JPS6029800A - Voice analysis system - 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] [Technical Field of the Invention] This invention converts an analog signal including an audio signal recorded on a magnetic tape or the like into a digital signal by performing analog-to-digital conversion (hereinafter referred to as AD conversion). The present invention relates to a voice analysis method in which only voice data is extracted from a digital signal and stored in a storage device.

[Technical background of the invention]

Recently, various devices have been developed that have a digital voice memory and realize a voice response function using the digital voice memory. The first step is to analyze the audio signal and convert it into digital data. The voice analysis method is used to perform the overcasting process.

By the way, some devices are capable of pronouncing various sentences by combining many words, and in this case, many words are recorded and sent in the form of an analog signal to a recording/playback device using a recording medium such as magnetic tape. This involves recording, analyzing each of these analog signals, and converting them into digital data.

However, such analog signals have unrecorded bands containing noise between each audio signal, and when performing audio analysis, it is necessary to remove these unrecorded bands and extract only the audio signal.

Therefore, in the conventional voice analysis method, the operator listens to the reproduced sound and determines the moment when the voice starts and ends.

[Problems with background technology]

However, this work relies on the operator's judgment, requiring several trials and errors, and creating a large amount of audio data requires a large amount of time, making it extremely inefficient.

[Purpose of the invention]

The purpose of this invention is to eliminate the above-mentioned drawbacks and provide a voice analysis method that can create lean voice data in a single operation by automatically detecting the start and end points of voice. shall be.

[Summary of the invention]

In this invention, an audio signal is converted into a digital signal and inputted in each minimum unit, and it is determined whether the digital signal is silent data or sound data, and the sound data continues for a first predetermined time t' or more. If so, storage of the digital signal is started, and either the silent data or the sound data that does not continue for 11 or more times in the first predetermined time period is stored for a second predetermined time period t2 (
>tx), the storage of the digital signal is terminated.

[Embodiments of the invention]

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

FIG. 1 is a block diagram showing an embodiment of a speech analysis device to which the method of the present invention is applied. The voice analysis device of this embodiment includes an operator console 1, a recording/playback device control section 2. Recording and playback device in which a large number of voices are recorded 3. AD conversion unit that converts audio from analog to digital 4. DA converter 5° speaker 6 for digital-to-analog conversion of audio data of digital signals; audio data start/end point detector 7. Central processing unit8. Audio memory for storing one audio data9. Program memory 10. It includes an audio data storage device 11 that stores a large amount of audio data.

Figure 2 shows an example of a large number of nine sounds recorded in the recording/playback device 3, where the horizontal axis represents time T and the vertical axis represents time T.
It shows the volume of sound that changes in response to changes in.

The audio component 12 is scattered in the time period from time T1 to time T2, and the noise component 13 and non-speech component 14 are scattered.

Now, in the block diagram shown in FIG. 1, when the operator console 1 instructs voice analysis, a control signal is output from the recording/playback device control section 2 and the recording/playback device 3
is driven and audio playback begins. Here, the audio reproduced by the recording/reproducing device 3 corresponds to, for example, FIG. 2. This playback starts from time To. The playback output of the recording/playback device 3 is transmitted to the AD converter 4. The AD conversion unit 4 sequentially converts the playback output of the recording/playback device 3 into digital signals, and the digital signals are transmitted to the start/end point detection unit 7.
It will be done. The start/end point detecting section 7 sequentially inspects the digital signal converted by the AD converting section 4, and detects the starting point of the audio component 12, that is, the time T! Detect. Detection of the start point of the audio component 12 by the start point/end point detection section 7 is performed according to the flowchart shown in FIG.

The start point/end point detection section 7 first inputs the digital data converted by the AD conversion section 7 in minimum units and judges whether or not this data is silent data. This is done by comparing with Threshold E/l/Drepel. That is, the input data -! If it is smaller than the threshold level, it is determined to be silent data, and if it is larger than the threshold level, it is determined to be sound data. In this judgment, if the input data is silent data, the process returns to the beginning of this flowchart and repeats the above operation again. In addition, in this judgment, if the input data is not silent data, that is, if it is sound data, then it is judged whether or not this sound data lasts for more than time t1, and if it does not last for more than time t1, then this Returning to the beginning of the flowchart, the above operation is repeated, but if it continues for more than time t1, it is determined that the audio data has started, and storage of the digital audio data in the audio memory 9 is started. That is, the time point T1 detected by the start point/end point detection section 7 is transmitted to the central processing section 8, and the audio data from the AD conversion section 4 is stored in the audio memory 9 from time point T1.

This storing operation in the audio memory 9 is continued until the end point of the audio component 12 is detected by the start/end point detection unit 7. The end point detection operation of the audio component 12 in the start/end point detection section 7 is performed according to the flowchart shown in FIG.

The audio data from the AD conversion unit 4 is added to the start/end point detection unit 7 in minimum units, and the start/end point detection unit 7 first determines whether or not this input data is silent data. If it is not silent data, that is, if it is sound data, then it is determined whether or not this active data lasts for more than time t1, and the sound is more than 91 pairs of time t□. If it is red, it is determined that the audio component is still 12, the time count is cleared by 1 for the audio memory portion of the audio data, and storage of the audio data for the audio memory portion is continued. Further, in the case where the input data is determined to be silent data in determining whether or not the input data is silent data, and when the above-mentioned sound data is determined to be silent data at time 1. In determining whether or not the sound data lasts for more than 12 hours, if the sound data does not last for more than 12 hours, time measurement of time t2 is started. Then, it is determined whether or not the time t2 has been reached, and if the time has not been reached, the process returns to the beginning of this flow and the above operation is repeated again. However, if the time t2 has been reached, the audio component 12 is Decide that it is finished.

Based on this determination, the central processing unit 8 finishes storing the audio data, and the recording/playback device control section 2 controls the recording/playback device 3.
In other words, in this embodiment, the AD converter 4 serially processes the digital data that has been AD converted, skips silent data and sound data that does not last at the constant value t1, and sets the constant value t1. It finds the beginning of the audio component 12 that exceeds the limit and starts storing the audio data in the audio memory 9, and when silent data and sound data that last only at a constant value t1 are read at t2, it is determined that the end point is reached and the audio data is stored in the audio memory 9. I am trying to finish storing the audio data to.

In addition, when the operator console 1 issues an audio playback command, it outputs the analyzed audio data from the audio memory 9, outputs the data to the DA converter 5, and reproduces the analyzed audio data from the speaker 6. If it is good, the audio data in the audio memory 9 is stored in the audio data storage device 11 and used for the next audio data analysis.
[Effects of the Invention] As explained above, according to the method of the present invention, waste-free voice data can be created in a single operation, which has the advantage of enabling efficient voice analysis work.

[Brief explanation of drawings]

Fig. 1 is a diagram showing an example of a speech analysis device to which the method of the present invention is applied, Fig. 2 is a graph showing an example of speech, and Fig. 3 is a graph showing the starting point of riverbank data in the same embodiment. Flowchart for detection, 1...O(rate console), 2...Recording and playback device control unit, 3...Recording and playback device, 4...AD conversion unit, 5 Audio data storage device, 12
...Audio component, 13...Noise component, 14...Silent component. Representative Patent Attorney Kensuke Norichika (J, L or grade) Figure 3 Figure 4

Claims (1)

[Claims] The audio signal is converted into a digital signal and inputted in each minimum unit, and it is determined whether the minimum unit required signal is silent data or audio data, and if the audio data continues for a first predetermined time of 11 or more, the i-voice data is input. The storage of the digital signal is started by determining the start point of the digital signal, and either the silent data or the sound data that does not continue for 11 or more predetermined times is determined as the second predetermined end point, and the digital signal is stored. A voice analysis method that terminates signal storage.