JPH0540494A - Composite voice tester - Google Patents

Abstract

PURPOSE:To decide whether a composite voice is satisfactory or not, based on a result of matching with a pattern sequence of an inputted composite voice by registering in advance the pattern sequence of a standard composite voice. CONSTITUTION:The tester is constituted of a voice input part 1 for digitizing an inputted composite voice, a voice analyzing part 2 for extracting a feature pattern sequence of an input voice, a pattern memory part 3 for storing in advance a standard feature pattern sequence, a pattern matching part 4 for executing DP matching between the input and the standard feature pattern sequence, a result deciding part 5 for comparing similarity of a matching result with a prescription and deciding it, and a whole control part 6 for executing control of each constituting unit and a communication to a host.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech synthesizer automatic test apparatus.

[0002]

2. Description of the Related Art Conventionally, in the inspection and testing of a speech synthesizer, it is necessary for an inspector to actually listen to a synthesized speech output signal of a word or a phrase determined as a final test and check that the synthesized speech has no error. I have confirmed.

[0003]

However, this conventional inspection method has a problem in that the inspector must perform the inspection on the device under test all the time, which increases the number of inspection steps.

Further, in the test of the data memory for storing the segment data of the speech synthesizer, the signal processor for synthesizing, etc., although the software for automatic diagnosis is introduced, the final result is obtained. Since it requires manpower only to check the output of the synthesized voice, the throughput of the entire inspection process is not improved.

The present invention has been made in view of the above-mentioned conventional circumstances, and therefore, an object of the present invention is to provide a novel synthetic speech tester capable of solving the above-mentioned problems inherent in the prior art. To provide.

[0006]

In order to achieve the above object, a synthetic speech tester according to the present invention comprises a speech input section for digitizing a synthesized speech signal to be input and determining the start and end points,
A voice analysis unit that extracts an acoustic characteristic pattern sequence from the input digitized synthetic speech, a pattern memory unit that stores and registers the characteristic pattern sequence, a characteristic pattern sequence that is registered in advance, and an input synthetic speech. A pattern matching unit that performs pattern matching with the characteristic pattern sequence extracted from the result pattern, a result determination unit that determines the validity of the input synthesized speech from the similarity between the pattern sequences obtained as a result of matching, And an overall control unit for controlling the constituent units.

[0007]

BEST MODE FOR CARRYING OUT THE INVENTION The present invention will now be described in detail with reference to the accompanying drawings with reference to the accompanying drawings.

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

Referring to FIG. 1, an output synthesized voice from a voice synthesizer (not shown), which is an object to be tested, is input to a voice input section 1 through a microphone 7, where the synthesized voice is digitized and the whole process is started. Edge detection is performed. The synthesized voice digitized by the voice input unit 1 is used as the voice analysis unit 2
And is converted into a feature pattern series by acoustic analysis such as mel cepstrum analysis. Pattern memory unit 3
Is a memory for storing the characteristic pattern series extracted by the voice analysis unit 2 in advance. The pattern matching unit 4 executes DP matching between the characteristic pattern series of the input synthesized speech obtained by the speech analysis unit 2 and the characteristic pattern series registered in the pattern memory unit 3.
The similarity between the pattern sequences obtained by the DP matching of the pattern matching unit 4 is compared with the prescribed similarity in the result determination unit 5, and it is determined that the input synthesized speech showing the similarity higher than the prescribed is acceptable. Then, the overall control unit 6 is notified of the result.

The overall control unit 6 registers a characteristic pattern sequence in the pattern memory unit 3 based on an instruction from the host 8,
The pattern matching unit 4 specifies the characteristic pattern series used for the matching template, controls the operation sequence of each constituent unit, and the like.

The operation of this embodiment will be briefly described below.

When the output synthesized speech of the speech synthesizer is inspected according to this embodiment, it is necessary to register the standard characteristic pattern series of synthesized speech in advance. For this purpose, first, a standard synthesized voice is input to the voice input unit 1 through the microphone 7 in units of words or phrases, digitized and the start and end are determined, and then converted into a feature pattern sequence in the voice analysis unit 2. And stores it in the pattern memory unit 3. At this time, the overall control unit 6 receives the phrase number corresponding to this standard synthesized voice from the host 8 and registers it in the pattern memory unit 3 together with the characteristic pattern series.

When the output synthetic speech of the speech synthesizer is inspected, the overall control section 6 first receives the phrase number of the synthetic speech input from the host 8 and designates it to the pattern matching section 4. Next, the synthesized voice input to the voice input unit 1 through the microphone 7 is digitized, and the start and end are determined, and the voice analysis unit 2 converts it into a characteristic pattern sequence and transfers it to the pattern matching unit 4. The pattern matching unit 4 executes DP matching between the characteristic pattern sequence corresponding to the phrase number designated by the overall control unit 6 and the characteristic pattern sequence sent from the voice analysis unit 2,
The result judging unit 5 calculates the similarity between the pattern series and outputs the result.
Send to.

The result judging section 5 compares the sent similarity with the similarity defined in advance by the overall control section 6, and if it is larger than the specified value, the pass determination is made. Send to. The overall control unit 6 adds the received pass / fail judgment to the corresponding phrase number
And waits for the next instruction from the host 8.

[0015]

As described above, according to the present invention, a standard acoustic feature pattern sequence of synthesized speech is registered in a memory in advance, and the input of the synthesized speech corresponding thereto is obtained. In order to verify the correctness of synthesized speech based on the degree of similarity of the results of pattern matching between feature pattern series, it becomes possible to automatically inspect the speech synthesizer without human intervention by the inspector. The effect is that the number of inspection steps can be reduced.

Further, according to the present invention, considering the high reproducibility and stability peculiar to speech synthesis, the correctness of the synthesized speech output, that is, the correctness of the synthesized speech output, is shown because a very high degree of similarity is shown for a correct synthesized speech. The effect that the quality of the device under test can be determined with high accuracy is obtained.

[Brief description of drawings]

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

[Explanation of symbols]

 1 ... Voice input unit 2 ... Voice analysis unit 3 ... Pattern memory unit 4 ... Pattern matching unit 5 ... Result determination unit 6 ... Overall control unit 7 ... Microphone 8 ... Host

Claims (2)

[Claims] 1. A means for digitizing a synthetic voice input from a microphone, a means for extracting an acoustic feature pattern from a digitized voice signal, a storage means for storing the acoustic feature pattern, and a pre-stored unit. Means for performing pattern matching between the feature pattern being input and the feature pattern obtained from the input voice, means for determining the validity of the input synthetic voice from the similarity of the pattern matching results, It has a control means for controlling each unit, registers a correct characteristic pattern of synthesized speech in advance, and performs pattern matching with the synthesized speech output from the synthesized speech apparatus to be tested to inspect the apparatus to be tested. A synthetic speech tester characterized by performing automatically. 2. The control means further receives a phrase number corresponding to a standard synthesized voice from the host, and registers the phrase number in the storage means together with the characteristic pattern series. The synthetic speech tester described.