JPS5837700A - Voice recognition unit - 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 The present invention relates to a speech recognition device, and particularly to a speech recognition device that can efficiently update registered patterns.

The speech recognition device compares the characteristics of a plurality of registration turns created in advance with the characteristics of the input speech pattern of the speaker's voice, and identifies which registration turn the input voice corresponds to. , the corresponding registration/arm turn is output to operate external equipment, etc. (for example, a welding machine). In a speech recognition device, it is necessary to create a registration/mother turn in advance, and conventionally, a registration pattern has been created by a speaker inputting speech a plurality of times and taking the average of the input speech.

However, because the voice used when creating the registration pattern differs from normal voice (for example, due to the speaker's nervousness), optimal registration and turns cannot be obtained, resulting in erroneous recognition, and long-term interruptions while operating the device. There was a problem in that the voice of the user changed over time, resulting in erroneous recognition or inability to recognize it. For this reason, conventionally, in order to improve the recognition rate, from time to time (for example, every 2 to 3 hours),
I had to update or partially update my registration and turns.

However, when updating the registered pattern, it is necessary to interrupt the voice input for operating the equipment, which causes the work process to stop, and this drawback is particularly serious in the assembly process. Furthermore, when updating the registration turn by stopping the work process, the noise component is not included, and therefore, the registration turn that does not include the noise component and the registration turn when the work process is in progress. Since the input speech and speech patterns in noisy environments are different, the recognition rate in noisy environments decreases.

The present invention has been made in view of the above-mentioned conventional problems, and its purpose is to improve voice input by using voice input for device operation for registration/lean update.
An object of the present invention is to provide a speech recognition device that efficiently updates registered patterns without interrupting input.

In order to achieve the above object, the present invention includes a voice analysis section that analyzes the voice signal from the microphone, a local memory section that stores the input voice nof turn signal from the voice analysis section, and a voice pattern created in advance. In a speech recognition device, the speech recognition device has a turn memory unit for registering and a logic processing unit that compares and identifies an input speech pattern with a registered/return pattern and supplies an output signal to an output display unit and an external device,
An update instruction signal input section is provided to supply an update instruction signal to the logic processing section, and the update is performed when the audio signal is correctly recognized.
The update instruction signal is used to average the multi-input voice pattern and the registered pattern, and the averaged turn is registered as a new turn in the registered pattern memory section, and the registered pattern is updated. If the audio signal is originally registered iJ? The turn is instructed by an update instruction signal, the registered pattern and the input voice pattern are averaged, and the 24 turns after the average are registered as a new registered pattern in the registered pattern memory section.Register! It is characterized by updating the turn.

Hereinafter, preferred embodiments of the present invention will be described based on the figures.

The figure shows a block circuit for speech recognition according to the present invention.

In the illustrated speech recognition device 10, a speaker (not shown) inputs speech into a microphone 12, and a speech signal 1 from the microphone 12 is output.
00 is the input speech pattern signal 10 from the speech analysis section 14 whose features are extracted and converted into patterns by the speech analysis section 14.
2 is stored in the local memory unit 16. On the other hand, in the registration/turn memory section 18, voice patterns created in advance are registered/turn 20-1, 20-2, etc.
. . . 20-n, and the logic processing section 22 uses the registered pattern from the registered pattern memory section 18 and the pattern of the turn signal 104 and the input voice from the local memory section 16. Compare the patterns of the signal 102 to determine which registered patterns 20-1, . . .
.....Identifies whether it corresponds to 20-n. Logic part 1'! As a result of the comparison and identification, the Ii section 22) supplies the output signal 106 of the corresponding registered pattern to the external device 24 to operate the external device 24, and further outputs the output signal 108 to a speech synthesis section, etc. One sound signal is supplied to the output display section 26 to display the output result.
In this device e, 20 registered patterns are used to improve the recognition rate.
-1, . . . , 20-n needs to be updated. Conventionally, as mentioned above, voice input is interrupted to register iJ? However, the present invention is characterized in that registration and turn updates can be performed without interrupting voice input by using voice input for device operation to update registered patterns. do.

In the figure, an update instruction signal input section 28 consisting of a key 7 is provided for the key 7 that supplies an update instruction signal 110 to the logic processing section 22. When updating the fourth turn of registration, the update instruction signal input section 28 An update instruction signal 110 is supplied from.

The speech recognition device according to the present invention has the above configuration, and its operation will be explained below.

The speaker inputs voice into the microphone 12, and the logic processing unit 22
The input voice pattern signal 102 is compared with the pattern of the lean registered pattern signal 104, and an output signal 106 of the corresponding registered pattern is supplied to the external device 24, and an output signal 108 is further supplied to the output display section 26. The speaker compares the input audio to the microphone 12 and the display on the output display section 26, and if the audio signal is correctly recognized and the registered pattern is updated, the speaker updates the registered pattern. The instruction signal 110 instructs the logic processing section 22 . Based on the update instruction signal 110, the logic processing unit 22 averages the input audio pattern in the local memory unit 16 and the registration/4' turn in the memory unit 18, and updates the averaged pattern. It is registered again in the registration/interval/memory unit 18 as an L1 registered pattern.If the audio signal is correctly recognized, the updating of the registered pattern is completed as described above.

On the other hand, if the voice signal is incorrectly recognized, the speaker informs the logic processing section 22 of the update instruction signal 110.
instruct. In response to the update instruction signal 110, the logic processing unit 22 averages the correct registration pattern that the audio signal should originally specify and the input audio pattern in the local memory unit 16, and takes the averaged pattern as a new registration/mother turn. Register again in the registration/IP turn memory unit 18. Further, the logic processing unit 22 supplies a correction signal 112 to the external device 24 to correct the pattern to the registered pattern specified by the speaker.

Note that the external device 24 is provided with a correction function. If the audio signal is erroneously recognized, the registration/renewal is completed as described above.

As described above, according to the present invention, registered patterns can be updated without interrupting voice input for device operation.

In the figure, when updating the registered Noreen, the local memory expansion section 30 stores the next input voice pattern signal 102, and the input sound p) + turn signal 102 from the voice analysis section 14 is stored in the local memory section 16. A selection switching unit 32 is provided for switching between supplying the data and supplying the data to the local memory expansion unit 30. Then, when the next input voice pattern signal 102 is input while the registered pattern is being updated, the logic processing section 22 supplies the switching signal 114 to the selection switching section 32, and the next input voice pattern signal 102 is inputted.
is stored in the empty local memory expansion section 30.

As explained above, according to the voice recognition device according to the present invention, voice input for device operation is used to update registered i4 turns, so voice input is not interrupted and the work process is not stopped. . In this way, since the registered pattern is updated while the work process is in progress, there is an advantage that work efficiency is improved, and this advantage is particularly great in the assembly process.

Furthermore, since the registered pattern is updated while the work process is moving, the registered pattern includes a noise component, and therefore the input voice/e-turn and the registered A' turn under noisy conditions when the work process is moving. However, the recognition rate under noisy conditions is improved.

[Brief explanation of the drawing]

The figure is a block circuit diagram of a speech recognition device according to the present invention. 10... Voice recognition device, device, 12... Microphone, 14.
...Voice analysis section, 16...° Local memo, Ri section, 18
···Registration·! Turn memory section, 20-1.20-2, 20-n
...Registration V-turn, 22...Logic processing section, 24...External device, 26
... Output display section, 28... Update instruction signal input section, 30... Local memory expansion section,

Claims (3)

[Claims] (1) A voice analysis section that analyzes the voice signal from the microphone, a local memory section that stores the input voice pattern 41'1 from the voice analysis section, and a registered pattern memory that registers the voice/zoo turn created in advance. and a logic processing section that compares and identifies the input speech pattern and the registered pattern and supplies an output signal to the output display section and external equipment, the update instruction signal is supplied to the logic processing section. An update instruction No. 48 input section is provided, and when the audio signal is correctly recognized, the update is performed by inputting the input voice/yaturn and registering A? by the update instruction signal. The averaged pattern with the turn is registered as a new turn (Registered as a turn/4') Registered pattern in the turn memory section and updated the registered pattern.Updating when the audio signal is recognized incorrectly is as follows. Registration that the audio signal should originally be specified) J? Instructs the turn by an update instruction signal, averages the registered pattern and the input voice note and turn, registers the average note and turn as a new registered pattern in the registered pattern memory section, and updates the registration and turn. A speech recognition device characterized by: (2) In the device according to claim +11,
During the registration pattern update, the next input audio pattern signal is
The present invention is characterized by being provided with a local memory expansion section for storing the audio data, and a selection switching section for switching whether to supply the input voice pattern signal from the voice analysis section to the local memory section or to the local memory expansion section. speech recognition device. (3) A speech recognition device according to claim (1) or (2), characterized in that when a speech signal is erroneously recognized, the logic processing section supplies a correction signal to an external device. .