JPS63149699A - 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] [Summary] In a voice input/output device, an already registered voice dictionary is
When learning to increase the recognition rate during operation after a few days, I forgot the nuance of the utterance at the time of registration.
In order to solve the problem of having to utter and input the same voice many times, a means is provided for storing the voice data at the time of registration as sample voice playback data. During learning, it is outputted aloud, and the operator inputs the voice while confirming the utterance method at the time of registration.

[Industrial application field]

The present invention relates to a voice input/output device, and particularly to a voice output processing (prompt processing) method used as a sample when learning a voice dictionary.

With the recent improvement in the processing power of computer systems and the growth in technology, it has become popular to input and process speech as data.

As an example, there is a voice input/output device for sorting work in a factory. In this case, when the voice of a specific speaker registered in the voice dictionary is confirmed through learning, the voice from the specific speaker is input and the voice input/output device for the above-mentioned sorting operation is used. However, when learning over a number of days, it is difficult to produce the voice under exactly the same conditions as at the time of registration, so even in such cases, try to use the voice dictionary as much as possible when you registered it. A voice learning method that allows learning under the same conditions is needed.

In particular, since the learning process of the speech dictionary is merely a preparatory work for increasing the recognition rate when operating the speech input/output device, it is desirable that the operation be performed as quickly and efficiently as possible.

[Prior art and problems to be solved by the invention] FIG. 3 is a processing flowchart showing the configuration of a conventional audio input/output device.

First, in a voice dictionary registration process 1, a specific speaker sequentially utters and inputs voices to create a voice dictionary.

Next, when the voice input/output device is used continuously for the sorting work in the factory mentioned above, learning process 2 is performed on the previously registered voice dictionary. In accordance with the "reading character string" for learning displayed there, utter and input the audio, match it with the audio at the time of registration using, for example, a known pattern matching method, and when all matches, the actual The sorting will involve operational processing such as voice input for work.

Therefore, in conventional voice input/output devices, a specific speaker registers a voice dictionary in advance, and during operation, learning is performed on the registered voice dictionary to stabilize the recognition rate during operation. However, after the registration process,
In the learning process after a certain number of days, there is a problem in that it is difficult to input into the same speech dictionary even if the utterances are the same, but the tone of the utterances are similar.

In addition, since it is impossible to check the registered contents, even if there is noise in the background that a specific speaker would not notice, it will be registered in the voice dictionary, and no matter how many times the same word is uttered or input during learning, it will be recorded. However, a situation may also occur in which it is not possible to match the voice data for registration.

As a result, a problem has arisen in that speech is not stable when registering the speech dictionary and during operation, and this causes problems in speech recognition.

In view of the above-mentioned conventional drawbacks, the present invention provides a learning method for smoothing the operation during learning and stabilizing the recognition rate during operation when using an already registered speech dictionary after a number of days in a speech input/output device. The purpose is to provide a method.

[Means for solving problems]

FIG. 1 is a process flowchart showing the configuration of the audio input/output device of the present invention.

In the present invention, there is provided a voice input/output device having a function of registering a voice dictionary, performing learning 2 for the registered voice dictionary, and then performing voice recognition, wherein at the time of registering the voice dictionary, means 12 for holding the input audio data as sample audio reproduction data; and means 21 for outputting the sample audio reproduction data as a sample audio when inputting audio at the time of learning 2; The system is configured to learn by inputting a voice with a utterance similar to that at the time of registration while listening to a sample voice.

[Effect]

That is, according to the present invention, when learning is performed to increase the recognition rate when using an already registered speech dictionary after a few days in a speech input/output device, it is possible to forget the nuance of the utterance at the time of registration. In order to solve the problem that the same voice may have to be uttered and input many times, a means is provided for storing the voice data at the time of registration as data for sample voice reproduction. This allows the operator to input the voice while checking the utterance method used during registration, making the operation smoother during learning, and making it easier to utter the utterances during voice recognition during operation. It has a stabilizing effect. Also, it is possible to detect audio data that requires re-registration.

〔Example〕

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

The above-mentioned FIG. 1 is a process flowchart showing the configuration of the audio input/output device of the present invention, and FIG. 2 is a diagram schematically showing an embodiment of the present invention, and the steps in FIG. 12.
21 are the means necessary to carry out the present invention.

Hereinafter, voice registration and learning processing in the voice input/output device of the present invention will be explained with reference to FIG. 2 while referring to FIG.

First, the specific speaker using the audio input/output device
In the speech dictionary registration process 1 shown in the figure, a user's own speech dictionary is created and registered in advance via the speech recognition device (2). (See step 10.11) At this time, in the present invention, the registered audio data is converted into audio playback data in a data format different from the registration data (for example, a data format converted into a simple digital signal). For example, save it on a microdisk (.Furthermore, with this voice output device, display data can also be saved as
Save it as a “reading character string” on the microdisk ■ above. (Refer to step 12) The above registration data is in a data format in which voice recognition, which will be described later, is efficiently performed, specifically, for example, in a 6-data format for performing well-known pattern matching. It is registered on disk ■ and is different from the data format for audio playback described above.

Next, the case of operating the audio input/output device will be explained.

First, when the specific speaker ■ who is the operator presses the learning operation key, the screen ■ on the display shows the following information as shown in the figure.
For example, the reading is ``ho no kaido''.

A learning screen for the character string "Hokkaido" is displayed. (See step 20.) Next, in the voice recognition device ■, ', the display content, the content (
Registered audio playback data compatible with "Hokkaido") is output from the speaker ■. (See step 21) The specific speaker (operator) (2) learns by uttering a voice that is closer to the time of registration, keeping in mind the reading "Horakaido" on the screen of the display device (2) and the "Horakaido" in the audio output. The human-generated voice data for recognition is checked in the voice recognition device against the data registered in the microdisk (2).

Once a match is found, the next word will be learned. (Refer to Steps 22.23.24) In the message display field under ■ on the display screen above,
The learning result is, for example, "the vocalization is too loud",
A so-called warning message such as "Please speak again" is displayed.

As described above, in the audio input/output method by a specific speaker, when registering a speech dictionary for the specific speaker, the audio playback data and data are created and saved, and the During learning, the above-mentioned audio playback data corresponding to the learning words displayed on the display screen is output from the speaker, and based on the learning words on the display screen and the audio output data from the speaker, the audio data is recorded when registering the audio dictionary. The feature is that the voice input is performed in a similar way to the generation, and the learning process is performed smoothly.

〔Effect of the invention〕

As described above in detail, the voice input/output device of the present invention is useful for learning to improve the recognition rate when using an already registered voice dictionary after a few days. In order to solve the problem of forgetting the nuance and having to utter and input the same voice over and over again, the voice data at the time of registration above is retained as sample voice playback data. By providing a means to output the voice during learning, the operator can input the voice while confirming the utterance method at the time of registration, making the operation smoother during learning and improving the voice output during operation. It has the effect of stabilizing speech during recognition. or,
It is possible to detect audio data that requires re-registration.

[Brief explanation of the drawing]

FIG. 1 is a process flowchart showing the configuration of the audio input/output device of the present invention. FIG. 2 is a diagram schematically showing an embodiment of the present invention. FIG. 3 is a processing flowchart showing the configuration of a conventional audio input/output device. It is. In the drawings, 1 is a voice registration process, and 2 is a voice learning process. 10-13. 20-25 are each processing step. ■ is the display screen, ■ is the voice recognition device. ■ is the operator, ■ is the micro disk. ■ is a speaker. are shown respectively. Processing/I 5L bill map No. 5Il! l Yakoto Hiri-

Claims (1)

[Claims] A voice input/output device having a function of registering a voice dictionary (1) and performing voice recognition after performing learning for the registered voice dictionary (2), the voice dictionary having the function of performing voice recognition. means (12) for retaining the input audio data as data for sample audio playback at the time of registration (1);
), and means (21) for outputting the sample voice reproduction data as a sample voice when inputting voice during the learning (2).
A voice input/output device comprising: a voice input/output device which inputs a voice for learning while listening to the sample voice.