JPS62102291A - Automatic diagnosing apparatus for voice input/output 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 [Field of Application of the Invention] The present invention relates to an audio input/output device, and more particularly to a method for diagnosing input functions in an audio input/output device.

[Background of the invention]

Conventionally, methods for testing the hardware of a speech recognition device include, for example, inputting digital test data from a circuit after A/D conversion and checking whether the pattern matching result is as expected. There is a method of inputting a standard tape, registering it, recognizing it, and checking it.

In the former case, test equipment and preparation are required to connect a device that inputs digital data.

Tests are complicated. In the latter case, it is also necessary to prepare, connect, and operate a tape recorder, etc. Furthermore, the general method for speech synthesis devices is to actually output speech and check it by a human.

Furthermore, as disclosed in Japanese Patent Application Laid-open No. 56-89799, there is a known device equipped with an error recognition cause analysis unit, but analysis is difficult and it is difficult to accurately know the error. There is a drawback that it cannot be done.

[Purpose of the invention]

ゝ-1 ゛An object of the present invention is to provide an automatic diagnostic device that can easily test the recognition function of a voice input/output device.

[Summary of the invention]

A speech recognition device generally performs a matching process between a pre-registered pattern and a uttered pattern to recognize speech. For this reason, the environment at that time, the psychology of the person speaking, and the physical condition of the speaker may cause differences from the registered patterns, which will affect the recognition rate. This is particularly true for less skilled vocalists, although it is far greater.

When evaluating the functionality of the entire device in conjunction with the human factor, it is necessary to always determine whether a decline in recognition rate or malfunction is caused by the device itself or the human factor. There is. Therefore, it is necessary to be able to easily check the recognition function. Furthermore, the audio input/output device has a function of synthesizing and outputting audio. Therefore, this synthesized sound is looped in the device itself, the synthesized sound is registered in the recognition device, and then the recognition module 11
・Comparison of various 17 recognition results allows you to dynamically check 17 recognition functions. Additionally, the audio output function can be checked at the same time, allowing comprehensive diagnosis.

[Embodiments of the invention]

Embodiments of the present invention will be described below with reference to FIGS. 1 and 2. First, an overview of the operation of the audio input/output device will be briefly described.

The voice input from the microphone is amplified by the analog interface section 5 and input to the voice recognition section 18. Thereby, the speech recognition unit performs recognition processing and outputs the recognition result. The control unit 2 takes in this result, processes it, and sends it to the speech synthesis unit 3.
Give output instructions, etc. to.

The voice synthesis section 3 synthesizes the instructed voice and outputs the synthesized voice to a speaker, a Herad set, etc. via an analog interface section.

Here, a switch and a level conversion circuit are provided in the analog interface section 5 so that the output is connected to the microphone input. FIG. 2 shows the inside of the analog interface section, where 52 is a changeover switch and 51 is a level conversion circuit.

By operating this switch, the audio output and audio input are connected internally, allowing the control unit to output any audio and have it recognized. The level conversion circuit 51 adjusts the speech synthesis output to the appropriate level of the speech recognition input.1 The amplifier or attenuator is selected depending on the input and output levels of both.1 The normal switches 52 and 53 are 502 and 504. It is connected to the side and performs voice response processing. Now, in order to perform a recognition function check, switches 52 and 53 are switched to the 501 and 503 sides. In this state, the control unit instructs the recognition unit to register the input voice. On the other hand, the control section instructs the synthesis section of the word to be tested and outputs the speech. As a result, the audio created within the audio input/output device is registered. Next, the recognition state is instructed to the recognition unit, and the synthesis unit is made to output the same registered word. Here, the output speech is recognized, and the recognition unit outputs the recognized words, similarity, etc. to the control unit.

Generally, synthesized speech and recognized speech go through an analog @path, so the degree of similarity calculated using digital values is not the same even for the same word. However, recognition results with extremely high similarity can be obtained.

Therefore, the control unit diagnoses the recognition unit by checking whether the similarity value in addition to the recognition word falls within the criteria for determining that the recognition function is normal.

By providing such a circuit, the recognition function of the voice input/output device itself can be easily checked, and since 5 synthesized sounds are also output, the voice synthesis section can be checked at the same time.

In the above embodiment, a circuit for looping one synthesized sound to the recognition section is provided in the analog interface section, but a loop connector may be connected to an external terminal or the like. That is,
As shown in Figure 3, connect the loop connector with a built-in resistor to the external connector for audio input/output. This loops the audio output to the microphone input side, which is equivalent to switching the switch, and the recognition function can be checked in the same way.

〔Effect of the invention〕

According to the present invention, the recognition function can be easily and automatically checked using the output function of the voice input/output device.
It becomes easier to maintain hardware and isolate troubles caused by human factors.

[Brief explanation of drawings]

FIG. 1 is an overall configuration diagram of the audio input/output device, FIG. 2 is a schematic internal configuration diagram of an analog interface section, and FIG. 3 is an internal connection diagram of a loop connector. DESCRIPTION OF SYMBOLS 1... Speech recognition section, 2... Control section, 3... Speech synthesis section, 4... Memory section, 5... Analog interface section. 51...Level conversion circuit, 52.53...Switching switch, 54.55...Amplifier, 10...Loop connector.

Claims (1)

[Claims] 1. In an audio input/output device consisting of a speech recognition section, a speech synthesis section, a control section, a memory section, and an analog signal interface section, a means is provided for inputting the speech output from the synthesis section to the recognition section, and An automatic diagnosis device for a voice input/output device, characterized in that the function of a voice recognition unit is checked by voice.