JPH11175096A - Voice signal processor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To realize a voice signal processor automatically and quantitatively evaluating a sentence understanding degree. SOLUTION: A voice signal outputted from a DAT 1 is inputted to a voice recognition device 12, and an evaluating sentence (voice signal) analog outputted from this device 12 is radiated from a speaker 11, and on the other hand, after the evaluating sentence is converted to the evaluating sentence data through an A/D converter 5, is ID added by an ID adder 6 to be supplied to a sentence data comparator 7, and is compared with the reference sentence data inputted through a reference voice coder 2, the A/D converter 3, the ID adder 4 one sentence each. Then, the sentence data comparator 7 stores the compared result at every sentence, and percentage displays the compared result on a display device 10 based on the stored contents. Thus, the sentence understanding degree is evaluated automatically and quantitatively.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech recognition technology, and more particularly, to a speech signal processing device capable of automatically evaluating sentence intelligibility.

[0002]

2. Description of the Related Art Heretofore, a speech editing device and a speech input type sentence processing device using a speech recognition technology have been known, and this type of technology is disclosed in, for example, JP-A-61-23809.
No. 8 and JP-A-62-191896. In the audio editing apparatuses disclosed in these publications, a silent section between words extracted from an evaluation audio signal source is interpolated by a low-level noise signal, and the original audio is detected by keeping the audio detection threshold at low-level noise. Signal sections are normally detected to improve the speech recognition rate. on the other hand,
In the voice input type sentence processing apparatus, voice patterns of a plurality of persons are registered in advance, the input voice patterns are matched with the registered patterns, and a registered voice pattern having the highest matching rate is automatically extracted to perform voice recognition. It has become.

[0003] In order to evaluate the sentence intelligibility using the speech recognition technology realized by each of the above devices, for example, as shown in FIG. 4, a sentence (speech) recorded on a DAT or a tape recorder is used. Is reproduced, input to an audio signal processing device to which the audio recognition technology is applied, and subjected to audio recognition, and the result is output again as audio from a speaker, thereby determining sentence intelligibility based on human hearing. .

[0004]

However, in the above-described audio signal processing apparatus, since the sentence intelligibility is ultimately evaluated based on human hearing, only a qualitative evaluation can be performed. There is a problem that evaluation cannot be performed.

The present invention has been made in view of such circumstances, and has as its object to provide an audio signal processing apparatus capable of automatically and quantitatively evaluating sentence intelligibility.

[0006]

In order to achieve the above object, according to the first aspect of the present invention, a converting means for converting a reproduced audio signal into a first text data serving as a reference, First code adding means for adding an identification code to the output, generating means for generating second sentence data for evaluation from the audio signal, and second generating means for adding an identification code to the output of the generating means. 2 sign adding means, and the first
And sentence comparing means for determining sentence intelligibility by comparing the sentence of each of the first and second sentence data to which the identification code has been added via the second sign addition means for each corresponding identification code. It is characterized by.

According to the second aspect of the present invention, the first and second sign adding means include:
It is characterized in that an identification code is added during a silent section. According to the third aspect of the present invention, the sentence comparing means does not match the comparison result at a predetermined ratio in all bits forming the sentence data corresponding to one sentence. In some cases, the sentence intelligibility is determined to be "good", and when it is higher than that, it is determined to be "bad".

According to the present invention, an ID is added to the reference sentence data and the evaluation sentence data for each sentence to be compared so that the comparison judgment can be easily performed, so that the sentence intelligibility can be automatically and quantitatively determined. Can be evaluated.

[0009]

DESCRIPTION OF THE PREFERRED EMBODIMENTS An audio signal processing apparatus according to the present invention can be applied to various systems for voice recognition. Hereinafter, an audio signal processing device according to an embodiment of the present invention will be described as an example with reference to the drawings. (1) Configuration FIG. 1 is a block diagram showing the configuration of one embodiment according to the present invention. In this figure, reference numeral 1 denotes a DAT in which a reference text (audio signal) is recorded. 2 converts the analog signal format audio signal input from the DAT 1 into digital signal format voice data (hereinafter referred to as reference text data) by a predetermined encoding method, and converts the reference text data in advance. This is a reference speech encoder that decodes and outputs a reference text signal by a determined decoding method. In addition,
It is assumed that the reference speech encoder 2 used here has the same performance as the speech encoder used in the speech recognition device 12 (to be described later) as a test device.

Reference numeral 3 denotes an A / D converter, which converts the reference text signal coded / decoded by the reference speech coder 2 again into reference text data and outputs it. Also, this A / D
The converter 3 determines whether there is sound or no sound in FIGS.
It is configured to perform under the conditions shown in FIG. That is, when the input analog signal is within the level indicated by the point A in FIG. 2A, the A / D converter 3 determines that it is a silent section, and performs a silent section during the A / D conversion. Output a LOW level to indicate. 4 is an ID adder. As shown in FIG. 2C, when the output of the A / D converter 3 receives the next reference sentence data after a silence period of 3 seconds or more continues, as shown in FIG. An ID is added to the head of the text data. The reference text data to which the ID has been added in this manner is supplied to a text data comparator 7 described later and stored as reference text data.

Reference numeral 8 denotes a switch controller, which includes switches S4 and S5 for selecting the number of test sentences and start switches S2 and S3.
It is composed of In the case of this embodiment, the number of test sentences is 20.
The switch S4 is operated to select a sentence, and the switch S5 is operated to select 50 sentences. The sentence data comparator 7
The reference text data stored in advance is compared with the text data for evaluation input from the ID adder 6, and the result is output to the comparison result storage 9. Note that the evaluation sentence data referred to here is data obtained by outputting the voice signal output from the DAT 1 from the voice recognition device 12.

In the sentence data comparator 7, FIG.
The sentence data is compared by the method shown in (b). Here, the part a in FIG. 3B is compared with FIG.
Although there is an error, the human auditory sense cannot identify an error of about 1 bit, so that one sentence data (1
A determination criterion is set to “good” if all the bits of the sentence) are within 0.1%, and “no” if there is an error of more than 0.1%. . The comparison result storage unit 9 which receives the comparison result for each sentence from the sentence data comparator 7 stores the comparison result for each sentence, and displays the comparison result on the display unit 100 based on the stored content. .

According to such a configuration, first, the number of test sentences is selected by the switch controller 8, the switch S1 is set to the test side, and then the switch S of the switch controller 8 is set.
By operating either S2 or S3, a test for evaluating sentence intelligibility is started. The voice signal output from the DAT 1 is input to a voice recognition device 12, and an evaluation sentence (voice signal) analog output from the device 12 is output from a speaker 11.
, While being converted into evaluation text data via the A / D converter 5, the ID is added by the ID adding device 6, and then supplied to the text data comparator 7, where The sentence is compared with the reference sentence data input through the steps 2 to 4 one by one. As described above, the sentence data comparator 7 stores the comparison result for each sentence, and causes the display 10 to display the comparison result on a percentage basis based on the stored contents. This makes it possible to automatically and quantitatively evaluate sentence intelligibility.

As described above, according to the present invention, the ID is added to each of the reference text data and the evaluation text data for each sentence to be compared so that the comparison judgment can be easily performed. In addition, it is possible to quantitatively evaluate the sentence intelligibility.

[0015]

According to the present invention, comparison and judgment can be easily performed by adding an ID for each sentence to be compared to the reference sentence data and the evaluation sentence data. You can evaluate sentence intelligibility.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of an embodiment according to the present invention.

FIG. 2 is a diagram for explaining a data configuration of reference text data.

FIG. 3 is a diagram for explaining a comparison operation between reference document data and evaluation document data.

FIG. 4 is a diagram for explaining a conventional example.

[Explanation of symbols]

 Reference Signs List 1 DAT 2 Reference speech encoder 3 A / D converter 4 ID adder 5 A / D converter 6 ID adder 7 Text data comparator 8 Switch controller 9 Comparison result storage 10 Display 11 Speaker 12 Voice recognition apparatus

Claims (3)

[Claims] A converter for converting an audio signal to be reproduced into first text data serving as a reference; a first code adding unit for adding an identification code to an output of the converter; Generating means for generating second sentence data for evaluation from the program, second code adding means for adding an identification code to the output of the generating means, and via the first and second code adding means An audio signal processing device, comprising: sentence comparison means for determining sentence intelligibility by comparing sentences of the first and second sentence data added with the identification code for each corresponding identification code. 2. The first and second sign adding means,
2. The audio signal processing device according to claim 1, wherein an identification code is added during a silent section. 3. The sentence comparison means judges that the sentence intelligibility is “good” when the comparison result does not match at a predetermined ratio in all bits forming the sentence data corresponding to one sentence, An audio signal processing device characterized by sometimes determining "No".