JPH03276947A - Interactive automatic answering telephone set - Google Patents

Abstract

PURPOSE:To set each threshold level so that the effect of voice detection error is not made detrimental by providing a threshold level setting means able to set two kinds of voice detection level threshold level for interactive reply and pause compression independently. CONSTITUTION:A threshold level setting means 9A is provided, which is able to set two kinds of threshold levels as interactive reply voice detection threshold level and a pause compression voice detection threshold level independently. Through the constitution above, two kinds of interactive reply threshold level Vth-a and pause compression threshold level Vtn-n are prepared as the voice detection threshold levels Vth to be compared with the voice detection level and they are calculated separately and respectively and the voice detection sensitivity is decreased in order to prevent an event of detecting the voiced sound regardless when the caller makes no utterance in the interactive reply. That is, the threshold level Vth-a is set relatively to a larger value. On the other hand, in the pause compression, in order to prevent the detection of silence regardless when the caller makes utterance, the voice detection sensitivity is increased. That is, the Vth-b is set to a small value relatively. The voice level is detected based thereon.

Description

[Detailed Description of the Invention] [Industrial Application Field 1] The present invention relates to an interactive answering machine that records the caller's voice while exchanging messages between the caller and the telephone when a call is automatically received. In particular, for interactive answering machines that employ a pause compression recording method that compresses the silent sections in which the caller does not speak when recording the caller's business message, we have investigated the dialogue responses and pause compression. This invention relates to an interactive answering machine that realizes both of these functions with good performance.

[Prior Art] In general, interactive answering machines send response messages by
It is divided into a part that reveals the name of the person being called, and a part that states that the called person is away, and each part is sent at regular intervals. When the caller hears the part (■), he or she does not realize that the answering machine has answered, so there is a high probability that the caller will start speaking. However, in order to maintain a natural dialogue between the caller and the telephone, and to obtain more detailed information without making the caller feel uncomfortable, it is best to divide the response message into smaller parts. It is necessary to detect the timing at which the caller has finished speaking, and to synchronize the timing of sending the response messages after ① with the end of the caller's voice.

The end of the caller's utterance can be detected by executing a utterance end detection process using the line signal audio detection means. FIG. 5 shows a specific example of processing. This process is executed immediately after the telephone completes sending the response message. The processing shown in FIG. 5 will be explained below. Note that 81 to SIO indicate each step.

First, check whether the other party starts speaking. Specifically, the timer is reset/started (St),
After the start of this process, even if the silence determination time threshold T has elapsed, utterance does not start, that is, if the determination result of the voice detection means does not indicate the presence of sound (S2.S3
．． S4), it is determined that the other party has no intention of starting vocalization (S5).

If the telephone determines that the other party has no intention of starting speaking, the telephone typically proceeds with the process of sending the next response message.

If it is confirmed in step (S3) that the other party has started speaking, the timer is reset/started (S6), a voice detection process is entered (S7), and the duration of silence that appears while the other party is speaking is monitored. (S8). Vocalization includes silent periods required for breathing, thinking about the content, etc. The duration of such silent periods is generally not very long. On the other hand, if the other party finishes speaking, the silent state continues. Therefore, the utterance end judgment time threshold T @
nf1 is determined, and a single silent period is the utterance end determination time threshold T. Only when the number of utterances continues for 6 or more, it is determined that the other party has finished speaking (S9.5IO). When it is confirmed that the other party has finished speaking, the telephone transmits the next response message. The above is the explanation regarding FIG.

On the other hand, pause compression recording is a technology that does not record the silent periods caused by breathing pauses between vocalizations, thoughts about the content, etc., and stores only the vocal parts on the recording medium. It has the advantage of increasing efficiency and shortening the time required to play back recorded messages.

A typical audio detection means is roughly divided into two parts: a level detection circuit and a comparison section. The level detection circuit is a circuit that extracts the envelope waveform or power information of the input signal. The comparison section stores a voice detection level threshold, compares the output value of the level detection circuit with the voice detection level threshold, and compares the output value of the level detection circuit with the voice detection level threshold. If the output of the level detection circuit is smaller than the voice detection level threshold, the voice detection result is determined to be silent. Therefore, if the voice detection level threshold is set high, only signals with a relatively high level will be determined to be active, and if the voice detection level threshold is set low, signals with a relatively low level will be judged as sound. However, it will be determined that there is a sound.

In addition, it is possible to improve voice detection accuracy by setting the voice detection level threshold in accordance with the level of background noise that changes depending on the surrounding conditions of the caller under line condition 9 (Japanese Patent Application Laid-Open No. 1-307350).

[Problem to be solved by the invention] The caller's voice transmitted via the telephone line is subject to individual differences in the caller's voice level, differences in the telephones used by the caller, and losses in subscriber lines and relay systems. The S/N ratio changes in various ways depending on the difference in the caller's surrounding noise and line noise. Even under these conditions, the voice detection circuit can detect false detections (
However, it is difficult to completely eliminate the occurrence of detection errors. Therefore, even if a detection error occurs, the speech detection circuit should be designed so that the detection error does not have a fatal effect on the dialogue response and pause compression functions.

Here, the effects of detection errors in the voice detection circuit on dialogue responses and pause compression can be classified as follows.

■ If the voice detection result shows sound even though the caller is not speaking due to the influence of line noise or surrounding noise of the caller: Dialogue response The dialogue response algorithm Since it is determined that the utterance has not finished, the next response message transmission will not start no matter how long it takes. In addition, if the caller does not make any utterances and a single noise is detected as sound, the telephone determines that the caller has started speaking and has immediately finished speaking. The telephone proceeds to play the next response message even though the person is not speaking.

- Pause compression The pause compression process records the signal during this period even though the caller is not speaking.

■ If the caller's voice level is extremely low or the line loss is extremely large, and the voice detection result is silent even though the caller is speaking: ・Dialogue response Dialogue response The algorithm determines that the caller is in a silent state without starting to speak, and sends the next response message after the silence determination time T mugon has elapsed since the response message was sent.

- Pause compression Pause compression processing does not record even though the caller is speaking. When this part is played back, the caller's voice is dropped, making it difficult to understand what the caller is saying.

The following is a breakdown of which detection errors have a fatal impact on each process: dialogue response and pause compression.

When the first call detection error (2) occurs, the next response message is not reproduced even though the caller has finished speaking, and the process hangs up. If a single noise is considered to be a voice when the caller does not speak, the telephone automatically plays the next response message, causing a conflict between the caller's voice and the telephone's response message transmission. There is a strong possibility.

When detection error (2) occurs, a certain period of time T mugon necessary for silent determination is required from the end of sending a response message to the start of sending the next response message. In other words, this is a type of interactive answering machine that sends out response messages one after another at regular intervals without detecting the caller's voice, although there is a possibility that the telephone's response message transmission may interrupt the caller's voice. The operation is equivalent to .

From the above, detection error (2) has a more fatal impact on dialogue response processing. Therefore, when designing a voice detection circuit, the voice detection threshold should be set high.
It is safer to reduce the probability of detection error ■.

When a detection/output error ■ occurs, originally silent parts are also recorded, so the entire utterance of the caller is recorded, although the efficiency of using the recording medium is reduced. Therefore, it is possible to listen to all the content uttered by the caller when playing back the recorded content.

When the detection error ■ is uttered, the caller's utterance is omitted from the recording, so even if the recording is played back, the caller's utterance cannot be heard.

From the above, it can be said that the detection error (2) has a more fatal influence on the pose compression process. Therefore, when designing a voice detection circuit, it is safer to set the voice detection threshold to a lower value to reduce the probability of occurrence of detection error (2).

As described above, for dialog responses, setting the voice detection level threshold to a higher value reduces the probability of falsely detecting noise as voice, and for pause compression, setting the voice detection level threshold to a lower value. It is desirable to reduce the possibility of voice dropout by setting this value to a certain value. However, these requirements are contradictory, and there is a problem that it cannot be achieved using a single voice detection threshold.

[Means for Solving the Problems] The interactive answering machine according to the present invention has a threshold setting means that can independently set two types of voice detection level thresholds in voice detection processing, one for dialogue response and one for pause compression. It has been established.

[Operation] In the present invention, even if a detection error occurs in the voice detection result, the effect of this detection error on dialogue response and pause compression will not be fatal. Therefore, it is possible to realize both functions with good performance.

[Embodiment] FIG. 1 shows a block configuration of an embodiment of the present invention. 1
2 is a loop opening/closing unit that opens and closes the line; 3 is a telephone call circuit connected to the line via the loop opening/closing unit 2; 4 is the telephone call circuit; A response message sending section connected to the transmitting (T) terminal of the circuit 3; 5 a response message storage section storing a plurality of response messages; 6 a receiving (R) terminal of the telephone circuit 3;
A message recording section 7 is connected to the terminal and records the line signal in the message storage section, 7 is a message storage section for storing the message message, and 8 is connected to the line via the loop opening/closing section 2. Connected signal level measuring section, 9
9A is a control unit such as a CPU; 9A is a threshold setting means; 10
is a timer for measuring the duration of the sound/silence section. It is assumed that the timer 10 automatically counts the time from zero upon receiving an instruction from the control section 9 to reset the timer count. FIG. 2 shows a specific configuration of the signal level measurement section 8 and an example of the interface with the control section 9. In this circuit example, a half-wave rectifier circuit is configured by the operational amplifier OPI and the diode D around it, and a low-pass filter is configured by the operational amplifier OP2 and the resistor R of the capacitor C1 around it. At this time, the output of the operational amplifier OP2 becomes the envelope of the input signal. This envelope is sent to the A/D converter 8.
A converts the signal into a digital signal and inputs it into the control section 9 at regular intervals.

FIG. 3 shows an example of the internal configuration of the response message storage section 5. As shown in FIG.

In this example, three types of dialogue response messages are stored in advance.

FIGS. 4(a) and 4(b) are flowcharts explaining the operation of this embodiment. In addition, (Sll) ~ (S18)
and (S21) to (S40) indicate each step. The operation of the embodiment will be described below with reference to FIG.

When the ringing signal detection unit 1 detects that a ringing signal has arrived on the line (Sll), the control unit 9 operates the loop opening/closing unit 2 to close the loop (S12), and then the first Select the response message 313
), the first response message is sent from the response message sending unit 4 to the line via the communication circuit 3 (S.14).
. When the sending of the response message is completed, the control section 9 starts recording the business message (S15). If the response message number is not the third (=3), add LL to the response message number (S17), and then step (S14)
The process is repeated from and after, and when the response message number reaches the third (=3) in step (S16), the line is closed (
318).

The basic structure of the business message recording processing operation is the same as the utterance end detection processing shown in FIG. 5, and FIG. The compressed recording process is also specifically described. In addition, in the voice detection processing in this figure, the voice detection level is adjusted to match the background noise level so that the voice detection results are not adversely affected by background noise that constantly exists in the background, such as noise around the caller or line noise. The size of the threshold is changed.

Below, the operation of the message recording process is shown in Figure 4 (
This will be explained in detail with reference to b). In this process, the timer is reset/started (S21), and the control unit 9 reads the output of the signal level measurement unit 8 at a constant cycle, and (
S22), the difference Δ between the previous read value and the current read value is calculated for the line signal level (S23). Expressing this in the form of an equation, it becomes Equation 11. However, ■ o is the read value from the signal level measuring section 8 at time n, and V n-+ is the signal level one read cycle ago. It represents the read value from the measurement unit 8.

△＝■ゎ −■n−＋ ・・
(1) The difference value Δ is compared with the voice start detection threshold value V@Lag@ within the control unit 9. Here, V @
ag@ is a threshold value for detecting the start of a voice, and if the difference value Δ is greater than or equal to V sage (S24), it is considered that the voice section has started. That is, at a certain time k,
■, 6,. ≦△U=V, -V,, -・・・・・・(
When 2) holds true, it is determined that the voice section has started from time k. At this time, V k-1 can be considered to represent the level of background noise. The signal level when the other party's voice is present is such that the level caused by the voice is added to the background noise level, so the voice detection level threshold V th is set to a value larger than the background noise level. Should. From this, the voice detection threshold value ■th is generally calculated according to equation (3). However, the constant f is a value larger than 1.

V ih= f X V m−r
(3) In this embodiment, at this time, V th
The pose compression and dialogue responses are calculated separately. Specifically, in equation (3), the value of the constant f is expressed as f, , f
t.

Two types are prepared: a voice detection threshold for dialogue response, Vth-al, and a voice detection threshold for pause compression, Vt.
Calculate h−ゎ according to the following formulas (S2
5).

V tli-a= f a X V k-+
・・・・・・(4)V th−ゎ=fゎ
×■U-1・・・・・・(5) As explained earlier, regarding dialogue responses, in order to prevent the situation where the caller is detected as having spoken even though the caller is not speaking, , lower the voice detection sensitivity. That is, V th-a is set to a relatively large value by the threshold setting means 9A. on the other hand,
Regarding pause compression, voice detection sensitivity is increased to prevent voice from being detected as silent even though the caller is speaking and to prevent the voice from being dropped from the recording. That is, ■th
-5 is set to a relatively small value by the threshold setting means 9A. Therefore, between f and fb, f,≧fゎ (6)
There is a relationship between Hereinafter, the voice detection threshold value V th-Ill Vth-b calculated by equations (4) and (5)
Performs voice detection processing based on.

When performing the pause compression recording process, between the signal level ■o read from the signal level measurement unit 8 and the voice detection threshold for pause compression V th-b, ■o ≧V th-b
. . . It is confirmed whether the relationship (7) holds true (S26+).

(S27). If the formula (7) holds true, it is determined that the other party's voice should be recorded (S2'8), and the message recording section 6 records the line signal to the message storage section 7 via the telephone call circuit 3. I do. This recording is temporarily stopped when formula (7) does not hold529), and then the recording is repeated.
When the formula (7) is satisfied, the process is restarted.

On the other hand, when executing the utterance end detection process, the signal level ■ read from the signal level measuring section 8 is used. and the voice detection threshold for dialogue response processing V th-m, ■o≧V th-a ...-(
It is confirmed whether the relationship 81 holds true (S30).

The count value of timer 10, which was reset and started at the start of the message recording process, is the silent judgment time threshold T.
. □. . If Equation (8) does not hold true (S31), it is determined that the other party has no intention of speaking, and the message recording process is terminated (532), and the telephone sets the next response message. Move on to the sending operation. At this time, if the recording process of the business message recording unit 6 has already started, the response message is sent after finishing this recording process. In step Cs30), the count value of the timer 10 reaches the silence determination time threshold T, □. . If formula (8) is satisfied before the above occurs, it is determined that the other party has started speaking, and a process for detecting the end of speaking is started. Therefore, the timer 10 once resets the count value (333) and starts measuring the duration of the silent section that appears during utterance (S34) to (S37). At this time, if the formula (8) is satisfied (S38), the count of the timer 10 is reset again (333),
) is not satisfied (S38), compare the count value of the timer 10 and the utterance end judgment time threshold value T@n d (539), and when the count value exceeds T@In, the other party's It is determined that the utterance has ended, the message recording operation is ended (540), and the next response message is sent.

Incidentally, steps (S35, S36, 537) are a process in which voice recording is performed only in the section where equation (7) is satisfied at the same time as steps (S27) to (S29), and no recording is performed in other sections.

The above operations are repeated until the message is recorded after the final response message is sent. When the business message recording operation for the final response message is completed, the loop opening/closing section 2 is operated again, the line is opened, and the automatic response is terminated.

〔Effect of the invention〕

As explained above, the present invention provides a threshold setting means that can independently set two types of voice detection thresholds for the voice detection means: a voice detection threshold for dialogue response and a voice detection threshold for pause compression. Since the voice detection threshold for dialogue response and the voice detection threshold for pause compression can be set independently, even if voice detection malfunctions, the effect on each process will not reveal a fatal flaw. It becomes possible to set each threshold value.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing an embodiment of the present invention, FIG. 2 is a diagram illustrating a specific example of the configuration of the signal level measuring section and the interface with the control section, and FIG. 3 is the inside of the response message storage section. 4(a) and 4(b) are flowcharts showing an example of the operation of the embodiment of FIG. 1, and FIG. 5 shows how the caller starts speaking in response to the response message. 12 is a flowchart illustrating a conventional algorithm for determining whether or not the caller has finished speaking based on the voice detection result. In the figure, 1 is a calling signal detection unit, 2 is a loop opening/closing unit, 3 is a communication circuit, 4 is a response message sending unit, 5 is a response message storage unit, 6 is a message recording unit, and 7 is a message storage unit , 8 is a signal level measuring section, 8A is an A/D converter, 9 is a control section, Fig. 3, Fig. 3 (a)

Claims (1)

[Claims] It has an interactive response function that automatically receives a call, sends out multiple response messages stored in the response message storage unit in response to messages from the caller, and allows the caller to respond in an interactive manner. and a voice detection means for detecting the presence or absence of the caller's voice. In an interactive answering machine that also has a pause compression recording function, it is possible to independently set two types of voice detection thresholds for the voice detection means: a voice detection threshold for dialogue response and a voice detection threshold for pause compression. An interactive answering machine characterized by being equipped with a threshold setting means.