JPS61189600A - Voice storage reproducer - 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 Industrial Application] The present invention relates to a voice storage and playback device in a system that provides voice mail services, voice output services, and the like.

In particular, it is an object of the present invention to provide an audio storage and playback device that performs compression processing on silent sections with little deterioration in audio quality.

[Prior art] A conventional audio storage and playback device is disclosed in Japanese Patent Application Laid-open No. 1983-
20056. 58-40598゜58-40599
．． 58-40963゜58-40964. 58-
50597゜58-75360. 58-96446
゜No. 58-208917 discloses that the audio signal is digitized and stored in a storage device, but at that time, it is determined whether the audio signal is sound or silent, and the silent part is The system only stored information on the temporal length of silent parts, and the signal for silent parts was discarded and compressed to improve the efficiency of the storage device.

There is a time lag because it takes time to determine whether speech is voiced or silent, and it is difficult to detect low-level speech levels at the beginning and end of words. It has been pointed out that by adding a delay, it is possible to reduce deterioration in audio quality by performing compression processing that goes back in time from the time when the presence or absence of audio was determined.

(References: 1, “Silent interval compression processing in voice communication systems that allow delay” Journal of the Institute of Electronics and Communication Engineers (B)
July 1983 VoL, J66B-NO, 7,908 pages, 915 pages: 2, "Voice storage service device configuration" Research and Practical Report Vol. 33, No. 5 (1984), page 939 Published May 22, 1980 , Published by Nippon Telegraph and Telephone Public Corporation Research and Development Headquarters, 3. Research and Practical Application Report on "Voice Storage Method for Business Offices" No. 3
Vol. 3, No. 5 (1984), p. 967. ) The silent section compression process published in one of the above-mentioned references will be explained with reference to FIG.

(a) shows the audio signal, and (b) shows whether each block of A, B...K, which is the detection unit time of audio presence/absence, is active (V) or silent (P). It is a pattern,
(C) is a diagram in which blocks G and H of the silent section are compressed.

That is, the audio signal shown in (a) is divided into voice/silence detection unit times, and it is determined whether or not there is audio between them, and the result is the pattern shown in (b). Instead of storing the pattern shown in (b) as is in the storage device, as shown in (C), it is stored starting from block A, which is a silent section immediately before block B-H, which is a sound section. Among blocks F to 7r, which are silent sections, block F, which is a silent section immediately after a sound section, is accumulated as is, and above the blocks G and block H, which are accumulated after that, are the silent sections immediately before the sound section. block y of section
The silent section is compressed by overwriting ir and the blocks after block J in the sound section.

In this way, the silent sections A and F immediately before and after the sound section
, T blocks are treated and stored in the same way as voiced sections because there is a strong possibility that word beginnings and word endings exist there.

[Problems to be Solved by the Invention] However, as shown in FIG. 6, it is not necessarily sufficient to simply store blocks A and I immediately before a sound section.

This is because it is difficult to specify from which part of block A or Fr the voice becomes active, and there is a possibility that the beginning of the speech exists in a block even earlier than blocks A or I. cannot be eliminated. Therefore, it is inevitable that time-cut distortion of the audio occurs, resulting in a large deterioration of the audio quality. This problem is particularly noticeable when using a simple voice detection means that determines the presence/absence of voice only based on the level of the voice signal.

In order to prevent the beginning of a speech from being cut off, the amount of silent parts that should be accumulated immediately before a voice becomes active should vary depending on the noise level, the level of the audio signal, and the method of determining whether the voice is active or silent. However, in the conventional technology, as shown in FIG. 6, the object of accumulation in a silent section is one block of the silent section immediately before a voice becomes active;
There was a problem that complete determination of silence was not possible.

[Means for Solving the Problems] The present invention solves the problem by accumulating the silent section immediately before the voiced section over a plurality of silent sections, in order to avoid such speech cutting. .

[Effect] By doing this, not only one block of the silent section immediately before the voiced section but also the previous silent section can be stored, so cutting at the beginning of the speech can be avoided and the speech can be improved. It is possible to avoid generation of time cut distortion and prevent deterioration of voice quality.

[Example] An embodiment of the present invention will be described with reference to the drawings.

FIG. 1 is a circuit configuration diagram of a portion related to audio storage, and FIG. 2 is a circuit configuration diagram of a portion related to audio reproduction.

1 is an input/output terminal for audio codes, 2 and 3 are logical switches that are interlocked with each other, and a buffer memory 4. and 5 are mutually accessed. Reference numeral 6 designates a control device, 7 a transfer device, 8 a main storage device including, for example, a magnetic disk, 9 a voice detection circuit, and 10 a programmable address counter including an ending silence block counter and a repetition counter.

In FIG. 1, the digitized voice codes applied to the input/output terminal 1 are sequentially stored in the buffer memory 4 or 5 at the locations indicated by the programmable address counter 10.

The switching of the buffer 7 memory 4 or 5 to which data is written is performed by the switch 2 based on a command from the control device 6. Assuming that the switch 2 is in the state shown in FIG. 1 and the voice code has been written to the end in the buffer 7 memory 4,
An interrupt is generated to the control device 6 to notify it. In response to this interrupt, the control device 6 switches the switch 2 to the buffer memory 5 side to continue writing the voice code. Meanwhile, during this time, the control device 6 instructs the transfer device 7 to transfer the audio code written in the buffer memory 4 to the main storage device 8. When the switch 2 is switched to the 5 side, the switch 3 is switched to the buffer memory 4 side (switched).

As a result, switches 2 and 3 operate to select different buffer memories. The voice detection circuit 9 measures the level of the voice code applied to the input/output terminal 1, the number of polarity inversions, etc., determines whether there is a sound or not, and outputs the result.

FIG. 3 is a diagram for explaining an example of the structure of data stored in the buffer 1 memory 4 or 5, and will be explained below using this diagram.

(a) shows the audio signal before being encoded with the audio code applied to input/output terminal 1, and (b) shows whether the audio signal shown in (a) is voiced (V) or silent (V). P) Block A divided by unit time for detection of voice presence/silence;
B,...N, Q, R.

This is a hail pattern for S. (C) shows the internal configuration of buffer memory 4 or 5, and (
d) shows the case of overwriting what is shown in (C), and (e
) indicates the case where writing is continued in the area shown in (d), and (f>) has a structure in which each block of A-3 includes the flying light information 31, repetition information 32, and audio code writing range. (CJ) indicates the output of the voice detection circuit 9, which is at a high level when the voice signal shown in (a) has a sound, and at a low level when there is no sound. A predetermined time delay is generated with respect to the audio signal (a).

Figures 4A, 4B, 4C, and 4D are examples of flowcharts showing the operation during audio accumulation. Go through any of the flows from ko to end.

Now, a voice code is to be stored at the address of the buffer memory 4 or 5 indicated by the programmable address counter 10 (5TEP 40, FIG. 4A).
If the address counter 10 does not indicate the last address of one block (STEP 41), the programmable
The contents of the address counter 10 are incremented by 1 (STEP 42).

When the programmable address counter 10 indicates the last address of one block (STEP 41), the relative destination block number is set to O (STEP 43). here,
The relative jump destination block number is a number indicating how many blocks ahead of the block currently being written, and is indicated by the jump light information 31. Therefore, it is determined whether the current state is in the m block shown in FIG. 3, in the n block, or in the sound section (STEP 44).
.

If it is in the sound section, the output of the voice detection circuit 9 is at a high level (STEP 65, Figure 4B), and the switch 2
The repetition information 32 (FIG. 3(f)) of the block being accessed in the side buffer 7 memory 4 or 5 is set to 0. ! :L (
STEP70. (FIG. 4c), the light flying information 31 of the block is set to O, 5TEP71>, and a block ahead by the number of relative jump destination blocks from the block next to the block currently being accessed is accessed (STEP72). That is,
Access the next block. At this time, the block is connected to the buffer memory 4 or 5 connected to the switch 2 side.
If it does not exceed the memory size of (STEP 73)
, programmable address counter 10 is connected to switch 2
Set it to the beginning of the audio code writing range (FIG. 3(f)) of the block of buffer memory 4 or 5 on the side (STE
P74).

To summarize the above operation, when the audio signal is in the active state and the programmable address counter 10 indicates the last address of one block, the 5TEP40-41-4
3-44-65-70-71-72-73-74, the repetition information 3゛2 and the flying light information 31 of the block being accessed are set to O, and the voice code writing range of the next block is Set the programmable address counter 10 at the beginning. If the voice code has been written to the end of buffer 7 memory 4 or 5 (STEP 7)
3) The contents of the buffer memory 4 or 5 are transferred to the main memory 8.
Switches 2 and 3 are switched to transfer the data to 5TEP40-41-43-44- (STEP75).
Pass 65-70-71-72-73-75-76-74.

, it is determined that the audio signal is silent, and the audio detection circuit 9
When the output of becomes low level (S, TEP65), the current state is in m block (Figure 3 (C)) (STEP
66), 5TEP40-41-43-44-65-6
Low 67-70-71-72-73-74 or -73
-After 75-76-74, approximately m block frontage-
Check whether the level remains and 5TEP40-4
171-72-73-74 or -73-75-76-
Pass through 74.

Here, the count of m is made by the ending silence block counter included in the programmable address counter 10. During this time, the speech code is written into the μ-order buffer, memory 4 or 5, as shown in FIG. 3(C). Here, if the programmable address counter 10 does not indicate the last address of one block (the end of the audio code writing range, FIG. 3(f)), the 5TEP
40-41-42 and indicates the last address, the above 5TEP40-41-71-72-7
Pass through 3-74 or -73-75-76-74.

If the output of the voice detection circuit 9 becomes high level before the m blocks are written (STEP50), the current state is in the sound section (STEP61>, 5TEP71-72-73-
74 or -73-75-76-74, and the current state is processed as a sound section.

If the low level continues after m blocks have been written, the current state is in n blocks (Figure 3(d)).
, 5TEP53) The number of disappeared blocks is 1 and 5TE
P54>, memorize the block next to the m block (the block next to the block in Figure 3 (C)) 5TEP55)
, set the relative jump block number to -n5TEP56
), STEP40-41-43-44-50-51-
0102 which passes through 72-73-74 and is smaller than m
9 minutes, return the programmable address counter 10 and overwrite the previously written location as shown in FIG. 3(C) (FIG. 3(d)), 5TEP40-41-43.
Pass through -44-80-81-82-72-73-74. Then m-01029 minutes (blocks F and G in Figure 3)
>The part corresponding to the ending of the word is accumulated. If the output of the voice detection circuit 9 (Fig. 3 (g)) continues to be low level (STEP 80), it will be overwritten many times within the n block and the output will be 5TEP40-41-42 or 5TEP40-
Pass through 41-43-80-81-82-72-73-74 or -82-83-72-73-74. That is, the latest 01029 minutes of audio codes are stored in this n block. At the same time, the programmable address counter 10 stores the next block position to be accessed before going back n blocks, that is, the next block position after the block in FIG. 3(C).5TEP55
), roughly the block that disappeared due to overwriting (H,'
Count the total number Q of I', J) (STEP 8)
1). Here, Q is counted by a repetition counter included in the programmable address counter 10. .

If the output of the voice detection circuit 9 becomes high level (S
(TEP 80), the current state is a sound section (STEP
84), the next block to be accessed is the block following m block (the block to the right of the block in Figure 3 (C)) or not. Store the number Q of disappeared blocks in the repetition information 32 of the block (block K in (d)) (STEP 88 or 91) In this way, 5T
EP40-41-43-44-80-84-73-74
Or pass through -85-91-72-73-74. By processing in this way, the latest n-1 blocks (L, M, N) are stored as a portion corresponding to the beginning of a word.

Or, although it is different from the flow shown in FIGS. 4A to 4D,
From 070 onwards, the latest block in the past (block G in (d)) is deleted from the repetition information 32.

The number Q of destroyed blocks may be stored. In this case, the latest 01029 minutes (K, L, M, N) are stored as the part corresponding to the beginning of the word.

In addition, the access is moved by blocks to the block position that was scheduled to be accessed next (Fig. 3(,e')S
TEP86), the flying light information 31 of each block includes (A, B, . . . F, G, L,
After a certain block has been played, the flash light information 31 indicating the block to be played is sent to the next address so that the block is played back (in the order of M, N, Q, R, S, etc.). Memorize (STEP 8'7, 89.90
>.

When the period during which the output of the voice detection circuit 9 is at a low level is less than m blocks, and while the output of the voice detection circuit 9 is at a high level (STEP 65), the flying light information 31 contains the block to be accessed next. A specific value, for example O, is written as information indicating that is adjacent (STEP
71>,<A specific value, for example O, is written in the repetition information 32 as information indicating that the block is not repeated (STEP 70).

Here, 5TEP62, 63.64 is n block (third
This is to extend the m block part so that the switch 2.3 will not be switched after the part shown in Figure 4 (d) exceeds the size of the buffer 7 memory 4 or 5. be.

Next, the operation during audio reproduction will be explained with reference to FIG. 2 and FIGS. 5A and 5B showing flowcharts.

Every sampling period of the audio code output to the input/output terminal 1, one of the flows from [start of reproduction] to [end] is passed.

5TEP100 when the programmable address counter 10 does not indicate the most aggressive address of one block.
-101-"110, the contents of the programmable address counter 10 are incremented by 1 from the buffer memory 4 or 5 indicated by the programmable address counter 10, and are sequentially read out and output to the input/output terminal 1. When the programmable address counter 10 indicates the last address of one block (STEPlol), check the contents of the <repeat counter, and if it is not O (STEP102), decrease the <repeat counter by 1 (STEP103).
), 5T returns the programmable address counter 10 to the beginning of the audio code writing range of the block being accessed.
EP104), 5TEPI00-101-102-10
3-104, and this operation is repeated to reproduce the corresponding block.

When the repetition counter is O (STEP 102)
, the access is moved to a predetermined block according to the value of the flying light information 31 (5TEP105), and it is checked whether the next block exceeds the size of buffer memory 4 or 5 (5TEP106). At the same time, the contents of the main memory 8 are transferred to the buffer memory 4 or 5 (STEP 10).
7) , Move the access to the first block of buffer memory 4 or 5 5TEP108) , Repetition information 3 of the block being accessed in buffer memory 4 or 5
Set 2 on the counter repeatedly (STEP 109)
. Then, the programmable address counter 10 is set to the beginning of the audio code writing range of the block being accessed (STEP 104). In this way, 5TEP10
9-104 or -106-107-108-109-
104.

Although the flow is different from that shown in Figures 5A and 5B,
In this repeated reproduction, it is also possible to repeatedly reproduce a predetermined code sequence such as a silent pattern or a noise pattern such as a dorsal am. In this way, the data is played back in the chronological order in which it was stored.

It goes without saying that, although not shown in the flowchart, each variable required for processing is set prior to processing both during storage and playback.

In the above description, a serial number indicating the time order of the blocks stored in the flying light information 31 may be set and used. Although the repetition information 32 shows the number Q of erased blocks, it is also effective to set an upper limit to this value so that a long period of unnecessary silence is not reproduced as it is. It is also possible to roughly embed the flying light information 31 and repetition information 32 only in necessary blocks together with a flag indicating the existence of the flying light information 31 and repetition information 32. Further, the flying light information 31 and the repetition information 32 may not be embedded in each block, but may be grouped together in the directory or header part of one message.

[Effects of the Invention] As is clear from the above description, since there are a plurality of blocks (n blocks) at the beginning of a word that has been determined to be a silent portion, truncating the beginning of a word can be almost completely prevented. Roughly speaking, the amount accumulated as the beginning of a word is not fixed to one unit of time for detecting speech presence/absence, but depends on the speech presence/non-speech judgment method, the input level to the device, and noise. It has the advantage of being able to be set freely depending on the level etc.

Therefore, according to the present invention, it is possible to avoid the problems caused by the delay in the output of the voice detection circuit 9 or the difficulty in detecting the low voice level part at the beginning and end of a word. This has the extremely great effect of economically realizing an audio storage and playback device with little deterioration in quality.

[Brief explanation of the drawing]

Figures 1 and 2 are circuit configuration diagrams of parts related to audio storage and audio playback, respectively, and Figure 3 is a buffer memory 4.
, 5. Figures 4A, 4B, 4C, and 4D are flowcharts for the audio storage operation, and Figures 5A and 5B are for the audio playback operation. The flowchart in FIG. 6 is a diagram for explaining the prior art. 1... Input/output terminal, 2.3... Switch, 4, 5.
... Buffer memory, 6... Control device, 7... Transfer device, 8... Main storage device, 9... Audio detection circuit, 1
0...Programmable address counter, 31...
- Flying light information, 32... Repeated information.

Claims (3)

[Claims] (1) A voice detection means that divides a code sequence of a digitized audio signal into blocks each consisting of a fixed number of codes and determines whether each block is a silent block or a sound block, and a storage device that stores the code sequence. A voice storage device comprising a word-final silence block counter and a repetition counter, wherein the count of the word-final silence block counter during storage indicates that silence blocks do not continue beyond a predetermined number of m blocks. and when the voice detecting means detects a sound block, the code sequence is stored in the storage device, and when played back, the code sequence is played back in the order in which it was stored, and when the m blocks are stored, silent blocks continue. When the ending silence block counter detects this, it counts the number of blocks exceeding the m blocks, and stores the code sequence of the latest n blocks whose number is smaller than the predetermined m blocks. The audio is stored in the device, and the code sequences of the exceeded blocks past the n blocks are discarded, and when played back, the code sequences of the exceeded blocks are repeatedly reproduced. Storage and playback device. (2) The audio storage and playback device according to claim 1, wherein the block may include information on the number of repeated playbacks and playback order information. (3) The repeated code sequence is the code sequence of the oldest block in terms of time among the n blocks, or the code sequence of the latest block in the past than the n blocks;
Alternatively, the audio storage and playback device according to claim 1, which uses a predetermined code sequence.