JPH04200049A - Voice dialer - Google Patents

Abstract

PURPOSE:To recognize a key word in a memory area registered by the user without any special operation on each occasion of change in users by providing a voice data analysis means, a voice data storage means and a voice data control means to the dialer. CONSTITUTION:The dialer is provided with a bank memory 7 storing voice data for plural persons, each bank memory 7 is provided with an area storing a voice recognition characteristic data and a characteristic data for talker collation, and voice analysis means 3-5 are used to extract the voice recognition characteristic data from a designated voice signal and to extract the talker collation characteristic data and they are stored in the bank memory 7. Thus, when voice start is implemented by the recognition of a specific password, the key word is recognized and the talker is collated and the bank memory 7 registered by the user is selected and the voice dial mode is transited. Thus, the key word in the memory area registered by the user is used for an object of recognition without any special operation.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a voice dialing device that uses voice recognition to simplify the dialing operation of a telephone device or the like.

[Prior Art] There is a demand for a voice dialing device consisting of a voice recognition device and a telephone device to be usable by multiple people6. The reality is that G-devices must ensure their performance by identifying speakers, limiting the number of recognized words, and so on.

In this way, the number of recognition target words that can be handled at a time may be limited, so in order for multiple people to be able to use it, prepare a memory area (bank) for each user and use this. You will need to switch and use it each time the person changes.

[Problem to be solved by the invention] In such a configuration, when using voice dialing,
The user must tell the system which bank memory to use. This can be done by specifying a phone number such as Sui-Sochi, but using a switch for voice dialing, which should be hands-free and eyes-free, loses its convenience.

Also, by voice recognition of a specific password, it is possible to have the system automatically perform voice recognition to switch bank memories, but this requires an extra operation (uttering a voice to switch bank memories). There is a problem.

On the other hand, in order to shift from a normal mode using a telephone to a voice dialing mode using voice recognition, the user needs to convey his or her intention to the system in some way.

Conventionally, the transition to voice dialing mode has been achieved by pressing and holding a button on a telephone for a certain period of time or by voice recognition of a specific keyword.

If the former method is used, hands-free operation is no longer possible due to the operation of pressing keys. In the latter method, voice activation may be delayed due to erroneous recognition.

In addition, even in the case of voice recognition that specifies a speaker, recognition may occur if a person with a similar voice quality utters the voice.As a result, voice activation may occur even if the voice is uttered by a person different from the person who registered the voice. I end up. A similar situation may also occur when erroneous recognition is caused by ambient noise.

Therefore, as a result of misrecognizing utterances such as pranks or other sounds, there is a problem in that the device may be activated when it is not necessary.

OBJECTS OF THE INVENTION An object of the present invention is to provide a voice dialing device that can recognize keywords in a memory area registered by the user without any special operations each time the user changes. be.

[Means for Solving the Problem] In order to achieve such an object, the present invention includes a bank memory that stores voice data of a plurality of people, and each bank memory has feature data for voice recognition and speaker verification. An area is provided to store feature data for voice recognition, and a voice analysis means is used to extract feature data for voice recognition from a specified voice, as well as feature data for speaker verification. It is characterized by being stored in.

[Function] The present invention stores characteristic data for voice recognition and speaker verification in a bank memory, and performs speaker verification at the same time as keyword recognition when voice activation is performed by recognizing a specific password. You can switch to the bank memory registered by the person and switch to voice dial mode.
Keywords in the memory area registered by the user can be targeted for recognition without any special operations.

[Example] Embodiments of the present invention will be described in detail below with reference to the drawings.

FIG. 1 is a system configuration diagram of an embodiment of the voice dial IT according to the present invention, where ■ is a microphone, 2 is an amplifier, 3 is a voice analysis section, 4 is a feature extraction section for voice recognition, and 5 is a feature for speaker verification. Extraction unit, 6 is first control unit, 7 is first button memory, 8 is external ■/○, 9 is speech synthesis unit, 10 is second memory, 11 is second control unit, 12 is dial control unit,
13 is a telephone, and 14 is an external telephone line.

In such a configuration, when a voice is input from the microphone 1, the voice is converted into an electrical signal, amplified to a desired amplitude level by the amplifier 2, and input to the voice analysis section 3.

The audio analysis unit 3 is a group of filters for analyzing audio signals, and divides the audio band into multiple channels using bandpass filters, converts the output of each channel into an absolute value, and outputs a signal smoothed using a low-pass filter. do.

That is, the audio analysis unit 3 outputs an envelope of power occupying each band of the audio signal.

The speech analysis section 3 extracts the envelope of the speech signal spectrum and inputs it to the speech recognition feature extraction section 4 and the speaker verification feature extraction section 5, respectively.

These extraction units 4 and 5 are controlled by a control unit 6, which performs sound I recognition, speaker verification, and communication with external devices.

The feature extractor 4 extracts feature amounts for use in speech recognition based on the signal input from the speech analyzer 3, and stores them in a predetermined area of the button memory 7.

Furthermore, when a specification is received from an external device via the external l108, the speaker verification feature extraction unit 5 extracts parameters (features) for speaker verification from the signal input from the speech analysis unit 3. , stored in a predetermined area of the button memory 7.

The voice synthesis section 9 is for the system to output a guidance message to the user, record the voice at the time of registration, and echo it back when using voice dialing so that the user can confirm the keyword. be. The memory 10 stores these fixed message data and voice data.

There are two main types of voice data for voice recognition handled by this system: control words and keywords.

Among these, the control word controls the flow of processing in the system control sequence, for example,
This is a word sound used when transitioning from normal mode to voice control mode by voice activation. When registering this control word, data for voice recognition is extracted and at the same time data for speaker verification is extracted and stored in a predetermined area of the button memory 7. When recognizing a control word, speaker verification is performed at the same time, and the corresponding bank memory is switched based on the verification result.

On the other hand, the keyword is a word sound for speech recognition that controls each operation of the controlled device. At the time of registration, only voice data for voice recognition is registered, and the data is stored in the vanniglia corresponding to the specified user.

FIG. 2 is a flowchart showing the main operation of the system in FIG. 1, and FIGS. 3 and 4 are flowcharts showing control word and keyword registration operations, respectively.
FIG. 5 is a flowchart showing the operation of the speaker switching mode.

FIGS. 6 and 7 are diagrams showing examples of the contents of the memory 7 in which control words and keywords are registered, respectively.

Hereinafter, an example of automatic dialing by voice recognition of a telephone according to the present invention will be explained in detail with reference to FIGS. 2 to 7.

The voice recognition system is activated to start voice dialing, but a control word is required for this purpose.

Here, when a predetermined specific function code (for example, registration mode Func"1 is assigned) is input from the button key attached to the telephone 13, the dial control unit 12 of the first TXi inputs this code. The function code is detected (step 21 in FIG. 2), and the control unit 11 is notified that this function code has been detected, and the control unit 11 then
The process moves to the control word registration mode (step 22 in FIG. 2).

The operation of the second control word registration mote will be explained with reference to FIG.

Control Registration Mode (In this case, under the control of the control unit 11, the speech synthesis unit 9 sends a voice message, for example, "Control word registration mode" to the speaker, to inform the user that it is in this mode. 15 (step 40).

At this time, this guidance may be provided on a display unit attached to the telephone 13, and in that case, it may be used in conjunction with a voice message.

Next, a voice message such as "Please say something" is output from the speaker 15, a registration number is displayed on the display, and the user is informed of the voice that should be registered as a control word. A voice request instruction is issued (step 41).

According to the message, the user selects the control word,
For example, if you say "telephone" (step 42.
43), as described in fm, extract parameters for speech recognition and data for speaker verification (steps 44 and 45).
)6 FIG. 6 shows an example of a part of the button memory 7 when such a control word is registered.

Speech recognition data and speaker verification data for the control word "telephone" are stored as a pair for each user (speaker in the figure). In this case, there is only one type of control word, but a plurality of control words may be registered (steps 46 and 47). Once the required number of controls are registered, exit the current mode and return to normal mode.

Next, the user selects a specific function word (eg, Func*#2.
(to which speaker switching mode is assigned),
Shift to speaker switching mode (step 23.24)
.

The operation of the speaker switching mode will be explained with reference to FIG.

When the speaker switching mode is entered, the currently selected bank number is notified to the user (step 71). This can be done by outputting a voice message such as "The current bank number is..." using the above-mentioned sound synthesis method, or by displaying the message if the telephone 13 has a display device.

When in speaker switching mode, for example, if a number key is pressed (step 72), the bank number is incremented (step 73), and if the bank number does not exceed the maximum value (step 74), the current bank number is If the maximum value is exceeded, set the bank to 1 (step 75),
Bank memories are switched according to the hang number (step 76).

When the bank switching is completed, the bank number is notified again by voice message (step 71).

If the numeric key is pressed again (step 72), the same process as above is performed, and if the END key is pressed (step 77), the current mode is terminated.

Next, when the function code assigned to the keyword registration mode (for example, Func"3:) is input, the mode shifts to the keyword registration mode (step 25).
．． 26).

The operation in this mode will be explained with reference to FIG.

When the keyword registration mode is entered, an entry number input request is made on the display or by voice message (step 52). If the entry number is input and it is valid, the process moves to the next step, and if it is not determined, the number input process continues (steps 53 and 54).

Next, under the control of the control unit 11, a voice request such as a voice message "Keyword please" is made (step 55). When the user utters a keyword (for example, "Torihikisaki") following the message (step 56), data for speech recognition is extracted (step 57), and as shown in FIG. When the registration of one voice data stored in bank area 1 in (1) is completed, a message requesting registration of a telephone number will be displayed on the display attached to the telephone, or a voice message "li input telephone number" will be displayed. (Step 58)
).

While the input of the telephone number is not confirmed, the telephone number input process is performed (steps 59 and 60), and when the number is confirmed, the input data is stored in a predetermined memory area (step 61). The number is stored in association with the previously registered audio data. The memory for storing telephone numbers may be shared if the telephone supports speed dialing. Alternatively, a part of the memory 7 or 10 may be used. -It is also possible to register all of them together. - This method may be convenient when there is a large amount of data to be registered at one time - Keyword registration mode can be returned to normal mode by pressing the END key (Step 5).
1).

The automatic dialing operation will be explained based on the flowchart in FIG.

In a normal mode in which automatic voice dialing is not performed, a control signal is set in the dial control section 12 so that the telephone 13 operates in a single mode. Therefore, while the automatic dialing operation is not performed, the control unit 11
It only sets signals for operating the telephone 13 independently, and does not perform any other dial control.

Therefore, in the normal mode, the system starts speech recognition with the control word as the recognition target word, and only monitors the speech activation word "telephone" (step 20).

At this time, what is to be recognized is the speech recognition data for the control word "telephone" of all registered speakers.

If the voice recognition for "telephone" fails, try the recognition again and monitor the next utterance (step 27).
. While "telephone" is not set, check whether the function code is input (step 27.2).
0). When a function code is input, voice recognition is performed again and the "telephone" is monitored (steps 20 to 27).

When the speaker (1) utters the control word "telephone" for voice activation and the voice recognition is successful (Yes in step 27), the system performs speaker verification processing using data for speaker verification. (Step 28) When the speaker (ri) is selected based on the result of the 5 comparisons, the recognition target memory is switched to the registration bank of keyword data registered in advance by the speaker (1) (step 29).

When the voice activation is performed through the above-described process, a voice dialing sequence is executed and a call state is established (step 30).
). After that, the system goes into standby mode until you confirm that the call has ended, and then returns to normal operation mode (
Step 31).

Note that it goes without saying that the present invention is not limited to the embodiments described above, and that various modifications are possible. Since the hang memory used by the user is determined and switched, when multiple users have registered keywords in each bank, voice recognition can be performed without switching bank memories each time the user changes. I can do it.

Furthermore, when switching between bank memories, it is possible to use voice instead of manually operating a switch, making it possible to achieve complete hands-free operation.

Furthermore, even if there is a misrecognition due to ambient noise or the voice of a third party, the user can be rejected as an unregistered user through speaker verification, reducing the chances of voice activation due to misrecognition compared to conventional methods. be able to.

[Brief explanation of the drawing]

FIG. 1 is a system configuration diagram of an embodiment of a voice dialing device according to the present invention, FIG. 2 is a main flowchart, FIGS. 3 and 4 are flowcharts of control word and keyword registration operations, respectively, and FIG. The flowcharts of the user switching mode, FIGS. 6 and 7, are diagrams for explaining the registration status of control words and keywords, respectively. 3°...speech analysis section, 4-. feature extraction section for speech recognition, 5 feature extraction section for speaker verification, 13. telephone set. Patent Applicant Glarion Co., Ltd. Agent Patent Attorney Nagai 1) Takeshi Part 1 Figure 2 Figure 3 Figure 5 Figure 6 Figure 7

Claims (2)

[Claims] (1) A voice dialing device that performs voice recognition by comparing and collating the voice keyword to be recognized corresponding to a registration number with the input speaker's voice data, and performs a dialing operation of the registration number based on the voice recognition result. a voice data analysis means for extracting respective feature quantities of the voice recognition feature data and the speaker verification feature data of the voice data and outputting them as analysis results; an area for registering feature data, an audio data storage means for storing the feature data, and an audio data control for controlling the analysis results obtained by the audio data analysis means to be stored in the audio data storage means. 1. A voice dialing device comprising means for performing a dialing operation based on an analysis result of said voice data analyzing means. (2) When dialing the registration number using predetermined password recognition, speaker verification is performed along with the voice keyword recognition, and based on the verification result, the area provided in the voice data storage means is assigned to the speaker. 2. The voice dialing device according to claim 1, further comprising selection control means for corresponding selection.