JPS6031373B2 - Digital call audio introduction method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a digital call voice insertion method, and particularly to a method that can suitably prevent freeze-outs that occur at the beginning of a conversation when voice is inserted.

Digital call voice insertion method (DigtaISpeech)
Interpolation (abbreviated as DSI) originally means that in a telephone conversation, it is necessary to listen to what the other person is saying, so the average time each caller actually speaks is about half of the call time. It was developed by focusing on the redundancy that exists in calls, where there are pauses between syllables, words, or phrases even while speaking, which do not need to be transmitted as audio. It is used for effective utilization of.

In other words, if you instantaneously observe a large number of telephone lines on which calls are being made, you will find that voice signals exist only on approximately half or less of them.
The I device 1 is provided with a large number of DSI input trunks TK, ~TKn connected to the telephone line, and a smaller number of DSI output channels CH, ~CHm, and the digitized audio signal in each input trunk is is divided into unit blocks (for example, 4 ms) each consisting of a predetermined number of series of sample values, and a voice detector determines for each unit block whether or not voice exists within the unit block. Only the unit blocks for which it is determined that voice exists are extracted from each input trunk, inserted into each other's empty space, reorganized and transmitted to the output channel. In this case, for effective use of the transmission path, the number of input trunks n and the number of output channels m
It is desired that the ratio (n/m), that is, the DSI gain, be as large as possible. This DSI gain is limited by the average activity rate, which is the proportion of the detected audio portion to the total audio signal, e.g. 2-2.5 for an average activity rate of 40%.
The OSI gain is set to . However, when inserting audio, the number of active input trunks that contain audio in the input trunks exceeds the number of output channels, and some active input trunks cannot be connected to output channels. There are cases where the sound is cut off midway through.

This state is a freeze-out, and in order to avoid this freeze-out as much as possible, the DSI gain must be set on the safe side, and the efficiency of effective use of the transmission path is reduced accordingly. Furthermore, since the setting of the DSI gain is based on a statistical value called the average operating rate, the only way to completely eliminate freeze-outs is to set the DSI gain to 1, but this is unreasonable considering the purpose of adopting the DSI method. It is. Therefore, in reality, the DSI gain is set by allowing a slight freeze-out occurrence rate of, for example, 0.5% to 0.01%. However, if the beginning of the conversation is cut off due to freeze-out, it will have a negative impact on the conversation and is extremely important in terms of subjective evaluation.

The present invention aims to provide a digital call voice insertion method that can prevent such freeze-outs, which are unfavorable from a subjective evaluation, and this purpose is achieved by utilizing hangover in voice detection.

Therefore, the principle of the present invention will be explained in advance with reference to FIG.

The audio detector uses the magnitude of the level of the input signal as the first criterion, and for example, the intra-block power Pw given by the following equation {1} or {21 is determined by the threshold P as shown in the following equation {3i.
Basically, if it exceeds Th, it is determined that there is audio in this unit block. PTh<Pw
......{3} However, N' is the number of samples forming a unit block, and xj is the amplitude value of each sample. However, since determining that audio does not exist means not transmitting the entire unit block, the audio detector is designed to minimize the possibility of erroneously determining that audio is not present in the unit block where audio is present. Once detected, for example, there are several unit blocks that do not satisfy equation {3'
Even if the audio continues for several tens of seconds, these are considered to be unit blocks in which audio exists.

This is a hangover, and as shown in FIG. 2, the hatched area in the audio detection signal b for the audio waveform a is the hangover region. Note that FIG. 2c shows an allocation control signal for allocating the unit block corresponding to the "1" state to the output channel based on the voice detection signal b. Such hangovers occur not only in simple speech detectors that use equation {3} as the only criterion for judgment, but also in speech detectors that employ other characteristics such as the number of zero crossings or complex judgment logic. There is. On the other hand, when we examine the relationship between hangover areas and freezeouts, we find that if a freezeout occurs because the hangover area is quite long, there is a possibility that one of the active input trunks is in the hangover area. Extremely high.

Therefore, if the part of the hangover area that corresponds to the occurrence of freezeout is forcibly and temporarily excluded from channel allocation, freezeout in areas other than the hangover area can be prevented. An embodiment of the present invention will be described below.

FIG. 3 is a block diagram of an embodiment, in which 2 is a buffer memory, 3 is a voice detector, 4 is a channel assignment controller, 5 is a channel assignment status signal generator, 6 is a multiplexer, 7 is a block diagram of an embodiment; is a hangover detector. The digitized audio signal S in each input trunk (input trunk TK, ~TKn) is divided into unit blocks and sent to the audio detector 3.
and stored in the buffer memory 2.
Among the large number of unit blocks stored in the buffer memory 2, the unit blocks for which the voice detector 3 determines that voice exists are outputted to the output channels (CH, ~CHm) under the control of the channel allocation controller 4. Ru. For example, in the PCM signal of Iso HZ sampling, the above unit block is 32
The sample time is 4ms. The voice detector 3 shown as an example includes a power calculation circuit 3a, a zero-crossing counting circuit 3b, a temporary detection circuit 3c, a hangover time setting circuit 3d, and a voice detection determination circuit 3e. The power calculation circuit 3a is based on the previous formula'
The intra-block power Pw of the input signal S for each unit block of 4 ms length is calculated based on (1) and (2), and the drop crossing number counting circuit 3b counts the number of zero crossings for each unit block. The temporary detection judgment circuit 3c determines that a sound exists in the unit block when the intra-flock power Pw satisfies the condition of the front leg, or when the zero crossing number Zc satisfies any of the conditions of the following equation (4'). Temporary detection signal v with strong possibility
d"1" is output, and in other cases, a temporary detection signal vd "0" indicating that the possibility of the presence of voice is low is outputted. ZC
≦ZThl. rZT ground≦ZC……{4}
(However, ZTh, ZTh2) The hangover time setting circuit 3d outputs a hangover time signal Tm indicating a predetermined period (for example, 20 dishes s) from the time when the temporary detection signal vd changes from "1" to "0". .

The voice detection determination circuit 3e determines, for example, vd “1” based on the continuity of the temporary detection signal vd in the same input trunk.
If 2 or more of the 3 unit blocks following the unit block are vd "1", it is determined that voice exists in the unit block, and the voice detection signal VD is determined.
Outputs “1”. Furthermore, even if this continuity is not satisfied, it is assumed that voice is present within the time determined by the hangover time signal Tho, and VD "1" is output. Therefore, only when the continuity condition is not satisfied and it is outside the hangover time, the voice detection signal VD "0" which determines that there is no voice is output. FIG. 2b shows the audio output signal VD output in this manner. The hangover detector 7 detects a low-level unit block within the hangover area among the unit blocks determined by the audio detector 3 to have audio (VD "1"), and the power comparison circuit 7a and a gate circuit 7b.

The power comparison circuit 7a determines for each unit block whether the output Pw from the power calculation circuit 3a satisfies the following equation {5-. The gate circuit 7b outputs a hangover area signal HO indicating only those within the period determined by the hangover time signal Tho among the unit blocks determined to satisfy the equation {5}. Pw<PTh, (However, PTh.<PTh,)
...[5} The channel assignment controller 4 is the audio detector 3
The unit block in which it is determined that audio is present based on the audio detection signal VD “1” is normally assigned to the output channel.
When a freeze-out occurs, unit blocks that are determined to be within the hangover area by the hangover area signal HO are excluded from among the unit blocks allocated to this button and are allocated to the output channel.For example, the allocation control circuit 4a and the exclusion command It consists of circuit 4b.

The exclusion command circuit 4b inputs the freeze-out state signal FOS outputted from the allocation control circuit 4a and indicating that a freeze-out occurs, and the hang-over area signal HO from the hang-over detector 7, and receives the freeze-out state signal FOS and the hang-over area signal HO from the hangover detector 7. An allocation exclusion signal OA is output that causes unit blocks within the channel allocation to be excluded from channel allocation. The allocation control circuit 4a receives the audio detection signal VD and the allocation exclusion signal OA as input, and outputs the read control signal R only to unit blocks of VD "1" other than those specified by the allocation exclusion signal OA.
C and an allocation control signal AC. The buffer memory 2 is read out to each output channel of the unit block to be transmitted by the output control signal RC. Allocation control signal A
C indicates the correspondence between the input trunk and output channel of each unit block allocated by the read control signal RC, and is encoded by the channel allocation status signal generator 5 and output as an allocation status signal AS for each output. It is sent out to the transmission path through the multiplexer 6 together with the channel information. Therefore, when a freeze-out occurs, some or all of the unit blocks in the original hangover area of any input trunk are excluded from channel assignment, so that the unit block where voice truly exists (input trunk)
will not freeze out.

In addition, input trunks that have been once excluded from channel allocation will be automatically reset when the freezeout is released, the hangover area ends, or the intra-block power Pw within the hangover area exceeds the threshold PM.
If the voice detection signal VD "1" is set, channel assignment is performed again. By the way, many input trunks may become hangover regions when a freeze-out occurs.

In preparation for this case, it is preferable to provide some form of priority for allocation exclusion in the allocation control circuit 4a. In this embodiment, assignments are removed in order from the input trunk with the lowest number. As specifically explained above with the examples,
According to the present invention, it is possible to prevent freeze-out at speaking faces, which are important for subjective evaluation.

[Brief explanation of the drawing]

Fig. 1 is a diagram conceptually showing the DSI system, Fig. 2 is a diagram for explaining hangover, a, b, and c indicate a voice waveform, a voice detection signal, and an allocation control signal, respectively. FIG. 1 is a block diagram showing an example of the present invention. In the drawing, 2 is a buffer memory, 3 is a voice detector, 3a
3b is its power calculation circuit, 3b is a hangover time setting circuit, 4 is a channel assignment controller, 4a is its assignment control circuit, 4b is an exclusion command circuit, 6 is a channel assignment status signal generator, 6 is a multiplexing layer, 7 is a A hangover detector, 7a is its power ratio circuit, and 7b is a gate circuit. Figure 1 Figure 2 Figure 3

Claims (1)

[Claims] 1 Divide the digitized audio signal in the input trunk into a predetermined number of unit blocks, which are a series of sample values, and select only the unit blocks for which audio is determined to exist by the audio detector by the channel assignment controller. A hangover detector is provided for detecting a unit block in which it is determined that voice is present due to a hangover of the voice detector and the power within the block is less than or equal to a standard value. , a digital call voice insertion method characterized in that if a freeze-out occurs during voice insertion by the channel allocation controller, a unit block detected by the hangover detector is excluded from allocation to an output channel. .