and a word dictionary memory 527. The conventional speech recognition device configured as described above will be described below.

First, a voice is input to an input voice analysis unit 523. The input speech analysis unit 523 performs a linear prediction analysis (LPC analysis) on the input speech sequentially for each frame, similarly to the input speech analysis unit 511 of the conventional speech conversion apparatus described above, thereby obtaining spectrum information from the speech. Is extracted and output to the similarity calculation unit 524. Phoneme memory 5
25 stores in advance a plurality of phoneme names and spectrum information corresponding to each phoneme in association with each other. The similarity calculation unit 524 calculates the distance between each piece of spectrum information stored in the phoneme memory 525 and the spectrum information from the input speech analysis unit 523, and the calculated result has the shortest distance, that is, spectrum information. , Out of the plurality of pieces of spectrum information stored in the phoneme memory 525, the one that is most similar to the spectrum information from the input speech analysis unit 523 is selected. Then, the phoneme name corresponding to the selected spectrum information is stored in the phoneme memory 5.
25, and output. Thus, the similarity calculator 5
24 sequentially outputs phoneme names, and the series of phoneme names is given to the determination unit 526.

The word dictionary memory 527 stores a plurality of words and phoneme sequences corresponding to the respective words in association with each other. The determination unit 526 compares and compares a phoneme sequence composed of a series of phoneme names provided from the similarity calculation unit 524 with each phoneme sequence stored in the word dictionary memory 527, and determines the phoneme having the highest similarity. Select a series. Then, a word corresponding to the selected phoneme sequence is read from the word dictionary memory 527. The word read in this way is output as a recognition result of the input voice.

[0010]

In the conventional speech recognition apparatus as described above, when the input speech is analyzed to extract spectrum information, the input speech is directly subjected to linear prediction analysis (LPC analysis). If the voice is a whisper, the recognition rate is low. A whisper has a lower volume level, especially in a vowel portion, than a normal voice, and is called S / N.
This is because the ratio (signal-to-noise level ratio) is not good and the spectrum of the vowel sound is difficult to stabilize during utterance. Conventionally,
In order to recognize a whisper, a conventional voice converter for converting a whisper into a normal voice as described above is used in a stage preceding the conventional voice recognizer. That is, the whisper is converted into normal voice through a conventional voice conversion device, and then input to a conventional voice recognition device.

However, in the above-mentioned conventional voice conversion apparatus, in order to estimate a mapping function between a whisper and a normal voice, a large number of pairs of a whisper and a normal voice paired with the whisper are input before the apparatus is used. There is a need to. That is, it takes time to learn the user's voice in advance. In addition, since a circuit and a memory for learning are required, the configuration of the apparatus is complicated. In addition, when learning is performed in advance as described above, if the user changes, there is a possibility that voice conversion cannot be performed correctly. This is because voice characteristics differ from person to person. Learning many users' voices in advance requires a lot of trouble and also requires a large-capacity memory.

SUMMARY OF THE INVENTION Therefore, an object of the present invention is to provide a voiced sound forming apparatus capable of forming a voiced sound from a whispered voice with a simple configuration and without having to learn the user's voice in advance, and a use thereof. And a voice recognition device.

[0013]

Means for Solving the Problems and Effects of the Invention The first invention is based on a whisper having a spectrum in which a level difference between an amplitude level in a formant and an amplitude level in a band other than the formant is relatively small. A voiced sound forming device for forming a voiced sound having a spectrum in which the level difference is relatively large, wherein a whispered voice is input and a voice input unit that outputs a whispered voice signal corresponding to the whispered voice, a voice signal A whisper audio signal output from the audio input unit and a delayed audio signal output from the delay unit, and the whisper audio signal and the delayed audio signal An adder for adding and outputting is provided. The audio signal output from the adder is input to the delay unit. The whisper audio signal newly output from the audio input unit is added to the audio signal circulating through the closed loop that returns to the adder, so that the voiced sound and the corresponding voiced sound signal are added. It is characterized by being formed.

According to the first aspect of the present invention, the whispered voice signal is input to a closed loop that returns from the adder to the adder via the delay unit and is circulated (that is, the whispered voice signal is transmitted to the adder and the delay unit). ) Can form a voiced sound. For this reason, it is possible to provide a voiced sound forming device having a simple configuration that does not require the trouble of performing learning in advance as in a conventional voice conversion device and is easy to incorporate into a mobile phone or the like.

The second invention according to the first invention further comprises an output terminal for outputting a voice signal output from the adder to the outside as a voiced sound signal corresponding to a voiced sound.

According to a third aspect, in the first aspect, an output terminal for outputting an audio signal output from the delay unit to the outside as a voiced sound signal corresponding to a voiced sound is further provided.

According to the second or third aspect, a voiced sound signal formed while circulating through the closed loop can be output to the outside. When the second invention is compared with the third invention, the delay of the output timing of the voiced sound signal is smaller in the second invention.

In a fourth aspect based on the first aspect, a plurality of delay time lengths are prepared in advance, and any one of the plurality of delay time lengths is selected and the delay time length is selected. The delay unit further includes a delay length selection unit configured to set the delay unit, wherein the delay unit delays the audio signal output from the adder by a delay time length set by the delay length selection unit.

According to the fourth aspect, the delay length selection unit prepares a plurality of delay time lengths in advance, selects one of them, and sends the selection result to the delay unit. Since the pitch is set for the voiced sound to be formed, the pitch of the voiced sound to be formed can be selected from options prepared in advance. As a result, the frequency of the voiced sound to be formed can be properly used. As an example of proper use,
When the user is a male, a voiced sound having a relatively low pitch is formed, and when the user is a female, a voiced sound having a relatively high pitch is formed.

In a fifth aspect based on the first aspect, an arbitrary pitch is instructed and inputted, and a delay time required to form a voiced sound having the pitch is calculated, and the delay time is calculated. Is further provided for the delay unit, and the delay unit delays the audio signal output from the adder by the delay time set by the delay length input unit.

According to the fifth aspect, the delay length input section receives the instruction input relating to the pitch, calculates the delay time extension, and sets the calculation result to the delay device. The pitch of the voiced sound to be formed can be arbitrarily specified. As a result, a voiced sound whose pitch changes with time can be formed. For example, a voiced sound with a melody (scale) and a voiced sound with intonation (intonation) can be formed.

According to a sixth aspect of the present invention, a whisper having a spectrum in which the level difference between the amplitude level in the formant and the amplitude level in a band other than the formant is relatively small, the level difference is relatively large. A voiced sound forming device for forming a voiced sound having a spectrum, wherein a whispered voice is input and a whispered voice signal corresponding to the whispered voice is output, and a discrete output peak is provided at a constant frequency interval. And an output terminal for externally outputting a voiced sound signal formed by transmitting a whispered audio signal from the audio input unit through the comb filter unit. It has.

According to the sixth aspect, a voiced sound signal can be formed only by passing a whispered audio signal through the comb filter. This is a whispered audio signal,
In the band other than the formant, the spectrum fluctuates relatively randomly, so that there are few effective components passing through the comb filter.As a result, the amplitude level in the band other than the formant is smaller than the amplitude level in the formant. It is caused by lowering. For this reason, it is possible to provide a voiced sound forming device having a simple configuration that does not require the trouble of performing learning in advance as in a conventional voice conversion device and is easy to incorporate into a mobile phone or the like.

The seventh invention is characterized in that, in the sixth invention, the fixed frequency interval is determined in relation to the pitch of a voiced sound to be formed.

According to the seventh aspect, a voiced sound having a desired pitch can be formed by using the comb filter having an appropriate output characteristic (frequency interval between output peaks). . Further, by using a comb filter unit whose output characteristics cannot be changed, it is possible to temporally change the pitch of a voiced sound to be formed.

According to an eighth aspect, a whisper having a spectrum in which the level difference between the amplitude level in the formant and the amplitude level in a band other than the formant is relatively small, the level difference is relatively large. A voiced sound forming device for forming a voiced sound having a spectrum, wherein a whispered voice is input, and a whispered voice signal output from a voice input unit that outputs a whispered voice signal corresponding to the whispered voice and a whispered voice signal output from the voice input unit A voiced sound forming unit that forms a voiced sound signal corresponding to the voiced sound, the voiced sound forming unit delays the whispered voice signal output from the voice input unit, and outputs the delayed whispered voice signal to the voice input unit. Is added to the whispered audio signal output from the By going further adding the audio signal whisper is characterized by forming a voiced sound signal corresponding to the voiced.

According to the eighth aspect, a voiced sound signal can be formed from a whispered voice signal. For this reason,
It is possible to provide a voiced sound forming apparatus having a simple configuration, which does not require the trouble of performing learning in advance like a conventional voice conversion apparatus and is easy to incorporate into a mobile phone or the like.

A ninth invention is a speech recognition apparatus for recognizing a whisper having a spectrum in which a level difference between an amplitude level in a formant and an amplitude level in a band other than the formant is relatively small. A voice input unit for inputting and outputting a whisper voice signal corresponding to the whisper voice; a voiced sound signal corresponding to a voiced sound having a spectrum having a relatively large level difference from a whisper voice signal output from the voice input unit And a voice recognition unit that recognizes a whisper based on a voiced sound signal output from the voiced sound generation unit.

According to the ninth aspect, the whispered voice signal to be recognized is first converted into a voiced sound signal through the voiced sound forming unit, and the voiced sound signal is input to the voice recognition unit for recognition. Therefore, it is possible to provide a speech recognition device with an improved whispering speech recognition rate.

In a tenth aspect based on the ninth aspect, the voiced sound forming section delays the whispered voice signal output from the voice input section, and outputs the delayed whispered voice signal from the voice input section. The voiced sound signal corresponding to the voiced sound is formed by adding the whispered voice signal and further adding the whispered voice signal newly output from the voice input unit while sequentially delaying the voice signal after the addition. It is characterized by:

According to the tenth aspect, a voiced sound signal can be formed from a whispered voice signal. For this reason, the configuration of the voice recognition device is simplified as compared with the case where a conventional voice conversion device that converts a whisper into a voiced sound with reference to the learning result is used in a stage preceding the voice recognition unit.

[0032]

Embodiments of the present invention will be described below with reference to the drawings. (First Embodiment) FIG. 1 is a block diagram showing a configuration of a voiced sound forming apparatus according to a first embodiment of the present invention. In FIG. 1, the voiced sound forming device according to the first embodiment includes:
Whispering voice input unit 10, adder 11, and delay unit 1
2 is provided. The adder 11 and the delay unit 12 constitute a comb filter unit (not shown).

The whispering voice input unit 10 is constituted by a microphone, an amplifier, an analog-to-digital converter, etc., and converts the input whispering voice into whispering voice data and outputs it. The adder 11 is a whispering voice input unit 1
The output data of 0 and the output data of the delay unit 12 are added together and output. The delay unit 12 delays the output data of the adder 11 by a predetermined time (this is T milliseconds). That is, it receives the output data of the adder 11, holds the data for a predetermined delay time length (= T milliseconds), and outputs the data.

The operation of the voiced sound forming apparatus configured as described above will be described below. When the user makes a whisper, the whisper is sent to the whisper sound input unit 1.
Input to 0. The whisper voice input unit 10 converts the input whisper voice into whisper voice data. The whispered voice data output from the whispered voice input unit 10 is input to the delay unit 12 through the adder 11, where the whispered voice data is held for a predetermined delay time length (T milliseconds). Feedback will be given. Adder 1
1, the whisper voice data newly output from the whisper voice input unit 10 and the data fed back as described above (that is, the whisper voice input unit 1).
Whispered voice data output 0 to T milliseconds before).

That is, the whispered voice data output from the whispered voice input unit 10 is added to the adder 11 and the delay unit 12.
The signal is input to the comb filter unit composed of the following formulas and passes through a closed loop passing through the adder 11 and the delay unit 12. Then, the whisper data that is newly output from the whisper audio input unit 10 is sequentially added to the whisper data that circulates in the closed loop,
As a result, a voiced sound signal is gradually formed. At this time, if the output data of the adder 11 or the delay unit 12 is branched and extracted, a voiced sound signal formed from the whispered audio signal is obtained.

Next, in the above comb filter section,
The principle of forming a voiced sound from a whisper will be described in more detail. FIG. 2 shows the adder 11 and the delay unit 12 of FIG.
FIG. 4 is a diagram schematically illustrating a principle in which a voiced sound is formed from whispering voices in a comb-shaped filter unit made of. FIG.
(A) represents the frequency-amplitude spectrum of the whisper, (b) represents the frequency-amplitude spectrum of the comb filter, and (c) represents the frequency-amplitude spectrum of the voiced sound formed by the comb filter.

Here, the spectrum of the normal voice and the spectrum (output characteristics) of the conventional comb filter will be described with some supplementary explanations. Each vowel of a voice has a characteristic frequency component, and the band of the characteristic frequency component is generally called a formant. The spectrum of normal speech is a continuous spectrum and has a prominent peak in this formant. On the other hand, a conventional comb filter including a delay unit and an adder is used, for example, in a television receiving circuit to separate a luminance signal and a chrominance signal. It has a spectrum like a tooth.

Referring again to FIG. 2, the frequency amplitude spectrum of the whisper is a continuous spectrum as shown in FIG. 2A and has formants (indicated by F1 and F2 in the figure).
At the peak. However, unlike the spectrum of the normal voice, the peak appearing in the formant is not so remarkable (that is, the difference between the peak level of the formant and the level of the band other than the formant is relatively small as compared with the case of the normal voice).

On the other hand, in the spectrum of the comb filter section composed of the adder 11 and the delay unit 12, discrete peaks appear at regular frequency intervals as shown in FIG. However, the frequency interval of each peak depends on the delay unit 12
F [H determined from the delay time length (= T milliseconds)
z] (F is the reciprocal of T).

When a whisper having a spectrum as shown in FIG. 2A is input to a comb filter having an output characteristic as shown in FIG. 2B, it appears in a formant as the whisper circulates in the comb filter. The peaks become increasingly pronounced. As a result, a voiced sound having a pitch F [Hz] having a spectrum in which a remarkable peak appears in the formant as shown in (c) is formed. This is because in the case of a whisper, the spectrum fluctuates relatively randomly in bands other than formants,
This is because less effective components pass through the comb filter unit, and as a result, the amplitude level in a band other than the formant is lower than the amplitude level in the formant.

Next, the delay unit 12 will be described in detail. The delay unit 12 is constituted by a so-called shift register or random access memory. The sampling frequency of the entire voiced sound forming apparatus according to the present embodiment is, for example, 10
In the case of [kHz], when the delay time length T is 10 milliseconds, the pitch F of the voiced sound formed is 100 [Hz]. At this time, the memory used for the delay unit 12 is 100
(Words). Assuming that T is 10.1 milliseconds, 101 memories (words) are used.

If T is 10.05 milliseconds, the delay time length T cannot be expressed by an integer number of memories. In this case, however, the data stored in the 100th memory and 101 The interpolation may be performed using the data stored in the memory of the number. Alternatively, the phase delay amount of the all-pass filter may be controlled. It is a known technique to realize a desired amount of delay even when the number of required memories is not only an integer but also a fraction.

FIG. 3 is a block diagram showing an example of the configuration of the delay unit 12 shown in FIG. 1 (when a random access memory is used). 3, the delay unit 12 includes a distributor 40, an address generator 41, a writer 42, a delay memory 50 using a random access memory, a first reader 52, a second reader 53, an adder 60, 63 and multipliers 61 and 62.

The above sampling frequency is 10 [kHz]
z], when the predetermined delay time length T is 10.05 ms, the delay length L required as memory data is 10
0.5. When the voiced sound forming operation is started, the distributor 40 outputs the integer part M (= 100) of the delay length L to the address generator 41, and outputs the decimal part α of the delay length L to the multiplier 61, and (1-α) as a delay length complement multiplier 6
Output to 2.

The address generator 41 supplies the write address WP (= 0.1) every sampling period (= 0.1 millisecond).
The address i) and the read address RP (= address i−M) are output. The output write address WP is provided to the writer 42, and the read address RP is provided to the first reader 52 and the adder 60. In the adder 60, the read address RP
, And the result of the addition (RP + 1) is given to the second reader 53.

The writer 42 writes the output from the adder 11 in FIG. 1 to the delay memory 50 according to the given write address WP. At the same time, the first reader 52 reads the data stored at the given read address RP from the delay memory 50 and
On the other hand, the second reader 53 reads the data stored at the given read address (RP + 1) from the delay memory 50 and outputs the data to the multiplier 62.

The multiplier 61 multiplies the data input by the first readout unit 52 by the delay length fraction α input by the distributor 40, while the multiplier 62
Is multiplied by the fractional complement (1-α) input by the distributor 40 to the data input by the readout unit 53. The adder 63 calculates the multiplication result of the multiplier 61 and the multiplier 6
The result of multiplication by 2 is added to each other. Then, the addition result becomes the output of the delay unit 12 in FIG. Thus, the delay length integer part M (= 100) and the delay length integer part M + 1 (= 10
Using the data of 1), an interpolation calculation based on the delay length fractional part α is performed.

The address generator 41 increases the write address WP and the read address RP by one each for each sampling period (= 0.1 millisecond). Time length T (= 1
The data corresponding to the data input by the adder 11 before (0.05 msec) is output. In addition,
Normally, the address generator 41 outputs the write address WP
When the value of the read address RP reaches a predetermined value, those address values are reset. Thus, by using the delay memory 50 as a so-called ring memory, a finite storage amount of the delay memory 50 can be effectively used.

As described above, according to the present embodiment, the whispering voice input to the whispering voice input unit 10 is passed through the comb filter unit composed of the adder 11 and the delay unit 12 to provide a voiced sound. Can be formed. For this reason, it is possible to provide a voiced sound forming device having a simple configuration that does not require the trouble of performing learning in advance as in a conventional voice conversion device and is easy to incorporate into a mobile phone or the like.

The voiced sound formed while circulating through the above comb filter is obtained by extracting the output of the adder 11 or the output of the delay unit 12. However, extracting the output of the adder 11 causes a smaller delay in the output timing of the voiced sound. Therefore, when the delay in the output timing of the voiced sound becomes a problem, the output of the adder 11 is extracted and the voiced sound is output. It is desirable to output.

(Second Embodiment) FIG. 4 shows a second embodiment of the present invention.
It is a block diagram showing the structure of the voiced sound formation apparatus which concerns on embodiment. In FIG. 4, the voiced sound forming apparatus according to the second embodiment includes a whispered voice input unit 10, an adder 11, a delay unit 12, and a delay length selection unit 13. Upon receiving the delay length selection instruction, the delay length selection unit 13 selects one of a plurality of predetermined delay lengths (here, two delay lengths), and determines the delay length as a delay unit. 13 is set. The whispering voice input unit 10, adder 11, and delay unit 12 are the same as those in FIG.

The delay length selection unit 13 has a plurality of options (here, two of T1 and T2) as the delay time length T.
Are prepared in advance. Here, the delay length selection unit 13
Is realized by a switch (not shown) for switching between T1 and T2, and the user operates this switch to give an instruction to select one of T1 and T2. In addition, the delay length selection unit 13 has a memory (not shown), and stores a delay length L corresponding to each delay time length (here, L1 corresponding to T1, L1 corresponding to T1).
And L2 corresponding to T2) are stored in advance. Note that the delay length L can be determined from the delay time length T, the sampling frequency of the entire voiced sound forming apparatus, and the pitch of the voiced sound to be formed.

When the selection of the delay length is instructed, the delay length selection unit 13 first selects T1 (= 10 milliseconds) or T2 (= 5 milliseconds) as the delay time length T in response to the instruction. Then, the selection result and the delay length L corresponding to the selection result are read out from the above-mentioned memory and output to the delay unit 12. Thus, the delay unit 12 adds the delay time length corresponding to the set delay length L to the data input by the adder 11 and outputs the data (that is, the delay time corresponding to the set delay length L) After holding the data input by the adder 11 by the length, the data is output to the adder 11 again)
Performs the operation.

The operation of the voiced sound forming apparatus after setting the delay length is as follows, except that the delay unit 12 operates based on the delay length L selectively set by the delay amount selection unit 13.
This is the same as in the first embodiment. That is, the whispered voice data output from the whispered voice input unit 10 is input to the comb filter unit configured by the adder 11 and the delay unit 12 and circulates there. Then, every time one round, whisper audio data newly output from the whisper audio input unit 10 is added, thereby forming a voiced sound. In the voiced sound thus formed, T2 of 100 [Hz] when T1 is selected as the pitch corresponding to the delay length selectively set through the delay length selection unit 13, that is, T2 is selected. 2 if
It has a pitch of 00 [Hz].

As described above, according to the present embodiment, the whispered voice input to the whispered voice input unit 10 is simply passed through the comb filter unit including the adder 11 and the delay unit 12 to provide a voiced sound. Can be formed. For this reason, it is possible to provide a voiced sound forming device having a simple configuration that does not require the trouble of performing learning in advance as in a conventional voice conversion device and is easy to incorporate into a mobile phone or the like.

Further, the delay length selection section 13 stores a plurality of delay lengths in advance, selects any one of them, and sets the selection result to the delay device 12. Therefore, the pitch of the voiced sound to be formed can be selected from options prepared in advance. As an example, when the user is a male, a voiced sound of 100 [Hz] is formed, and when the user is a female, a voiced sound of 200 [Hz] is formed.

In this embodiment, two types of voiced sounds are selectively used, but more types of pitches can be used properly by increasing the number of options.

(Third Embodiment) FIG. 5 shows a third embodiment of the present invention.
It is a block diagram showing the structure of the voiced sound formation apparatus which concerns on embodiment. In FIG. 5, the voiced sound forming apparatus according to the third embodiment includes a whispered voice input unit 10, an adder 11, a delay unit 22, and a delay length input unit 23. The delay length input unit 23 receives an instruction about the pitch of the voiced sound to be formed, calculates a delay length necessary to obtain a voiced sound having the specified pitch, and outputs the calculation result to the delay unit. Set to 22.

The function of the delay length input section 23 can be realized by using, for example, a keyboard of an electronic musical instrument, a microprocessor, and a program memory. That is, the program memory stores a program for calculating the delay length according to the pitch. When the user instructs the pitch through the keyboard, the microprocessor operates according to this program to calculate the delay length.

The delay unit 22 receives the output data of the adder 11, holds the data for the delay time set by the delay length input unit 23, and outputs the data (the difference from the delay unit 12 in FIG. 1). Will be described later). The whispering voice input unit 10 and the adder 11 are the same as those in FIG.

The operation of the voiced sound forming apparatus configured as described above will be described below. When an instruction on the pitch of the voiced sound to be output is input to the delay length input unit 23, the delay length input unit 23 outputs the specified pitch (= F [H
z]), the delay length required to form a voiced sound having Such a delay length L can be calculated by dividing the sampling frequency of the entire voiced sound forming apparatus by the above pitch F [Hz].

The delay length input unit 23 sets the delay length L calculated as described above to the delay unit 22. Thereby, the delay unit 22 adds the delay time length corresponding to the set delay length L to the data input by the adder 11 and outputs the data (that is, the delay time corresponding to the set delay length L) After the data input by the adder 11 is held by the length, the data is output to the adder 11 again).

The operation of the voiced sound forming apparatus after setting the delay length is the same as that of the first embodiment except that the delay unit 22 operates based on the delay length L arbitrarily set by the delay length input unit 23. This is the same as the embodiment. That is, the whispered voice data output from the whispered voice input unit 10 is input to a comb filter unit configured by the adder 11 and the delay unit 22, and circulates there. Then, every time one round, new whispered voice data is sequentially added from the whispered voice input unit 10, thereby forming a voiced sound. The voiced sound thus formed is input to the delay length input unit 2
3 has a pitch corresponding to the arbitrarily set delay length.

In order to realize a voiced sound having a desired pitch, the delay unit 22 is configured as follows.
FIG. 6 is a block diagram showing an example of the configuration of the delay unit 22 shown in FIG. 5 (when a random access memory is used). 6, the delay unit 22 includes a distributor 40, an address generator 41, a writer 42, a delay memory 54 using a random access memory, and a first reader 5
2, a second readout unit 53, adders 60 and 63, and multipliers 61 and 62.

That is, the delay unit 22 shown in FIG. 6 is the same as the delay unit 12 shown in FIG. 5 except that a delay memory 54 is provided instead of the delay memory 50. The delay memory 50 and the delay memory 54 have different minimum memory areas. In the delay memory 50, the address (i
−M−1) to memory area (M + 1) from address i
Is secured at a minimum, but in the delay memory 54, a memory area (LL) from the address (i-N) to the address i is secured at a minimum.

The memory area LL is provided in the delay length input section 2
3 is obtained by dividing the sampling frequency of the entire apparatus by the minimum pitch FL, with the lowest pitch that can be instructed to be input to FL. For example, the lowest pitch FL is 50 [H
z], the sampling frequency of the entire apparatus is 10 [kHz]
, The memory area LL becomes 200. In the delay unit 22, during the voiced sound forming operation, the memory area LL is used.
Is always secured as a minimum necessary memory area. As a result, even when the pitch of the voiced sound to be formed is changed during the voiced sound formation operation, the data input by the adder 11 is held for a delay time length corresponding to the changed pitch. Will be able to

As described above, according to the present embodiment, the whispered voice input to the whispered voice input unit 10 is passed through the comb filter unit composed of the adder 11 and the delay unit 22 to provide a voiced sound. Can be formed. For this reason, it is possible to provide a voiced sound forming device having a simple configuration that does not require the trouble of performing learning in advance as in a conventional voice conversion device and is easy to incorporate into a mobile phone or the like.

Further, the delay length input section 23 calculates the delay length in response to the instruction input relating to the pitch and sets the calculation result to the delay unit 22, so that the sound of the voiced sound to be formed is formed. The height can be specified arbitrarily. As an example, it is possible to form a voiced sound whose pitch changes with time, that is, a voiced sound with a melody (scale) and a voiced sound with intonation (intonation).

(Fourth Embodiment) FIG. 7 shows a fourth embodiment of the present invention.
FIG. 2 is a block diagram illustrating a configuration of a speech recognition device according to the embodiment. 7, the voice recognition device according to the fourth embodiment includes a voiced sound forming unit 1 and a voice recognition unit 2. The voiced sound forming unit 1 has the same configuration as any of the voiced sound forming devices according to the first to third embodiments. The voice recognition unit 2 has, for example, the same configuration as the conventional voice recognition device shown in FIG.

A whispered voice is input to the voiced sound forming device 1, and the voiced sound forming unit 1 forms a voiced sound from the input whispered voice by the same operation as in the first to third embodiments. The voiced sound thus formed has a stable vowel spectrum and a large S / N ratio as compared to a whispered voice. The voice recognition unit 2 receives the voiced sound formed by the voiced sound generation unit 1, and the voice recognition unit 2 recognizes the input voice. At this time, the recognition operation performed by the voice recognition unit 2 is the same as the conventional operation, but a higher recognition rate can be obtained because the input voice is not a whispered voice but a voiced sound formed from the whispered voice as described above. .

The voiced sound forming section 1 outputs a delay length L generated in the process of forming a voiced sound, in addition to the formed voiced sound, and outputs the voiced sound and the delay length L. You may make it input into the speech recognition part 2. Thereby, the input voice analysis unit 523 of the voice recognition unit 2 (FIG. 9)
Is performed for each frame of a size synchronized with the pitch of the input voice, and as a result,
Analysis errors are reduced.

As described above, according to the present embodiment, the whisper to be recognized (whispered voice signal) is first converted into a voiced sound (voiced sound signal) through the voiced sound forming unit 1 and the voiced sound (voiced sound signal) is converted. (Voice signal) is input to the voice recognition unit 2 for recognition, so that it is possible to provide a voice recognition device with an improved whispering voice recognition rate. Further, if such a voice recognition device is incorporated, it is possible to realize a voice instruction input device that does not generate noise around when used in an office where many people are active.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of a voiced sound generation device according to a first embodiment of the present invention.

FIG. 2 is a diagram schematically illustrating a principle that a voiced sound is formed from a whisper in a comb filter unit including an adder 11 and a delay unit 12 in FIG.

FIG. 3 is a block diagram illustrating an example of a configuration of a delay unit 12 illustrated in FIG. 1 (when a random access memory is used).

FIG. 4 is a block diagram illustrating a configuration of a voiced sound generation device according to a second embodiment of the present invention.

FIG. 5 is a block diagram illustrating a configuration of a voiced sound forming device according to a third embodiment of the present invention.

6 is a block diagram illustrating an example of a configuration of the delay unit 22 illustrated in FIG. 5 (when a random access memory is used).

FIG. 7 is a block diagram illustrating a configuration of a speech recognition device according to a fourth embodiment of the present invention.

FIG. 8 is a block diagram illustrating an example of a configuration of a conventional voice conversion device.

FIG. 9 is a block diagram illustrating an example of a configuration of a conventional voice recognition device.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 ... Voiced sound formation part 2 ... Voice recognition part 10 ... Whispering voice input part 11 ... Adder 12, 22 ... Delayer 13 ... Delay length selection part 23 ... Delay length input part

Claims (10)

[Claims] 1. A whisper having a spectrum in which a level difference between an amplitude level in a formant and an amplitude level in a band other than a formant is relatively small, and having a spectrum in which the level difference is relatively large. A voiced sound forming device for forming a vocal sound, wherein the whispered voice is input, and a voice input unit that outputs a whispered voice signal corresponding to the whispered voice, a delay unit for delaying a voice signal, and the voice input unit A whispered voice signal output from the delay device, a delayed voice signal output from the delay unit is input, the whispered voice signal, the delayed voice signal and an adder that adds and outputs each other, An audio signal output from the adder is input to the delay unit, and the adder passes through the delay unit from the adder to the adder. A whispered voice signal newly output from the voice input unit is added to the voice signal circulating through the returning closed loop every time the loop is completed, thereby forming a voiced sound signal corresponding to the voiced sound. A voiced sound forming device, characterized in that: 2. The voiced sound forming apparatus according to claim 1, further comprising an output terminal for outputting a voice signal output from the adder as a voiced sound signal corresponding to the voiced sound to the outside. 3. The voiced sound forming device according to claim 1, further comprising an output terminal for outputting a voice signal output from the delay unit as a voiced sound signal corresponding to the voiced sound to the outside. 4. A delay length in which a plurality of delay time lengths are prepared in advance, and one of the plurality of delay time lengths is selected and the delay time length is set for the delay device. The delay unit further includes a selection unit, wherein the delay unit outputs the audio signal output from the adder,
The voiced sound forming apparatus according to claim 1, wherein the voiced sound forming apparatus is delayed by a delay time length set by the delay length selecting unit. 5. A delay for inputting an arbitrary pitch, calculating a delay time required to form a voiced sound having the pitch, and setting the delay time to the delay unit. The delay unit further includes a long input unit, wherein the delay unit outputs the audio signal output from the adder,
The voiced sound forming apparatus according to claim 1, wherein the voiced sound forming apparatus delays by a delay time length set by the delay length input unit. 6. A whisper having a spectrum in which a level difference between an amplitude level in a formant and an amplitude level in a band other than a formant is relatively small, and having a spectrum in which the level difference is relatively large. A voiced sound forming device for forming a vocal sound, wherein the whispered voice is input, and a voice input unit that outputs a whispered voice signal corresponding to the whispered voice, wherein output peaks appear discretely at regular frequency intervals. Comb-shaped filter unit having an output characteristic as described above, and an output terminal for externally outputting a voiced sound signal formed by transmitting a whispered sound signal from the sound input unit through the comb-shaped filter unit A voiced sound forming device comprising: 7. The voiced sound forming apparatus according to claim 6, wherein the predetermined frequency interval is determined in relation to a pitch of a voiced sound to be formed. 8. A whisper having a spectrum in which a level difference between an amplitude level in a formant and an amplitude level in a band other than a formant is relatively small, and having a spectrum in which the level difference is relatively large. A voiced sound forming device for forming a voice sound, wherein the whispered voice is input, a voice input unit that outputs a whispered voice signal corresponding to the whispered voice, and a whispered voice signal output from the voice input unit, A voiced sound forming unit that forms a voiced sound signal corresponding to the voiced sound, wherein the voiced sound forming unit delays the whispered voice signal output from the voice input unit, and converts the delayed whispered voice signal into the voice While adding the whispered audio signal output from the input unit and sequentially delaying the added audio signal, the audio input unit Et newly by going further adding the whispering voice signal output, and forming a voiced sound signal corresponding to the voiced, voiced forming apparatus. 9. A voice recognition apparatus for recognizing a whisper having a spectrum in which a level difference between an amplitude level in a formant and an amplitude level in a band other than the formant is relatively small, wherein the whisper is input, A voice input unit that outputs a whispered voice signal corresponding to the whispered voice, from a whispered voice signal output from the voice input unit,
A voiced sound forming unit that forms a voiced sound signal corresponding to a voiced sound having a spectrum in which the level difference is relatively large,
And a voice recognition unit that recognizes the whisper based on a voiced sound signal output from the voiced sound formation unit. 10. The voiced sound forming section delays a whispered voice signal output from the voice input section, adds the delayed whispered voice signal to a whispered voice signal output from the voice input section, A voiced sound signal corresponding to the voiced sound is formed by further adding a whispered voice signal newly output from the voice input unit while sequentially delaying the voice signal after the addition. The speech recognition device according to claim 9.