JPH03150953A - Automatic transfer telephone set - Google Patents

Abstract

PURPOSE:To surely and easily receive a call from a party desiring transfer with simple hardware constitution even when the arrival of the call comes from an external line or an extension by deciding whether or not a call arrived at present is to be transferred based on a voice signal specific to a transfer desiring person in advance. CONSTITUTION:Plural extension telephone sets 3A, 3B,... are connected to an exchange 2, which is provided with a storage section 5, an automatic reply section 6, a taker recognition section 8 and a reply control section 9. Whether or not the arrival of a call at present is to be transferred is decided based on a voice signal specific to a transfer desiring person in advance, then there is no limitation of making a call from a specific telephone set. Thus, even when the arrival of a call results from either an external line or an extension, the arrival of the call from a transfer desiring person in advance is surely and easily obtained with simple hardware constitution in the automatic transfer telephone set.

Description

[Detailed description of the invention] [Industrial application field] TECHNICAL FIELD The present invention relates to automatic transfer telephone equipment.

[Prior Art] Conventionally, there is an automatic transfer charging device as described in Japanese Patent Application Laid-Open No. 63-203047.

This automatic forwarding telephone device stores forwarding source extension information, forwarding destination extension information, and forwarding party extension information in a storage unit connected to the exchange by a special operation from the telephone of an extension subscriber who is attempting to use the call forwarding service. Register. When a call is received on the forwarding extension telephone, the extension information of the call is compared with the extension information of the forwarded party, and on the condition that the extension information is the one registered as the forwarded party, the incoming call is is forwarded according to the forwarding destination extension information.

[Section H trying to invent or solve] However, the prior art has the following problems (1) and (2).

■Extension information is used to determine whether or not the current incoming call can be forwarded, and incoming calls from outside will not be forwarded.

■Similar to (■) above, if you wish to receive a transfer service, you must make a call from a specific telephone that has registered extension information. Conversely, anyone can use a registered telephone to receive a forwarding service.

The present invention provides an automatic forwarding telephone device that can reliably and easily receive a call from a person who wishes to forward the call, regardless of whether the current call is received from an outside line or an extension line, using a simple hardware configuration. The purpose is to

[Means for Solving the Problems] The present invention according to claim 1 stores transfer source extension telephone information and transfer destination extension telephone information, and registers a specific transfer applicant in advance. In an automatic forwarding telephone device that automatically forwards the incoming call to a forwarding destination extension when a call is received from a transfer requestor to the forwarding extension, an automatic answering section that automatically answers the incoming call to the forwarding extension; , a speaker recognition unit that recognizes whether or not the voice of the caller who has previously registered a specific speaker who is a person who wishes to transfer the call and who has responded to the automatic response unit upon receiving the current call is the voice of the registered speaker;
and a response control unit that connects the incoming call to the forwarding destination extension telephone based on the recognition result of the speaker recognition unit, on the condition that the caller at the time of the current incoming call is the registered speaker, and the speaker recognition unit This system uses a neural network to recognize the speaker from the caller's voice.

The present invention according to claim 2 provides, as inputs to the neural network, [3] temporal changes in voice frequency characteristics, ■ average linear prediction coefficient of voice, ■ average PAR (:OR) of voice. coefficient, [4] Average frequency characteristics of voice and pitch frequency, ■ Average frequency characteristics of voice waveform with high frequency emphasis,
and (1) one or more of the average frequency characteristics of voice is used.

According to a third aspect of the present invention, the neural network is a hierarchical neural network.

[Function] According to the present invention as set forth in claim 1, the following (1) to (3)
It has the function and effect of

(1) Determine whether or not this incoming call should be forwarded.
Since the call is made using a voice unique to the person who wishes to transfer, which has been specified in advance, there is no restriction that the call must be made from a specific telephone.

(2) According to (1) above, incoming calls from not only internal lines but also external lines are subject to transfer. Further, those who receive the transfer service can be specified only to speakers who have been registered in advance as transfer applicants.

(3) Since a neural network is used as a means of speaker recognition, there are the following advantages.

■Deterioration of normal operation rate over time is extremely small. This has been confirmed by the experimental results described below, and is presumed to be because the neural network can have a structure that is less susceptible to fluctuations due to differences in audio timing.

■Neural networks are, in principle, simple and quick to perform calculations on the entire network.

■In principle, each unit that makes up a neural network operates independently, and parallel arithmetic processing is possible. Therefore, calculation processing is quick.

(2) With the above (2) to (3), automatic forwarding telephone equipment can be easily processed in real time without using a complicated processing device.

Further, according to the present invention as set forth in claim 2, the above (1) to (3)
In addition to the effects of ), there is also the effect of (4) below.

(4) Since one or more of each of the elements (■ to ■) described in claim 2 is used as input to the neural network, the preprocessing for obtaining the input is different from conventional complex feature extraction. Therefore, the process is simple, and the time required for this preprocessing is short.

Further, according to the present invention according to claim 3, the above (1) to (4)
), there is the following effect (5).

(5) Regarding hierarchical neural networks, a simple learning algorithm (pack propagation) as described later has been established, and it is possible to easily form a neural network that can achieve a high recognition rate.

[Example] FIG. 1 is a schematic diagram showing an example of an automatic transfer telephone device to which the present invention is applied, and FIG. 2 is a schematic diagram showing an example of a speaker recognition unit.
Fig. 3 is a schematic diagram showing input audio, Fig. 4 is a schematic diagram showing the output of a bandpass filter, Fig. 5 is a schematic diagram showing a neural network, and Fig. 6 is a schematic diagram showing a hierarchical neural network. FIG. 7 is a schematic diagram showing the structure of the unit.

Prior to describing specific embodiments of the present invention, the configuration of the neural network and the learning algorithm will be described.

(1) Due to its structure, neural networks are
It can be roughly divided into two types: the hierarchical network shown in FIG. 5(A) and the interconnected network shown in FIG. 5(B). Although the present invention may be constructed using either of these networks, the hierarchical network is more useful because a simple learning algorithm as described below has been established.

(2) Network Structure A hierarchical network has a hierarchical structure consisting of an input layer, an intermediate layer, and an output layer, as shown in FIG.

Each layer is composed of one or more units. The connections are forward connections from the input layer to the middle layer to the output layer, and there are no connections within each layer.

(3) Structure of the unit The unit is a model of a neuron in the brain and has a simple structure as shown in FIG. It receives input from other units, sums it up, transforms it using a certain rule (conversion function), and outputs the result. Each connection with another unit is given a variable weight that represents the strength of the connection.

(4) Learning (pack propagation) Learning of a network is to bring the actual output closer to the target value (desired output). Learn by making changes.

Further, as a learning algorithm, for example, Rumel
hart, D.E., McClelland,
J, L, and thePDP Re5earc
h Group, PARALLEL DISTRIB
UTED PROCESSING, the MIT P
ress, 1986. Pack propagation as described in .

A specific embodiment of the present invention will be described below.

As shown in FIG. 1, the automatic transfer telephone device 1 includes a plurality of extension telephones 3A, 3B, 3C, .
, a response control section 9.

(1) The storage unit 5 stores transfer source extension telephone information and transfer destination extension telephone information.

(2) The automatic response unit 6 automatically responds to incoming calls on the transfer source extension telephone.

(3) Using a neural network, the speaker recognition unit 8 registers in advance a specific speaker who wishes to transfer the call, as will be described later, and the caller who responded to the automatic response unit 6 when receiving the current call. It is recognized whether or not the voice is the voice of the registered speaker.

At this time, the speaker recognition section 8 is configured by combining a preprocessing section 1o, a neural network 20, and a determination circuit 30.

The preprocessing unit 10 performs simple preprocessing on the input audio as described below.

The neural network 20 performs the following learning operation (2) and the following evaluation operation (2).

■Learning Learning words, for example, common "transfer" for all registered speakers
shall be.

The target value (target output value of each output unit constituting the output layer) is set to (1, O) for registered speakers and (0, 1) for other speakers.

The preprocessing section 10 performs preprocessing on the input voice "transfer" of the registered speaker, and the preprocessing result is input to the neural network 20. And neural network 20
Output value (output value of each output unit that makes up the output layer)
The conversion function and weight of each unit of the neural network 20 are modified so that the value approaches the target value.

Repeat this learning action.

■Evaluation Preprocessing is applied to the words spoken by the caller at the time of this call arrival, and the preprocessed word speech is input to the neural network 20 to obtain the output value of the neural network.

If the output value of the neural network 20 exceeds a certain threshold and is close to (lO), the determination circuit 30 recognizes the caller 45 in the current incoming call as a registered speaker.

(4) Based on the recognition result of the speaker recognition unit 8, the response control unit 9 connects the incoming call to the transfer destination extension telephone on the condition that the caller at the time of the current incoming call is the registered speaker.

Hereinafter, a case will be described in which a hierarchical neural network 20 is used as the speaker recognition unit 8, as shown in FIG. A specific example will be described.

The preprocessing section 10 of the speaker recognition section 8 is composed of a voice input section 12, a low-pass filter 13, a band-pass filter 14, an averaging circuit 15, and a memory 16, as shown in FIG.

(A) Learning of neural network 20 ■Registered speaker 5
and 25 other speakers each input the word "transfer" into the voice input section 12 ten times.

■The high-frequency component of the human-powered audio signal is cut by the low-pass filter 13. Then, this input audio is temporally equally divided into four blocks as shown in FIG.

■As shown in Fig. 2, the audio waveform is passed through a band pass filter 14 of multiple (n) channels, and each block, that is, each fixed time, is passed through the band pass filter 14 as shown in Figs. 4 (A) to (D). Obtain frequency characteristics.

At this time, the output signal of the bandpass filter 14 is averaged by the averaging circuit 15 for each block, that is, for a certain period of time.

Through the above preprocessing, the "temporal change in the average frequency characteristics of audio within a certain period of time" was obtained.

The output of the averaging circuit 15 is transferred directly to the neural network 20 or indirectly via the memory 16.

(2) The neural network 20 is composed of a three-layer hierarchical neural network. The input layer 21 is composed of 4×n units corresponding to 4 blocks of preprocessing and n channels. The output layer 22 is composed of two units: a registered speaker group and other speaker groups.

Set the target value of the output layer 22 to (1,0) for registered speakers.
For other speakers, it is set to (0, 1).

The weights and conversion functions of the neural network 20 are corrected so that the output value of the output layer 22 approaches the target value. This operation is repeated until the error in the output relative to the input converges to a certain level, and the system learns to construct a network that can guarantee a certain recognition rate.

(B) Control of the automatic transfer telephone device 1■ When determining the first extension telephone 3A as the transfer source, first use the telephone 3A to enter the transfer destination telephone number and the voice of the transfer requestor (or persons) to "transfer". Enter.

As a result, the extension information of the first extension telephone 3A as the transfer source and the extension information of the specified transfer destination telephone are stored in the storage unit 5.

Incidentally, the voice "transfer" of the person requesting transfer inputted into the first extension telephone 3A as described above is inputted into the preprocessing section 10 of the speaker recognition section 8, and is processed by the neural network described in (A) above. This is provided as audio input for learning 20.

■After that, from the internal or external line,
When the extension telephone 3A receives a call, the automatic answering section 6 is activated and automatically responds with the guidance message "Yes, Sahara Tesu. Please talk to me."

■Here, when the caller is a registered speaker and utters "transfer," this voice is subjected to the same preprocessing as in (A) above in the preprocessing section 10 of the speaker recognition section 8. , this preprocessing result is input to the neural network 20 of the speaker recognition section 8.

■The determination circuit 30 of the speaker recognition unit 8 receives the output of the neural network 20, and if the output of the neural network 20 is close to (1, 0), it determines that the caller at the time of the current call is the registered speaker, and (0 , 1), the caller at the time of the current call is determined to be another speaker.

■If the recognition result of the ■speaker recognition unit 8 is "registered speaker," the response control unit 9 controls the exchange 2 to
The incoming call is connected to the transfer destination extension telephone stored in the storage unit 5, and if the recognition result of the speaker recognition unit 8 is “other speaker”, the automatic response unit 6 is activated and the message “I’m home” is sent. I'm out of town.'' and the call ends.

(C) Experiment As a result of using the above automatic transfer telephone device 1, the normal operation rate was 100% immediately after learning, but the normal operation rate was 99 after 3 months.
%, and it was recognized that the deterioration of the normal operation rate over time was extremely small.

Further, the processing speed (the time required to recognize the utterance of one word) of the automatic forwarding telephone device 1 was within 1 second.

Next, the operation of the above embodiment will be explained.

(1) Determine whether or not this incoming call should be forwarded.
Since the call is made using a voice unique to the person who wishes to transfer, which has been specified in advance, there is no restriction that the call must be made from a specific telephone.

(2) According to (1) above, incoming calls from not only internal lines but also external lines are subject to transfer. In addition, those who receive the transfer service can be specified only to speakers who have been registered in advance as transfer applicants.

(3) Using the neural network 20 as a means of speaker recognition has the following advantages.

■Deterioration of normal operation rate over time is extremely small.

This is presumably because the neural network 20 can have a structure that is less susceptible to fluctuations due to differences in audio timing.

(2) In principle, the neural network 20 has simple and quick calculation processing for the entire network.

(2) In principle, the neural network 20 operates independently of each unit that constitutes it, and is capable of parallel arithmetic processing. Therefore, calculation processing is quick.

(2) With the above (2) to (2), the automatic transfer telephone device 1 can be easily processed in real time without using a complicated processing device.

■As the input to the neural network 20 is the temporal change in the frequency characteristics of the voice, the preprocessing required to obtain the input is simpler than the conventional complex feature extraction. The time is short enough.

At this time, as an input to the above neural network,
Furthermore, since "temporal changes in the average frequency characteristics of audio within a certain period of time" are used, the processing in the neural network 20 is simple, and the time required for this processing is shorter.

Since we used a zero-layer neural network 20,
Currently, a high recognition rate can be achieved using a simple learning algorithm (pack propagation) that has already been established.

In the implementation of the present invention, as inputs to the neural network, [3] Temporal changes in frequency characteristics of voices, [3] Average linear prediction coefficients of voices, and [3] Average PARCOR coefficients of voices. , [4] Average frequency characteristics of voice and pitch frequency, ■ Average frequency characteristics of voice waveform with high frequency emphasis,
and [3] one or more of the average frequency characteristics of the voice can be used.

Then, just as the element (■) above was further used as "temporal change in the average frequency characteristics within a certain period of time", the element (■) above is also used as "the average linear prediction coefficient within a certain period of time". "Temporal change in
``Temporal change in the average PARCOR coefficient within a certain time period of audio'', the above element (■) is ``the average frequency characteristic and temporal change in pitch frequency within a certain time period of audio'', and the above element (■) is: It can be used as a "temporal change in the average frequency characteristics within a certain period of time of a voice waveform that has been subjected to high-frequency emphasis."

Incidentally, the linear prediction coefficient of (2) above is defined as follows.

That is, it is known that there is generally a high proximity correlation between sample values (χ0) of audio waveforms.

Therefore, it is assumed that the following linear prediction is possible.

Linear predicted value χ, = -ΣαLχt-1...(1) Linear prediction error ε(=χ, -χt...(2) Here, χt: sample value of the audio waveform at time t, (αt
) (i=t,...,p): (9th order) linear prediction coefficient Now, in the implementation of the present invention, 2 of the linear prediction error εt
The linear prediction coefficient (αi) is determined so that the root mean value becomes the minimum value.

Specifically, (e t)2 is calculated, and its time average is (ε
t) expressed as 2, θ(εta” / a a 1 =
O,t=1.2. ..., p, (α,) can be obtained from the following equation.

Σ'(! 1VII-Jl ='+ J=1 + 2
+ "'+ p..." (3) Also, PARC of the above ■
The OR coefficient is defined as follows.

That is, [kn] (n=1..., p) is (9th order) P
When using ARCOR coefficient (partial autocorrelation coefficient), PAR (
: In the OR coefficient, +□ is the forward residual ε due to linear prediction,
The normalized correlation coefficient between fl and backward residual εt-+n+, (b) is defined by the following equation.

ε also -(n◆1) ...(4) Here, ε, (t'=χt-Σ α・χt-1゛(αl
): Forward prediction coefficient, ε t-+n◆th (b) = χ t -L n *
l ) −, EllJ13 J + χ t−j
, (βj) : Backward prediction coefficient Also, the pitch frequency of the voice in the above (■) is the reciprocal of the repetition period (pitch period) of the vocal cord wave.

In addition, since pitch frequency, which is a basic parameter of the vocal cords that differs between individuals, was added as an input to the neural network, it is possible to improve the recognition rate of speakers, especially between adults/dwarfs and male/female. .

Furthermore, the above-mentioned high frequency enhancement (2) is to compensate for the average slope of the spectrum of the audio waveform to prevent concentration of energy in the low frequency range. However, the average slope of the spectrum of the speech waveform is common to all speakers and is unrelated to the speaker's recognition.

On the other hand, if we manually input a speech waveform whose average slope of the spectrum has not been compensated for into a neural network, the feature of the average slope of the spectrum will be extracted when the neural network learns. It takes time to extract the peaks and valleys of the spectrum necessary for speaker recognition. On the other hand, when high-frequency emphasis is applied to the input to a neural network, it is possible to compensate for the average slope of the spectrum that is common to all speakers and is unrelated to recognition, but that affects learning, which speeds up the learning process. or faster.

[Effects of the Invention] As described above, according to the present invention, regardless of whether the current incoming call is from an outside line or an extension, incoming calls from a person requesting transfer can be received reliably and with a simple hardware configuration. You can get a call forwarding device that is easy to use.

[Brief explanation of the drawing]

FIG. 1 is a schematic diagram showing an example of an automatic transfer telephone device to which the present invention is applied; FIG. 2 is a schematic diagram showing an example of a speaker recognition unit;
Fig. 3 is a schematic diagram showing input audio, Fig. 4 is a schematic diagram showing the output of a bandpass filter, Fig. 5 is a schematic diagram showing a neural network, and Fig. 6 is a schematic diagram showing a hierarchical neural network. FIG. 7 is a schematic diagram showing the structure of the unit. DESCRIPTION OF SYMBOLS 1... Automatic forwarding telephone device, 5... Storage unit, 6... Automatic response unit, 8... Speaker recognition unit, 9... Response control unit, 10... Preprocessing unit, 20 ... Neural network, 30... Judgment circuit.

Claims (3)

[Claims] (1) When the transfer source extension telephone information and the transfer destination extension telephone information are stored, and a specific transfer applicant is registered in advance, and the transfer source extension telephone receives a call from a registered transfer applicant, The automatic forwarding telephone device that automatically forwards the incoming call to the forwarding extension telephone has an automatic answering section that automatically answers the incoming call to the forwarding extension telephone, and a specific speaker who is the person who wishes to forward the call, which is registered in advance, and this time. A speaker recognition unit recognizes whether the voice of the caller who responded to the automatic answering unit when the call arrives is the voice of the registered speaker, and the recognition result of the speaker recognition unit identifies the caller at the time of the current call. and a response control unit that connects the incoming call to the transfer destination extension telephone on the condition that the caller is the registered speaker, and the speaker recognition unit recognizes the speaker from the voice of the caller using a neural network. An automatic forwarding telephone device characterized by: (2) As inputs to the neural network, [1] Temporal changes in frequency characteristics of voice, [2] Average linear prediction coefficient of voice, [3] Average PARCOR coefficient of voice, [4] Voice using one or more of the following: average frequency characteristics and pitch frequency; [5] average frequency characteristics of high-frequency emphasized speech waveform; and [6] average frequency characteristics of voice. The automatic transfer telephone device according to claim 1. (3) The automatic transfer telephone device according to claim 1 or 2, wherein the neural network is a hierarchical neural network.