JPH0728486A - Voice compression device - Google Patents

Abstract

PURPOSE:To provide a voice compression device which detects an unnecessary silence interval, eliminates the interval from voice information and performs a store control without going through a host CPU. CONSTITUTION:The device consists of a power detecting circuit 11 which puts voice signals into blocks and outputs the power of a single frame, an averaging circuit 12 which outputs an average power of m frames, a discriminating circuit 13 which compares the average power value and a threshold value to discriminate sound and silence intervals and outputs the result, a compensation circuit 14 which only discriminates a silence interval when silence intervals continue for n times and outputs the result, a delay circuit 15 which synchronizes the timing of inputted voice signals and the output of the circuit 14 and a storage means 16 which takes the output of the circuit 15 as input data and makes the output of the circuit 14 as a write control signal input.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice data compression apparatus, and more particularly to a voice compression apparatus for compressing and storing voice data in an answering machine, voice mail (voice mail) or the like.

[0002]

2. Description of the Related Art A voice storage device is installed in an exchange,
Various voice mail services have been realized, such as accumulating messages by voice, delivering the message to the other party at the desired time on behalf of the sender, or allowing the other party to receive the message at a convenient time. .

Due to the fact that a high compression rate has been made possible with the progress of research on the voice compression system, and the voice compression circuit has been integrated into a circuit, the voice compression storage device can be used not only in exchanges but also in home telephones and the like. Will also be built in,
High-performance telephones that store messages by voice and play a role as an answering machine and transfer messages have become common.

Various methods are used to compress the audio information in the audio compression device, and there are, for example, information compression for band-compressing a digital audio signal, information compression for a silent section, or a combination of both.

As the information compression in the exchange, since the use of the input information is not always clear, a compression method based on the complete restoration of the input information is adopted. For example, there are methods of compressing information for silent sections, such as replacing the silent section with a silent code, or deleting the information of the silent section. Has been adopted.

Here, in order to describe a voice compression device built in a telephone or the like, a process for deleting unnecessary information of a silent segment with respect to a silent segment will be described with respect to a conventional voice compression device.

First, an unnecessary silent section will be described. The silent section is divided into two types: a silent section that is absolutely necessary for accurately transmitting the message content, and an unnecessary silent section that can be deleted for the purpose of transmitting the message.

Concretely speaking, the unnecessary silence section is, for example, in the voice generated by the caller, such as a long silence between sentences, or selection of the caller's words in the middle of a sentence. There is a silent section that occurs due to thoughts or hesitation.

[0009] This kind of silent section has no problem for message transmission even if it is not reproduced. Therefore, when such an unnecessary silent section is detected, it is deleted without being stored in the memory.

A conventional voice compression apparatus (hereinafter referred to as "conventional example 1") will be described with reference to FIG. FIG. 2 shows a schematic configuration of a voice compression device incorporated in a telephone or the like, which deletes information of unnecessary silent sections from the silent sections.

In FIG. 2, the digitized audio signal is input from the input terminal 20, and the power detection circuit 21
The input audio signal is divided into blocks and a power value per frame time is output.

The judgment circuit 23 inputs the power value output from the power detection circuit 21, compares the power value with a threshold value for judging a voiced section / silent section, and outputs the result.

The control means 27 outputs the output of the decision circuit 23,
The digital audio signal input from the input terminal 20 is input, and the writing control signal to the storage means 26 and the digital audio signal are output.

The storage means 26 stores the digital audio signal in the memory based on the write control signal for the memory output from the control means 27.

With reference to FIG. 5, the operation of the first conventional example will be described focusing on the control means 27.

The power detection circuit 21 inputs a digital audio signal from the input terminal 20, extracts a power value for one frame time (here, 20 msec), and outputs it.

The judgment circuit 23 compares the threshold value (T) for judging the voiced section / silent section with the power value output from the power detection circuit 21, and if (power value) ≧ T, the voiced section. Then, a logical value “1” is output, and if (power value) <T, it is determined to be a silent section and a logical value “0” is output.

The control means 27 inputs the judgment result of the judgment circuit 23, monitors it, and outputs a write control signal to the memory.

FIG. 5A shows an example of the output signal of the decision circuit 23 of FIG. As shown in FIG. 5 (A), when the output of the determination circuit 23 is "0", that is, when there are five or more silent sections in succession, the control means 27 does not output the write control signal (see FIG. 5 (B)). Activate.

The control means 27, at the time when the output signal of the judgment circuit 23 changes to "1" again in FIG.
The write control signal is activated and output from "0" twice before, and the input digital signal is output in synchronization with the timing of the write control signal.

The storage means 26 receives the digital voice signal and the write control signal output from the control means 27, and stores the digital voice signal in the memory when the write control signal is in the activated state.

When the digital audio signal stored in the memory is restored, the contents of the memory are only sequentially decoded.

Next, as a voice compression / accumulation device, for example, in Japanese Patent Laid-Open No. 3-825 (hereinafter referred to as "conventional example 2"), a block is provided by adding a circuit between a buffer memory and an input / output terminal.・ The size can be set independently of the buffer memory, magnetic disk, etc., and the information about voiced blocks and silence blocks is inserted into the voice code so that the computer control unit does not need to know anything. A storage and reproduction method has been proposed.

In the conventional example 2, it is determined whether a voice signal is a voiced block or a voiceless block for each block, the coded sequence is stored as it is for the voiced block, and the voiced block relates to the number of consecutive flags and voiceless blocks. Only the information is stored in the storage device, and at the time of reproduction, the flag is detected from the code sequence extracted from the storage device, the silent block is found, and the silent block is inserted based on the information about the number of silent blocks attached to the flag. For voiced blocks, a voice storage / playback system is disclosed in which a code sequence is extracted from a storage device and played back.

[0025]

In the above-mentioned conventional example 1, the control means 27 which is a part of the host CPU is used,
Judgment and detection of silent sections that do not require storage is performed.

The control means 27 monitors the judgment output of the judgment circuit 23, counts the number of "0" s indicating a silent section, judges an unnecessary silent section, and outputs a write control signal based on the judgment result. It is necessary to generate and also temporarily store the audio signal corresponding to this section.

Further, the control means 27 outputs a write control signal and a digital signal corresponding to the write control signal to the storage means 26 to control the storage of audio information in the memory.

As described above, in the conventional example 1, the host CPU performs all the processes from the determination of the unnecessary silence section to the control of the storage means, and therefore has a problem that a large load is applied to the CPU.

Further, in the second conventional example, a configuration is disclosed in which a predetermined flag and a continuous number of the silent sections are stored in the storage device for all the silent sections including the silent sections which need not be stored. However, the conventional example 2 is different from the conventional example 1 in that the silent section is divided into two types, and unnecessary silent sections are deleted to enable efficient use of the memory space. Is. Furthermore, in the second conventional example, a flag control circuit and the like in the storage and reproduction system are required.

Therefore, the present invention solves the above problems,
A voice compression device for detecting an unnecessary silent section when storing a voice signal and deleting the unnecessary silent section from the voice information, so that the storage control can be performed without a host CPU with a simple circuit configuration. It is an object of the present invention to provide a voice compression device.

[0031]

In order to achieve the above object, the present invention is a voice compression apparatus for compressing and storing a voice signal, comprising an input terminal for inputting a digitized voice signal,
A power detection circuit that blocks the audio signal input from the input terminal into a frame composed of a predetermined number of samples and outputs a power value in frame units, and an output of the power detection circuit is input, and an average power value of m frames An averaging circuit that outputs the averaging circuit, a determination circuit that inputs the output of the averaging circuit, compares the average power value with a threshold value, determines a voiced section / silent section, and outputs the determination result, Input the output of the judgment circuit and the judgment result of the silent section is n (n> 0)
A correction circuit that determines a silent period only when the sound is consecutive and outputs the determination result; a delay circuit that is connected to the input terminal and that synchronizes the timing of the input audio signal with the output signal of the correction circuit; and the correction circuit. And a storage means for receiving the output of the delay circuit as the write control signal and using the output of the delay circuit as the input data.

[0032]

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

In FIG. 1, an audio signal is input from the input terminal 10, and the power detection circuit 11 detects the power value of the input digital signal per frame.

The averaging circuit 12 temporarily stores the power values of the m frames in a buffer and outputs the power average value among the m frames. In this embodiment, one frame has 20
msec.

The judgment circuit 13 compares the power average value between the m frames and the threshold value to judge the voiced section and the silent section, and outputs the judgment result.

The correction circuit 14 corrects the judgment result of the judgment circuit 13 and outputs a write control signal to the storage means 16.

The delay circuit 15 receives the digital audio signal from the input terminal 10 and synchronizes with the write control signal output from the correction circuit 14.

The storage means 16 is a memory for accumulating the output signal of the delay circuit 15 in accordance with the write control signal.

Next, with reference to FIG. 4, an explanation will be given of the process of extracting an unnecessary silent section in which the transmission message is not changed even if it is deleted.

FIG. 4 (A) shows an example of a voice signal waveform of a transmission message, and FIG. 4 (B) shows a range of the voice signal which is stored in the memory and a voice signal which is not stored in the memory. The range to be deleted is shown.

As a process of extracting an unnecessary silent section, a time point when a voice signal is near 0 level continuously for a predetermined time (TE) for a predetermined time (TE) or more is set as a start point, and a predetermined time (when a voice signal waveform rises again) This section is targeted for deletion, with the time point before TS) as the end point.

In the present embodiment, the process of detecting a removable silent section from the input digital audio signal will be described below.

The input signal is processed in blocks by one frame. The power value per one frame time is calculated in the power detection circuit 11, and the averaging circuit 12 stores the power value in m buffers and outputs the average power value between consecutive m frames.

The averaging can reduce the impact of the power value of each frame. Further, by averaging the power values of the frames, a large number of voice signals can be captured, and this can be identified before the voice signal waveform actually rises.

Therefore, the averaging circuit 12 can detect the rising change of the voice signal waveform at a predetermined time point before the voice signal waveform actually rises.

The decision circuit 13 determines whether the voice signal has a voice /
This is a circuit for determining a silent section, and compares a threshold value for determining whether a sound signal is detected / not detected with the input average power value. , And a logical value “0” is output in the silent section. The threshold value for determining the presence / absence of a voice signal has a predetermined margin.

The correction circuit 14 immediately outputs the output of the determination circuit 13 as it is so as to immediately respond to the change from the silent section to the voiced section.

The correction circuit 14 monitors the output of the judgment circuit 13 for n frames when the voiced section is changed to the silence section, and only when n frames are continuously in the silence section, the correction circuit 14 does not perform the operation. A detection signal is output (n =
5).

The operation of this embodiment will be described with reference to FIGS.

The power detection circuit 11 receives the digital audio signal (see FIG. 3) from the input terminal 10 in synchronization with the sampling cycle.
(See (A)) is input and the power value for one frame time is calculated.

The power detection circuit 11 squares the sample value of the audio signal and sums up for one frame, for example,
This sum of squares is output as the power value of one frame.

That is, as shown in FIG. 3 (B), the power detection circuit 11 uses the total sum An (n = 1, 2,
...) are sequentially output as power values.

The averaging circuit 12 sequentially inputs the output of the power detection circuit 11, writes and saves the input An for k frames in a buffer (not shown), and outputs the average power value Bn (n = 1, 2) for m frames. , ...) is output (see FIG. 3C).

As shown in FIG. 3C, the averaging circuit 12 outputs the output sequence An (n = 1, 2,
, Are sequentially shifted, and a moving average Bn is output by averaging a fixed number (m = 6 in this embodiment). The averaging circuit 12 attenuates the high frequency component by taking the moving average of the power value An, and makes it possible to determine the majority of the audio signal waveform.

In the decision circuit 13, the power average value B
n is compared with the threshold value T, and if Bn ≧ T, it is determined that the signal is detected (voiced section) and the logical value is “1”, and if Bn <T, it is determined that no signal is detected (the silent section) and the logical value is “1”. 0 "is output.

In FIG. 3D, the level of a predetermined threshold value in the judgment circuit 13 and the input power average value Bn are shown.
An example showing the relationship with (n = 1, 2, ...) Is shown, and further, an output signal string of the determination circuit 13 that is output by comparing these is shown.

The correction circuit 14 inputs the detection / non-detection signal which is the output of the judgment circuit 13, and in the case of the change from the non-detection to the detection (that is, the change from "0" to "1"), the detection signal as it is. Is output.

When the detection changes to the non-detection (that is, the change from "1" to "0"), the non-detection signal is not immediately output, and the non-detection is continuously performed for 5 sections, that is, the silent section is continuous. The undetected signal “0” is output only when the interval is reached.

The delay circuit 15 synchronizes the digital audio signal input from the input terminal 10 with the detected / undetected signal output from the correction circuit 14, and outputs the signal after delaying it by a time corresponding to k frames. .

The memory 16 inputs the detected / undetected signal which is the output signal of the correction circuit 14 and the digital audio signal which is the output signal of the delay circuit 15, and uses the detected / undetected signal as a write control signal to the memory. The digital audio signal output from the delay circuit 15 is stored and stored in the memory.

In this embodiment, according to the processing described above,
When restoring the digital audio signal stored in the memory,
It is only necessary to sequentially decode the stored memory contents.

As described above, in this embodiment, the averaging circuit,
With the configuration of the correction circuit and the delay circuit, it is possible to reduce the load of the above-described conventional example such as the process of discriminating the silent section by the host CPU and the control of the storage unit, and to directly connect to the storage unit without the host CPU. There is.

In this embodiment, if the input digital audio signal is band-compressed, a further compression effect of the memory space in the storage means can be expected.

In the averaging circuit 12 of the present embodiment, the number m of frames subjected to moving average processing is appropriately determined according to the length of one frame section, the characteristics of the audio signal to be processed and the system.

Further, in the present embodiment, it is determined that there is an unnecessary silence section in the case where there are five silent sections in succession, but this is determined depending on the length of one humm section, etc.
Also, it may be adaptively determined by the system. Further, the threshold value for determining the voiced section / silent section may be appropriately determined according to the characteristics of the system, or may be variable according to background noise or the like.

[0066]

As described above, according to the present invention, an averaging circuit, a correction circuit, and a delay circuit are added to a conventional voice compression device for eliminating unnecessary silence intervals when storing a voice signal. With such a configuration, there is an advantage that writing of the digital audio signal to the storage means can be controlled without going through the host CPU, the load on the CPU is reduced, and the speed of the system is increased.

Further, the averaging circuit in the voice compression apparatus of the present invention makes it possible to identify the rising edge of the input voice waveform in advance, alleviate the sudden impact generated in the input signal, and correct it. The circuit monitors the silent section for a certain period of time, determines that the silent section does not need to be stored only when the silent section continues for a predetermined section, and deletes the input signal without storing it in the memory space. It has an advantage of enabling efficient use. further,
In the present invention, if the band compression is performed on the input digital signal, a further memory space compression effect can be expected.

Further, the power circuit, the averaging circuit, the judgment circuit, the correction circuit and the delay circuit in the present invention have a simple structure which can be realized by the arithmetic processing and the delay element, and can be easily realized by the digital signal processor. .

Further, in the present invention, the detection / non-detection signal of the correction circuit is directly input as the write control signal to the memory, and the digital signal which is the output of the delay circuit can be used as the input data to directly connect to the memory circuit. There is also the effect.

[Brief description of drawings]

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

FIG. 2 is a block diagram showing a configuration of a conventional example.

FIG. 3 is an explanatory diagram for explaining the operation of the embodiment of the present invention. (A) is a figure which shows a digital audio signal waveform.
(B) is a diagram showing a signal sequence of the power output value An of the power detection circuit 11. (C) is an explanatory view showing the relationship between the power average value Bn and the power value An of the averaging circuit 12. FIG. 3D is an explanatory diagram showing the relationship between the input signal and the threshold value and the relationship between the comparison output and the judgment circuit 13.

FIG. 4 is an explanatory diagram for explaining the operation of the embodiment of the present invention. (A) is a figure which shows a digital audio signal waveform.
(B) is an explanatory view showing a memory accumulation and deletion range.

FIG. 5 is a diagram for explaining the operation of a conventional control circuit. (A) is a diagram showing an output signal sequence of the determination circuit 23.
(B) is a timing diagram showing a write control signal (activated by “1”) of the memory corresponding to (A).

[Explanation of symbols]

 10, 20 Input terminal 11, 21 Power detection circuit 12 Averaging circuit 13, 23 Judgment circuit 14 Correction circuit 15 Delay circuit 16, 26 Storage means 27 Control circuit

Claims (2)

[Claims] 1. An audio compression apparatus for compressing and accumulating audio signals, comprising: an input terminal for inputting a digitized audio signal; and an audio signal input from the input terminal is divided into frames each consisting of a predetermined number of samples. A power detection circuit that outputs a power value in units of frames; an averaging circuit that inputs the output of the power detection circuit and outputs an average power value of m frames; and an input of the output of the averaging circuit, A determination circuit that compares the average power value with a threshold value to determine a voiced section / silent section and outputs the determination result, and an output of the determination circuit is input and the determination result of the silent section is n.
A correction circuit that determines a silence section only when (n> 0) times consecutively and outputs the determination result, and a delay circuit that is connected to the input terminal and that synchronizes the timing of the input audio signal and the timing of the output signal of the correction circuit. And a storage means for inputting the output of the correction circuit as a write control signal and using the output of the delay circuit as input data. 2. The voice compression apparatus according to claim 1, wherein the correction circuit immediately outputs a detection signal of a voiced section when the voiceless section is changed to a voiced section.