JPH0437997B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Technical Field of the Invention] The present invention relates to a speech recognition device that can effectively and reliably recognize input speech uttered in synchronization with a signal tone.

[Technical background of the invention and its problems]

Recently, the development of speech recognition technology has been remarkable, and some devices such as speech recognition devices that recognize isolated speech with high accuracy and voice typewriters that input speech that is divided into monosyllables are now in practical use. It has come to be By the way, in the case of the above-mentioned word speech recognition device, it is not enough to recognize only one uttered sound, but when recognition of one sound is completed, it is necessary to immediately perform recognition processing for the next sound. . Conventionally, therefore, a signal sound such as a "beep" is output every time one sound is recognized, and the next sound is prompted to be uttered in synchronization with this signal sound. Also, for voice typewriters,
Since syllables that are continuously uttered are handled, in order to alleviate the problem of segmentation, measures are taken to output the signal tone at a predetermined cycle to indicate the timing of voice utterance. Therefore, the speaker only needs to monitor the signal tone and utter the voice in synchronization with the signal tone.

Now, regarding conventional general speech recognition devices,
The results of recognition of input speech are displayed using a display device such as a CRT character display attached to the above-mentioned device. For this reason,
In order to confirm the recognition result, the speaker needs to visually check it, and the advantages of the voice input method, such as high speed, are significantly impaired. In order to eliminate such problems, Japanese Patent Application Laid-Open No. 56-138799, etc.
It is disclosed that two signal tones, ``pitsu'' and ``boo'', are prepared and the signal tones are made different depending on whether or not a voice input in synchronization with the signal tone is correctly recognized. According to such a means, effects such as being able to prompt the user to re-input a voice that could not be recognized can be obtained. However, if things continue as they are, it is unclear how to pronounce them to ensure correct speech recognition, and there is a problem in that there is little hope of preventing errors.

[Purpose of the invention]

The present invention has been made in consideration of these circumstances, and its purpose is to easily inform the speaker of the recognition status of input speech and to effectively prevent recognition errors. An object of the present invention is to provide a highly practical speech recognition device capable of speeding up speech input.

[Summary of the invention]

In order to achieve the above object, the speech recognition device according to the present invention includes: a means for outputting a signal sound prompting voice input; a means for recognizing and processing a voice signal input as a voice in synchronization with the signal sound; This means corresponds to the average recognition reliability that is the average of the recognition reliability from the recognition processing performed at the present moment and the recognition reliability from the recognition processing performed within a period that goes back a certain number of recognition processings from the present time. and output form variable means for changing the output form of the signal tone.

〔Effect of the invention〕

Thus, according to the present invention, the recognition status of the input voice, that is, the recognition reliability, is notified to the speaker as a change in the signal sound, and the speaker is encouraged to produce the correct voice, thereby preventing erroneous recognition and improving the accuracy of the voice input. It becomes possible to increase the speed. In particular, the average recognition reliability obtained by averaging the recognition reliability obtained by the recognition processing performed at the present time and the recognition reliability obtained by the recognition processing performed within a period extending back a certain number of recognition processings from the present time is calculated as described above. Since the output form of the signal tone is variable, when inputting voice over a long period of time, the speaker can be informed of how the vocalization tendency is changing over time. As a result, the user's confusion that tends to occur when prompting for input using a signal tone can be alleviated, and the ease of use can be improved. In addition, since it is only necessary to vary the signal sound using the recognition reliability, for example, the similarity calculation result in the speech recognition processing process, the control is easy and highly practical.

[Embodiments of the invention]

Hereinafter, one embodiment of the present invention will be described with reference to the drawings.

Figures 1a to 1c show a signal sound a that is output at a predetermined period to prompt voice input, the power b of a voice input in synchronization with this signal sound, and similarities in the recognition processing of the above-mentioned voice. This shows the relationship with degree c. Here, in synchronization with the signal tone,
A change in voice power is shown when the following voices are inputted as utterances.

FIG. 2 is a schematic configuration diagram of the embodiment device, and 1 is a voice recognition section. This speech recognition unit 1 analyzes input speech to obtain its speech feature parameters, performs pattern matching with standard speech patterns registered in a dictionary by, for example, calculating similarity, and performs pattern matching with standard speech patterns registered in a dictionary, and performs pattern matching according to the similarity value. It recognizes the input voice. In this case, the reliability of the recognition is expressed by the similarity value obtained in the recognition processing process, the similarity difference between the first and second places obtained in descending order of similarity, and the like. Therefore, this reliability information is stored in the shift register 2 each time one sound is recognized. The similarity averaging circuit 3 calculates the reliability (similarity value) of the past few samples stored in the shift register 2.
This is then averaged to determine the speech recognition status. Reliability information indicating the recognition status thus determined is stored in the register 4 as control data.

The signal sound generation circuit 5 generates a signal sound to prompt the speaker to input the next sound in response to a signal outputted each time the speech recognition unit 1 finishes recognition processing for one sound of input speech. It is. The output form of this signal sound is variable according to control data stored in the register 4.

The output form of the signal tone is controlled as follows. In other words, as shown in FIG.
As described above, when accurate speech recognition is being performed, the frequency is controlled to be constant.
Then, as the reliability of the voice recognition increases, the signal sound frequency is variably controlled to correspond to the reliability, such as by lowering the frequency. Note that this signal sound may be controlled by, for example, changing the amplitude and output time width of the signal sound depending on the reliability, as shown in FIG. 4, for example. Furthermore, the signal tone output waveform may be distorted to indicate a decrease in reliability. Specifically, if the frequency of the signal tone is changed in a range of 2000Hz to 1000Hz corresponding to the reliability level of 0.1 to 1.0, the recognition can be easily performed.

In this way, according to this device in which the form of the signal tone is variably controlled according to the reliability of the voice recognition, when the speaker speaks in synchronization with the signal tone, the previous method of vocalization can be changed to the device. It is possible to easily know whether or not the voice is particularly easy to recognize, and it becomes possible to instantly input voice that is easy to recognize by the device by devising the method of voice input. The device can then reliably and reliably recognize this. Therefore, erroneous recognition can be effectively prevented and voice input can be performed quickly and efficiently. Moreover, unlike the conventional signal sound information that indicates whether recognition has been performed or not, the recognition status is informed, making it easier to devise vocalizations to obtain high recognition accuracy and to avoid unstable recognition judgment processing. It is possible to avoid this. and,
As described above, since variable control of the output form of the signal tone is simple, effects such as the ability to manufacture the device easily and at low cost can be achieved.

Note that the present invention is not limited to the above embodiments. For example, it is also possible to variably control the signal sound by dividing it into several stages depending on the reliability. Further, as the reliability information, other judgment factors used in speech recognition processing may be used, and it is also possible to map this to the reliability. In short, the present invention can be implemented with various modifications without departing from the gist thereof.

[Brief explanation of drawings]

FIG. 1 is a diagram showing the relationship between signal tones, input voices, and the reliability of their recognition results, FIG. 2 is a schematic configuration diagram of an apparatus according to an embodiment of the present invention, and FIGS. 3 and 4 are respectively for signal tones. FIG. 3 is a diagram showing the relationship between output format and reliability. 1...Speech recognition unit, 2...Register, 3...Similarity averaging circuit, 4...Register, 5...Signal sound generation circuit.

Claims (1)

[Claims] 1. Means for outputting a signal sound to prompt voice input;
A means for recognizing and processing a voice signal inputted in synchronization with the signal sound, and the recognition reliability of the recognition processing performed by this means at the present time, and within a period extending back a certain number of recognition processing times from the present time. 1. A speech recognition device comprising: output form variable means for varying the output form of the signal sound in accordance with an average recognition reliability obtained by averaging recognition reliability obtained by recognition processing performed on the signal sound. 2. The speech recognition device according to claim 1, wherein the output form variable means varies at least one of the frequency, amplitude, time width, and waveform of the signal tone.