JPS63183498A - Registration type voice input/output device - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of the Invention The present invention relates to a registered voice input/output device used for issuing commands to various devices by voice.

2. Conventional technology In recent years, voice information processing such as voice recognition and voice synthesis, and L
With the development of SI technology, voice recognition equipment.

Speech synthesis equipment began to be used in industrial equipment, consumer equipment, etc.
Registered voice input/output devices that combine a voice recognition device and a voice synthesis device have also begun to be put into practical use.

An example of a conventional registered voice input/output device will be described below with reference to the drawings.

FIG. 2 shows a block diagram of a conventional registered voice input/output device.

In FIG. 2, 1 is a microphone that converts air vibrations of sound into electrical signals, 2 is a sound section detection means that detects the sound section of the electrical signal of sound (hereinafter referred to as "sound signal") inputted by microphone 1, and 3 is the voice signal of the voice section detected by the voice section detection means 2, which is one of the voice characteristic parameters.
LPG cepstrum analysis means for converting into PG cepstrum coefficients, S is a recognition-registration switch, L
This is a changeover switch for sending the 1LPc cepstrum coefficients converted by the PG cepstrum analysis means 3 to the pattern mark 3 (described later) during recognition, to the notching means 4 (described later), and to the standard pattern storage RAM 5 (described later) during registration. Reference numeral 5 denotes a standard pattern storage RAM, in which the audio signal f input through the microphone 1 at the time of registration is converted into an ILPC Keb Nutram coefficient and stored therein.

4 is a pattern matching means, microphone 1 is used during recognition.
The degree of similarity between the LPG cepnutrum coefficients of the human-generated audio signal and the LPG cepnutrum coefficients of the standard pattern stored in the standard pattern storage RAM 5 is determined, and the most similar one is taken as the recognition result. Above is Mike 1. Voice section detection means 2, LPG cepstrum analysis means 3. Pattern matching means 4. The registration type speech recognition device is constituted by the standard pattern storage RAM 5 and the recognition/registration changeover switch S1.

Reference numeral 8 is a synthesis pattern storage ROM, which is a memory that stores human DPCM synthesis patterns for outputting information when creating a registered voice input/output device and for generating voice synthesis that provides complete guidance on handling the registered voice recognition device. , 7 is an ADPCM synthesis means, which reproduces the synthesis pattern stored in the synthesis pattern storage ROM 5 into an audio electrical signal. A speaker 6 converts the electric signal of the sound reproduced by the ADPCM synthesis means into air vibration of the sound. The speech synthesis apparatus is constituted by the synthesis pattern storage ROM 8 and the ADPCM synthesis means synthesis means 6.

Reference numeral 11 denotes a controlled device, which is controlled based on the recognition result of the pattern matching means 4 of the registered voice recognition device. Note that the controlled device 11 issues a command to the ADPCM synthesis means of the speech synthesizer as to which synthesis pattern in the synthesis pattern storage ROM 5 is to be reproduced.

The operation of the conventional registered voice input/output device configured as described above will be explained.

At the time of registration, the controlled device 11 is ADPCM all PCM 7
The audio signal reproduced by the OADPCM synthesis means 7 issues a command to read and reproduce the synthesis pattern to cause the registration voice to be uttered from the synthesis pattern storage ROMA (for example, "Please say it") to the speaker 6. The vibrations are converted into air vibrations and reach the user's ears. The user utters a registration voice in response to this synthesized voice at 5/. This voice is converted into an electrical signal by a microphone 1, a voice interval is detected by a voice interval detection means 2, and the voice interval is converted into an LPC cepstrum coefficient by an LPC cepstrum analysis means 3. The recognition-registration changeover switch S is for standard pattern storage RA during registration.
It is connected to send the LPG cepnutrum coefficients to M5, and the LPG cepnutrum coefficients are stored in the standard pattern storage RAM 6 as a standard pattern. The above operation at the time of registration is performed by the number of instructions sufficient for the registered voice recognition device to control the controlled device 11. Next, at the time of recognition, the controlled device 11 is
At step 7, a command is issued to read and reproduce a composite pattern for uttering all recognition voices (for example, "Please give me a command") from the composite pattern storage ROM 5. The audio signal reproduced by the human DPCM synthesis means 7 is converted into air vibrations by the speaker 6 and reaches the user's ears. The user utters a voice in response to this synthesized voice. This voice is converted into an electric signal 6 by the microphone 1, the voice zone is detected by the voice zone detecting means 2, and the voice zone is detected by the LPC cepstrum analysis means 3.
It is converted into a G kepnutrum coefficient.

The recognition/registration changeover switch S is connected to send the LPG cepstrum coefficients to the pattern matching means 4 during recognition, and the pattern matching means 4 selects the standard stored in the standard pattern storage RAM 5 during registration. The degree of similarity between the LPC kepnutrum coefficient of the pattern and the LPG kepnutrum coefficient of the input voice is determined, the most similar one is taken as the recognition result, and the user is asked to confirm whether the recognition result is correct (for example, by ×Isn't it?
'') is read by all ADPCM PCMs 7 from the composite pattern storage ROM 8 and reproduced. The following is the same as the above recognition. The user uses words that mean confirmation or correction (for example, ")・i".

Say "Yes"). If this recognition result means confirmation, the controlled device 11 is controlled based on the previous recognition result, and if it is a correction, the user is made to utter a command to control the controlled device 11 again and go to step 7. 7 RU0 Problems to be Solved by the Invention However, in the above configuration, the synthetic pattern that prompts the user to say whether the recognition result is correct is a synthetic pattern that is stored in advance in the synthetic pattern storage ROM. When a user instructs a registered voice input/output device to perform a recognition operation based on the user's own voice commands, the user must use the synthetic pattern meant by the voice commands, which may be troublesome to confirm. It had a point.

In view of the above-mentioned problems, the present invention allows a voice synthesizer to reproduce the recognition result as the user's own voice command even when the user causes the registered voice input/output device to perform a recognition operation using the user's own voice command. The present invention provides a registered voice input/output device that can be used as a registered audio input/output device.

Means for Solving the Problems In order to achieve the above object, the registration type voice input/output device of the present invention includes a registration type voice recognition device and a registering type voice input/output device that records the voice of the speaker used at the same time as registering for recognition at the time of voice registration. The present invention is characterized by comprising a speech synthesis device which converts and stores feature parameters effective for synthesis and reproduces them into speech when used.

According to the above-described configuration, the present invention first converts the voice of the speaker in use into feature parameters effective for speech recognition and registers the same at the time of registration, and at the same time converts this speech into feature parameters effective for speech synthesis. Remember it. Then, during recognition, the degree of similarity between the input voice and each of the registered feature parameters effective for voice recognition is determined, and the most similar one is taken as the recognition result. At this time, the user's unique voice command can also be answered by reproducing feature parameters effective for voice synthesis corresponding to the recognition result using the voice synthesizer.

Embodiment Hereinafter, a registered voice input/output device according to an embodiment of the present invention will be described with reference to the drawings.

FIG. 1 shows a block diagram of an embodiment of the registered voice input/output device of the present invention.

In FIG. 1, 1 is a microphone, 2 is a voice section detector 9 vane mountain means, 3 is an LPG cepstrum analysis means, 4 is a pattern matching means, 5 is a standard pattern storage RAM, S,
is a recognition-registration changeover switch, and the above is the same as the conventional example, and the registration type speech recognition apparatus is configured by the above.

Reference numeral 8 is a composite pattern storage ROM, 7 is an ADPCM composition means, and 6 is a speaker, which is the same as the conventional example. 9
is an ADPCM analysis means, which converts the speech signal of the speech section detected by the speech section detection means 2 at the time of speech registration into feature parameters using human DPCM which is effective for speech synthesis. Reference numeral 1o denotes a synthesis pattern storage RAM for storing feature parameters converted by the human DPCM analysis means, and the speech synthesis apparatus is configured as described above.

Reference numeral 11 designates a controlled device, which is similar to the conventional example, but includes a composite pattern storage RAM 1 in addition to the composite pattern storage ROM 5.
It also issues a command to reproduce the composite pattern in 0.

The operation of the registration type audio input/output device of this embodiment configured as described above will be explained.

10 To-7 At the time of registration, the controlled device 11 uses the ADPCM synthesis means 7
Then, a command is issued to read and reproduce a synthetic pattern indicating that a registration voice is to be uttered (for example, "Please say 1") from the synthetic pattern storage ROM 5. The audio signal reproduced by the ADPCM synthesizing means 7 is converted into synthesized audio by the speaker 6 and reaches the user's ears. The user utters a registration voice in response to this synthesized voice. This voice is converted into an electrical signal by a microphone 1, a voice interval is detected by a voice interval detection means 2, and is converted into an LPG cepnutrum coefficient by an LPG cepstrum analysis means 3. The recognition/registration changeover switch S1 is connected to send the LPG cepnutrum coefficients to the standard pattern storage RAM 5 during registration, and the LPC cepnutrum coefficients are stored in the standard pattern storage RAM 5 as a standard pattern.

On the other hand, the voice section detected by the voice section detection means 2 is converted into characteristic parameters by the ADPCM analysis means 9 and stored in the synthetic pattern storage RAM 10. The above operation at the time of registration is performed by the controlled device 1111.

A sufficient number of instructions are executed for the registered speech recognition device to control.

Next, at the time of recognition, the controlled device 11 issues a command to the ADPCM synthesizing means 7 to read and reproduce a synthetic pattern for uttering a recognition voice (for example, "Please give me a command") from the synthetic pattern storage ROMB. The audio signal reproduced by the ADPCM synthesizing means 7 is converted into synthesized audio by the speaker 6 and reaches the user's ears. The user utters the entire voice in response to this synthesized voice. This voice is converted into an electrical signal by the microphone 1, the voice interval detection means 2 detects the voice interval force, and the LPG cepstrum analysis means 3 converts the voice into an electric signal.
It is converted into a G kepnutrum coefficient.

The recognition/registration changeover switch S is connected to send the LPG cepstrum coefficient to the pattern matching means 4 during recognition, and the pattern matching means 4 selects the LPG cepstrum coefficient of the standard pattern stored in the standard pattern storage RAM 5 during registration. The degree of similarity between the cepstrum coefficient and the LPG cepnutrum coefficient of the input voice is calculated, the most similar one is taken as the recognition result, and the user is asked to confirm whether the recognition result is correct (for example, "It's XX"). ?”
) is read from the composite pattern storage ROM 8 and the composite pattern storage RAM 10 by the ADPCM composition means 7 and reproduced. The following is the same as the above recognition. The user utters all words meaning confirmation or correction (eg, "hi", "yes"). If this recognition result means confirmation, the controlled device 11' (i-control is performed based on the previous recognition result, and if it is a correction, the user is made to utter a command to control the controlled device 11 again. .

As described above, according to this embodiment, the characteristic parameters of the standard pattern used for speech recognition are stored in the standard pattern storage RAM 5 at the time of registration, and at the same time, the synthesis patterns used for speech synthesis are stored in the synthesis pattern storage RAM 10. You can answer back your own voice commands. Furthermore, since the LPG Kebnutram coefficient is used as a feature parameter for speech recognition, high quality speech recognition can be performed. Furthermore, since ADPCMi is used as the speech synthesis characteristic parameter, high quality synthesized speech can be reproduced.

Effects of the Invention The present invention converts the voice of the speaker used during registration into feature parameters effective for speech recognition and registers the same, and at the same time converts this speech into feature parameters effective for speech synthesis and stores the same. I'll keep it. Then, during recognition, the degree of similarity between the input voice and the registered voice is determined, and the most similar one is taken as the recognition result. On this occasion,
Feature parameters effective for speech synthesis corresponding to the recognition results are reproduced by the speech synthesis device. Therefore, the user's own voice commands can also be answered using synthesized voices.Furthermore, the characteristic parameters for voice recognition are those that are effective for voice recognition, and the characteristic parameters for voice synthesis are those that are effective for voice synthesis. The present invention is intended to realize a registered voice input/output device that can achieve a number of excellent effects, such as being able to perform high quality voice recognition and voice synthesis because of the use of characteristic parameters.

[Brief explanation of the drawing]

14.\2 FIG. 1 is a block diagram of a registered voice input/output device according to an embodiment of the present invention, and FIG. 2 is a block diagram of a conventional registered voice input/output device. 1...Microphone, 2...Voice section detection means, 3...LPG cepstrum analysis means, 4.
...Pattern matching means, 5...Standard pattern storage RAM. Sl...Recognition-registration switch, 6...
... Speaker, 7 ... ADPCM synthesis means, 8 ... Synthesis pattern storage ROM, 9 ...
. . . ADPCM analysis means, 10 . . . Composite pattern storage RAM, 11 . . . Controlled equipment. Name of agent: Patent attorney Toshio Nakao and 1 other person No. 1
Fig. 5 Igo y salt-=v ~ multi-November Jφ-nomata A solar special opening UG3-183498 (5) Fig. 2

Claims (1)

[Claims] A registration method that converts the user's voice into feature parameters effective for voice recognition and registers it in advance, then calculates the degree of similarity between the input voice and the registered voice at the time of use and outputs the most similar one as the recognition result. A voice recognition device, and a voice that converts and stores the voice of the user using the speaker into characteristic parameters effective for voice synthesis almost simultaneously with the registration, and outputs information and guides the handling of the registered voice input/output device. What is claimed is: 1. A registered voice input/output device comprising: a voice synthesizer which converts and stores characteristic parameters effective for voice synthesis into characteristic parameters, and reproduces the characteristic parameters into voice when used.