CA2011662C - Voice-controlled apparatus using telephone & voice-control method - Google Patents

Abstract

Abstract of the Disclosure A voice-controlled apparatus of this invention includes receiving/discriminating means for receiving a DTMF signal externally input through a telephone line and discriminating the DTMF signal, level detecting means for detecting a signal level of the DTMF signal input to the receiving/discriminating means, speech recognition means for recognizing the content of a voice signal externally input through the telephone line, input level adjusting means for adjusting a signal level of the voice signal input to the speech recognition means on the basis of the signal level detection result by the level detecting means, and function executing means for executing a function according to the content of the voice signal recognized by the speech recognition means.

Description

- 20~662 The present invention relates to a voice-controlled apparatus such as a so-called automatic answering telephone having an automatic answering function.
Some conventional automatic answering telephones have a function for receiving a push phone dial signal. ;~ -~
With a telephone of this type, a user can make a tele-phone call to this telephone from an outdoor telephone :, . ~ "-to hear a recorded message. In this case, the user inputs an ID number (identification data) by depressing the push buttons on the outdoor telephone. ~,~
The ID number is converted to a DTMF (Dual Tone -~
Multi Frequency) signal used as the push phone dial signal, and reaches the automatic telephone through the telephone line, where lt is encoded. When the encoded .: ~. .,;
DTMF signal coincides with a number reglstered in advance in the automatic answerlng telephone, the user `
can hear the recorded content. ~;
In order to hear the recorded content, push buttons . . ,, ' '.:: .~ `, ".,;
are depressed to designate a predetermlned operatlon. '"!''j.
For example, "l" designates a rewind operatlon; "2", a . ',:!~
fast-forward operation; "3", a playback operatlon; and ~ "4", a stop operation. The automatic answering ,~
; telephone receives these DTMF signals and executes the - . ~s .i -: .. .
designated function.
However, since the above operation functions are ; encoded by numerals, the user may tend to forget the correspondence between the operations and codes.
~ ~' ,~.'' ~ . : . ~, . . :. ..~-.. ~.` ' ~:.

Thus, if each operation is designated by a voice or voice input (e.g., a word ~rewind~, llfast-forward", "playback", or "stop"), and the designated operation is ~ -recognized by the automatic answering telephone, the ~ 1~
user can easily actuate the above operations. ~-- s In order to execute speech recognition with `~
existing techniques, a voice level input to a speech -~
recognition means must be appropriate. However, the level of a voice signal externally input through a telephone line varies considerably. causes of the -variation are mainly present in the state of the telephone line. More speclfically, when the telephone line suffers from a large signal transmission loss, the level of the voice signal reaching the speech recogni-tlon means is decreased. However, when a user makes a local telephone call, the transmission loss on the telephone line is very small, therefore a high level voice signal can be input to the speech recognition More specifically, when the telephone line is con~
nected to the speech recognition means, the input level `
of a voice signal varies considerably depending on the state of the telephone 11ne between the calllng party and the speech recognltion means. Therefore, this state is very dlsadvantageous for a speech recognition apparatus, and accurate speech recognitlon may not ; always be attained.
~: . ...:
','~'~ ,,;.',' :. ,," .....

-` 20116~2 - ~ .
It is therefore an object of the present invention to solve the problem of the level of a voice input to a speech recognition means varying depending on the state -~ -of a telephone line, disturbing accurate speech .: .,: .. . .. ~
recognition, and to provide a voice-controlled apparatus ~ s which can malntain a proper input level to the speech recognition means, so that highly accurate speech recognition can be achieved. ; .
A voice-controlled apparatus according to the pre-sent inventlon comprises: receiving/discriminating means . :.
for receiving and discriminating a DTMF signal exter- .
nally input through a telephone line; level detecting means for detecting the signal level of the DTMF signal input to the receiving/discrlminating means; speech .`~?
recognition means for recognlzing the content of a voice `.. ,,~
signal externally lnput through the telephone line; ,; .
input level ad~ustlng means for ad~usting the signal `i level of the voice signal input to the speech recogni- :.. ~`.
tlon means, on the basis of the signal level detection : 20 result of the level detecting means; and functlon executlng means for executing a function accordlng to the content of the volce signal recognlzed by the speech recognltion means.
In the above arrangement, an lnitlally input DTMF
signal ls discrlmlnated, and lts reception level is measured. The input level of the speech recognition : ~.
means ls ad~usted to a proper value on the basls of the ` - 2û116~2 detected reception level. Thereafter, a voice signal to be recognized is input to the speech recognition means.
Since a DTMF signal can be satisfactorily discrimi-nated by existing techniques, an ID number can be reliably discriminated. When the reception level of the !
DTMF signal is detected simultaneously with discrimina- ~ -tion of the DTMF signal, the degree of loss caused by ~-the telephone line can be measured. Thus, the sub~
sequent input level of the speech recognition means can be properly maintained by compensating for the measured `~
loss. As a result, speech recognition can always be highly accurately achleved.
This invention can be more fully understood from the following detailed descriptlon when taken in con- `~
~unction with the accompanying drawings, in which~
Fig. 1 is a block diagram showing a voice-controlled apparatus accordlng to an embodiment of the present lnventlon; ~
Flg. 2 shows detalls of a programmable gain ` ~ ;
amplifler used in the embodiment of Fig. l;
Fig. 3 shows a waveform diagram for explaining an ' operatlon for detectlng the level of a DTMF slgnal; and Flg. 4 ls a flow chart for explalnlng an operatlon of the embodlment of Fig. 1.
A volce-controlled apparatus accordlng to an embodlment of the present lnventlon wlll be descrlbed below wlth reference to the accompanylng drawings. ~ ;~

--` 201~662 ~

In Fig. 1, reference numeral 1 denotes a telephone at a distant place; and 2, a telephone line. Reference numeral 3 denotes a telephone with a network control unit 4 ( to be referred to as an NCU hereinafter). ;-Telephone 3 has a function of receiving a voice signal input from telephone line 2 or sending a voice signal -- -out on telephone line 2. ;
Reference numeral 5 denotes a DTMF signal receiver ','.J',' "'~"'.'"'.'-"'~
for receiving and encoding a DTMF signal. Reference -~-numeral 6 denotes a level detector for detecting the reception level of a DTMF signal input through NCU 4. , Reference numeral 7 denotes a speech responder havlng a ;~
conventional volce synthesizing circuit and a circuit for driving the synthesizing circuit. Speech responder 7 sends a voice message to the calling party. Reference numeral 8 denotes a PGA ~programmable galn control ampllfler) for controlling the magnltude of a volce level lnput to speech recognition unit 9 (to be descrlbed later). Reference numeral 9 denotes the speech recognltlon unlt for recognizlng a voice signal (word volce signal) input from telephone line 2 through ;~
NCU 4. Reference numeral 10 denotes a tape recorder (automatlc answerlng means) whlch ls controlled by controller 11 (to be descrlbed later). Reference numeral 11 denotes the controller for controlllng the~ ;
entlre apparatus shown ln Flg. 1. `~
Varlous speech recognitlon technlques may be "~, ', .~, .'`"' :`,'':'`'..

- 6 - ~ '~

executed by speech recognition unit 9. For example, a technique disclosed in the following reference:
Yoichi TAKEBAYASHI et al., ~TELEPHONE SPEECH ~-RECOGNITION USING A HYBRID METHOD", IEEE 7th - :r -''"`
International Conference on Pattern Recognition Proceedings, (July 30 - August 2, 1984), Montreal, -~
Canada, pp. 1232 - 1235 , The present specification incoxporates the disclo- ,;~
sure of the above reference.
Fig. 2 shows details of PGA 8 used in the embodiment shown in Fig. l.
An input DTMF signal is input to the inverted input terminal of operational amplifier 80 through resistor `~
Ra. The output from operational amplifier 80 is negati-vely fed back to the inverted input terminal of opera-tlonal amplifier 80 through feedback resistors RNF and ;~
Rb.
Feedback resistors RNF include n series-connected ~: ..... ~,.
resistors Rl to Rn. Resistors Rl to Rn are connected in parallel with electronic switches Sl to Sn. In this case, gain G of the entire circuit shown in Fig. 2 is glven by~
G - -(Rb + RNF)/Ra ........................ (1) -~
Fe~dback resistors RNF are expressed by:
RNF = Rl + R2 + .... + Rn ....................... (2) ~ -~
Swltches Sl to Sn are turned on/off ln response to `~

n outputs from switch driver 82. The output from switch -` 201~6~2 ;~ ~

.,-. ~ ~ ,',,. "
driver 82 is determined on the basis of a gain control ;;
instruction from controller 11.
Assume a simple example wherein the gain control - -instruction is 2-bit binary data [x,y], resistors RNF
include only two resistors Rl and R2, and the number of `
switches Sn is 2 (Sl and S2). When [x,y] = ~o,o], since both switches Sl and S2 are set ON, RNF = o. When ;i ;
[x,y] = [0,1], since switch Sl is set OFF and switch S2 ` ;
ls set ON, RNF = Rl. When [x,y] = ~1,0], since switch Sl ls set ON and switch S2 is set OFF, RNF = R2. When -! . .: . ', . j`, [x,y] = [1,1], since both switches Sl and S2 are set OFF, RNF = Rl + R2. In thls manner, since reslstors RNF -`-can have four reslstances, PGA 8 can have one of four galns according to equation (1).
As descrlbed above, the gain of PGA 8 is controlled ~;
on the basis of the galn control instructlon from ~ ;
controller 11.
Note that PGA 8 may be arranged as follows. That ls, four ampllflers havlng dlfferent fixed gains are ;
prepared, and a DTMF slgnal ls lnput to each of these `
ampliflers. A DTMF signal, amplifled with a predeter~
mined flxed galn selected based on galn control lnstruc-tlon, ls extracted from one of the amplifiers.
The operation of the arrangement shown in Fig. 1 ~ ;
will be described below, wlth reference to the waveform diagram of Flg. 3 and the flow chart of Fig. 4. A user - -makes a telephone call to telephone 3 uslng outside j ,-' ~,''.., - 8 - , telephone 1 (STlo). NCU 4 provided in telephone 3 is enabled to interrupt a CPU in controller 11 tSTll).
Thus, controller 11 enables speech responder 7 to send, to telephone 1 through telephone line 2, a message such as " I (owner) am out. Please leave a message on the -~
recorder, or please input ID number." (ST12). When the user depresses the push buttons of telephone 1 to input an ID number (ST13), level detector 6 detects the signal level of the DTMF signal (left waveform in Fig. 3) of the ID number (ST14). Parallel to the level detectlon, ~,;
DTMF slgnal receiver 5 discriminates the content of the received DTMF slgnal, and sends the discrimination ;;~
result to controller 11 (STl5).
The CPU in controller 11 compares the sent dlscri~
mlnation result (the ID number input by the user) with ;
ID number table lla registered in its own memory. If the dlscriminatlon result lndicates non-reglstration ~ -(NO ln step STl6), controller 11 enables speech ~ -responder 7 to send a message such as "Thls ID number ls ;~
invalid." (ST17), and subsequently cancels the teIephone ~;
call (ST18). When the telephone call ls cancelled, the operatlon of the apparatus shown ln Flg. l ls completed. i ;
If the sent dlscrlmination result indicates registration of the ID number (YES in step ST16), the ; ~
controller 11 sends a message such as "Please send an ~ ;
operation command." (STl9).
After the operation command message is sent, the ;~
, ., ,, . , ,. ., ,~ . . -: .::..,.:'' .

g CPu in controller 11 controls the gain of PGA 8 on the basis of the reception level of the DTMF signal detected in step ST14 (ST20). More specifically, when the ~ ;",.,!,~
reception level of the DTMF signal is low, the gain of -~-PGA 8 is increased; otherwise, the gain of PGA 8 is decreased, so that the input signal level of speech `~
recognition unit 9 always falls within a predetermined ''~
range (acceptable range in Fig. 3). After the gain control, the CPU in controller 11 enables speech recognition unit 9 (ST21).
The operations in steps ST20 and ST21 are executed within a very short period of time until the user sends the subsequent operation command. ~; `
The user sends a word (operation command) indicating a desired operatlon (e.g., RECORD, PLAYBACX, STOP, PAUSE, REWIND, FAST FORWARD, etc.) from telephone 1 ln response to the operatlon command request message ln step STl9 (ST22). Speech recognltlon unit 9 ;~
recognlzes the sent volce operatlon command, and inputs -the recognltion result to controller 11 (ST23). The CPU
ln controller 11 sends an operatlon lnstructlon corre~
spondlng to the input recognition result (e.g.~ RECORD) to tape recorder 10 (ST24). Tape recorder 10 executes an operation (RECORD) according to the sent operatlon command (ST25).
The operatlons in steps ST22 to ST25 contlnue as long as the telephone is connected (NO in step ST26).
" ',~ ',.,'~' ' ,,,~,: ,.. . "~ , .' ,,, '' ''.;~

~--` 2 ~ 2 -- 1 0 -- ` '`'' ' " ' If the telephone is hung up (YES in step ST26), the operation of the apparatus shown in Fig. 1 is completed. ~ -The user of telephone 1 need only speak a message -- `
. :.....
content when he or she wants to record a message, and ~ : -the voice content is recorded on tape recorder 10. When the message is over and telephone 1 is hung up, all the processing operations are completed, and telephone 3 returns to its initial state.
The above operations can be summarized as follows.
When the user wants to hear the recorded content using outside telephone 1, he depresses the push buttons of telephone 1 to input his ID number. For example, if the ID number is a four-digit number "1234", buttons "1", "2", "3", and "4" are depressed in turn. ~ -~
The input ID number is converted to a DTMF slgnal, and the DTMF signal is input to DTMF signal recelver 5 through telephone llne 2 and NCU 4 of telephone 3. DTMF
slgnal receiver 5 dlscrlminates the ID number from the lnput DTMF signal, and sends the discrimination result to controller 11.
When NCU 4 is enabled to lnterrupt controller 11, controller 11 enables level detector 6. Level detector : -:. :; ...
6 detects positive peak value C of the input DTMF slgnal ~ ; :
(right waveform ln Fig. 3) during a time period enabled by controller 11 (time period from measurement start up to completion of measurement)~ and sends the detection result to controller 11.

. ~'f~ ~ .

:. , . .... . . ', : -i' ' ~; ' ' " : ; ~ . :: : ' `` 2 0 ~ ~ 6 ~ 2 .~
',. ~-.,' When controller 11 determlnes that DTMF slgnal receiver 5 detects a DTMF signal of the predetermined ~ ~-number of dlgits, controller 11 sends a measurement end - i"
instruction to level detector 6.
When a telephone call is made from telephone 1, tape recorder 10 starts a recording operation. When DTMF signal receiver 5 detects that a DTMF signal is input, controller 11 immediately sends a pause command to tape recorder 10 to pause the recording operation. ~ ~-Controller 11 checks the discriminated ID number to determine if the ID number is valid. If the ID number is invalid, controller 11 enables speech responder 7 to ~ ~-send, to telephone 1, a message "This ID number is , r.
lnvalid.", and cancels the telephone call.
If the ID number is valid, speech responder 7 sends to telephone 1, a message such as "Thls ID number l ;
ls valld, please lnput a tape recorder operatlon command.
Controller 11 calculates the proper input level of j : :
speech recognltlon unlt 9 on the basis of peak value C
of the DTMF slgnal (e.g., an average value or an effectlve value) detected by level`detector 6. Peak value C corresponds to state of slgnal loss of active ~ , telephone llne 2 ~that ls, as the loss ls larger, peak value C becomes smaller). When the proper level is ~ -calculated ln thls manner, controller 11 ad~usts the galn Qf PG~ 8 90 that a vo1ce signal hsvlng the , . . .......

20116~

- 12 - ~-calculated proper level is input to speech recognition unit 9.
After the input level of speech recognition unit 9 ~ `;
ls ad~usted, controller 11 enables speech recognition ~ `
unit 9 to allow the voice signal to be input through telephone line 2, NCU 4, and PGA 8. The voice signal of the operation command spoken at the handset of telephone 1 is input to, and recognized by, speech recognitlon unlt 9. The recognltlon result ls supplled to `~
controller 11.
Note that the sound (voice) of an input operation command is a word, e.g., ~RECORD~', "PLAYBACK", "STOP", ;~
"FAST FORWARD", "REWIND", or the llke. `~
If speech recognltion unit 9 employs an unspecified speaker word recognition system (see the above IEEE ~ -reference "Telephone Speech Recognition Using a Hybrid *
Methodll)~ the sound of any word can be recognized, :, i..:
resultlng in an ideal apparatus. However, when only one user uses the automatlc answering telephone, a speclfled speaker word recognition system, in which user's volce pattern data is registered in a RAM dlctlonary (not shown) to allow only the user to use the apparatus, may be employed.
Controller 11 sends a command corresponding to an operation to be instructed to logic-controllad type tape ;. . :., recorder 10 on the basis of the recognition result from speech recognition unit 9. Thus, tape recorder 10 ~ -.

2 0 1 1 ~ ~ 2 `~

. ,~ ,.', :',.. ...

executes the instructed operation. This operation includes mechanical operations.
In recording, when controller ll sends a corresponding command to tape recorder 10, tape recorder 10 executes a recording operation. A voice signal input from telephone line 2 is recorded by tape recorder 10. ;
On the other hand, the voice signal is also input ~ -to speech recognition unit g through PGA 8. Speech recognitlon unit 9 monitors a voice OFF time using timer ;.
9a which is enabled when the level of an input voice envelope decreases below a predetermined level. When ' timer 9a detects that no voice signal is input over a predetermined period of time, signal e~a, indicating thls, is generated to inform controller 11 of the rest state. In this case, controller 11 determines this state as a state of waiting for the next operation `
command, and sends a recognition operation start enable ;,~;
command to speech recognition unit 9.
Thereafter, the voice signal input from telephone line 2 is recognized by speech recognition unit 9, and the recognition result is sent to controller 11. ;, Controller 11 then sends an operation command to tape recorder 10 to execute the predetermined operation, `~
which in this case ls, e.g., a pause operation. i, Simllarly, when a fast-forward or rewlnd command ls input from outslde telephone l, and a playback command i ;
ls sent thereafter, the recorded content can be checked.
,, .. . ,.. ~ .

2 0 1 ~ ~ ~ 2 -. ;~... ~.

Tape recorder lO can be separated from the main telephone body, and function as an independent unit.
More specifically, when the user returns home from outside, he can hear the recorded content upon operation of tape recorder lO. The method of operating tape recorder lO is exactly the same as a conventional tape recorder.
In this manner, an ID number is input using a DTMF
slgnal which can be reliably discriminated. Upon receptlon of the DTMF signal, the signal loss state of the telephone line, i.e. the reception level of the DTMF
signal, is detected. A control command of the tape recorder of the automatlc answering telephone is input ~`
by a voice from an outside telephone, and is recognized by recognition unit 9. As a result, an automatic answering telephone which is very simple and easy to ` ~`~
operate cdn be obtained. The voice level input to ;;
speech recognition unit 9 is ad~usted to a proper value ~ ,~
on the basis of the reception level of the DTMF signal ~ ~ ~3`'~
~loss state of the telephone line)~ thus maintaining the input level of the speech recognitlon unlt ln the proper state, corresponding to the signal loss state of the telephone line. Thus, highly accurate speech recognition can always be assured.
As described above, according to the present invention, there can be provided a voice-controlled apparatus which can always maintain the input level to ,, , ~

201~6~2 a speech recognition means in a proper state when the speech recognition means is connected, and can assure highly accurate speech recognition. ;~ `
Note that block 10 in Fig. 1 is not limited to a tape recorder. For example, when the apparatus shown in Fig. 1 is equipped in an automatic teller machine in a bank, a user can inquire about his or her balance using an outdoor telephone. When block 10 is equipped in a computer-controlled automatic baker, a computer~
controlled bath, or a computer-controlled VCR, a user can send an instructlon (e.g.~ baklng, bath preparatlon, and recordlng of a TV program by the VCR) to the appara-tus shown ln Fig. 1 by words, without uslng difficult ,~ ~!:.'",'`,,,"'~',,,~
operation commands. ~;
,"''`,'"',, .,. ,... ~. ..
,~
~
;
" ' ''.".

' ~
',-'~"''"' ''. .'.'.

K ., ~
' '' ,~,.~'' ' ,'`.`:'t.
'` `"' ~ ' ''.'~."'~
~,'`'"."' `'"`',"`''' ` .
~ '. ' ;. ' ' '''~ ~'.i'"''

Claims (15)

1. A voice-controlled apparatus adapted to a telephone system, comprising:
means for receiving and discriminating a content of a DTMF signal sent from a telephone line of the telephone system; means for detecting a reception signal level of the DTMF signal sent to said discriminating means to provide a result indicative of the reception signal level of the DTMF
signal; means for recognizing content of voice signal sent from the telephone line; means for adjusting a reception signal level of the voice signal sent to said recognizing means in accordance with the result of the reception signal level of the DTMF signal detected by the detecting means, so that an adjusted voice signal level of the voice signal is provided which is sufficient to ensure a proper recognition of the content of said voice signal by the recognizing means; and means for executing a specific function for controlling an operation of a computer defined by the content of said voice signal recognized by said recognizing means.

2. A voice-controlled apparatus according to claim 1, wherein said executing means includes:

means for recording and playing back voice information; and means for controlling an operation of said recording/playing back means in accordance with the content of said voice signal recognized by said recognizing means.

3. A voice-controlled apparatus according to claim 1, wherein said executing means includes:

a computer-controlled equipment; and means for controlling an operation of said computer-controlled equipment in accordance with the content of said voice signal recognized by said recognizing means.

4. A voice-controlled apparatus according to claim 1, wherein said recognizing means includes means for monitoring discontinuity of a signal envelope of the voice signal sent from the telephone line, and for generating a discontinuity signal for indicating when a period of absence of the monitored signal envelope exceeds a predetermined period of time, and said executing means includes means for activating said recognizing means in response to the generation of said discontinuity signal.

5. A voice-controlled apparatus according to claim 1, wherein said detecting means includes means for detecting a peak level of said DTMF signal.

6. A voice-controlled apparatus according to claim 1, wherein said detecting means includes means for detecting an rms level of said DTMF signal.

7. A voice-controlled apparatus according to claim 1, wherein said detecting means includes means for detecting an average level of said DTMF signal.

8. A voice-controlled apparatus according to claim 1, wherein said adjusting means includes a multi-gain amplifier having a plurality of selectable gain factors, each of said gain being selected in accordance with the result of said signal level detection.

9. A voice-control method adapted to a telephone system, comprising the steps of:

discriminating a content of a DTMF signal received form a telephone line of the telephone system; detecting a reception signal level of the DTMF signal to provide a result indicative of the reception signal level detected;
recognizing content of voice signal sent from the telephone line; adjusting a reception signal level of the voice signal in accordance with the result of the DTMF signal level detection, so that an adjusted voice signal level is provided which ensures a proper recognition of the content of said voice signal; and executing a specific function for controlling an operation of a computer defined by the content of said voice signal recognized by said recognizing step.

10. A voice-controlled method according to claim 9, wherein said executing step includes the steps of:

recording and playing back voice information; and controlling an operation of said recording/playing back step in accordance with the content of said voice signal recognized by said recognizing step.

11. A voice-controlled method according to claim 9, wherein said executing step includes a step of controlling an operation of a computer-controlled equipment in accordance with the content of said voice signal recognized by said recognizing step.

12. A voice control method according to claim 9, wherein said recognizing step includes a step of monitoring discontinuity of a signal envelope of the voice signal sent form the telephone line, and generating a discontinuity signal for indication when a period of absence of the monitored signal envelope exceeds a predetermined period of time; and said executing step includes a step of activating said recognizing step in response to the generation of said discontinuity signal.

13. A voice-control method according to claim 9, wherein said detecting step includes a step for detecting a peak level of said DTMF signal.

14. A voice-control method according to claim 9, wherein said detecting step includes a step for detecting an rms level of said DTMF signal.

15. A voice-control method according to claim 9, wherein said detecting step includes a step for detecting an average level of said DTMF signal.