Note : Les descriptions sont présentées dans la langue officielle dans laquelle elles ont été soumises.
2045959
SPFECX RECOGNITION APPARATUS
BACKGROUND OF T~: INVENTION
The present invention relates to a speech recognition
apparatus for use in a speech recognition answering system and,
more particularly, to a speech recognition apparatus having a
5 PB (PUSH BUTTON) signal detecting or recei~ing function.
A conventional speech recognition apparatus of the type
described has a speech recognition unit (SRU) and a PB sinal
receognition unit or PB receiver ~PBR), but it cannot determine
whether an input signal from a telephone channel is a PB signal
10 or whether it is a speech. It has been customary, therefore, to
assign an independent telephone channel to each of PB input and
speech input. A business application or similar software using
the apparatus monitors the telephone channels to determine
which of the telephone channels has received a call and
15 commands the apparatus to use only one of the recognition units
SRU and PBR associated with the telephone channel of interest.
Moreover, the conventional speech recognition apparatus
with a PB receiving function forces the business application to
execute processing matching the independent telephone channels.
2 0 On the other hand, the user has to select either on of two
'~
B~
2045 959
different telephone numbers assigned to speech input and P
input, resulting in limited serviceability. In a system wherein
the apparatus is expected to call the user, the user has to
register desired one of the speech input and PB input at the
5 system beforehand. Further, since the channels and the kinds of
input signals are fixedly held in one-to-one correspondence, an
idle channel cannot be efficiently assigned. Specifically, when
calls concentrate on either one of the speech input and PB input
channels, the user cannot take full advantage of the service.
SUMMARY OF THE INVENTION
It is therefore an object of the present invention to
provide a speech recognition apparatus capable of automatically
determining whether an input signal is a speech signal or whether
15 it is a PB signal.
It is another object of the present invention to provide a
generally improved speech recognition apparatus.
A speech recognition apparatus of the present invention
comprises a speech reco~nition unit for recognizing a speech
2 0 from an input signal and outputting the result of recognition, a
PB signal detection unit for detecting a PB signal from the input
signal and outputting the result of detection, and a control unit
for controlling the speech recognition unit and PB signal detection
unit to automatically determine whether the input signal is a
2 5 speech signal or whether it is a PB signal on the basis of the
~`
.,,
~3~ 2045959
result of recognition and the result of detection which the speech
recognition unit and PB signal detection unit output when used at
the same time.
BRIEF DESCRIPTION OF THE DRAWINGS
The above and other objects, features and advantages of
the present invention will become more apparent from the
following detailed description taken with the accompanying
drawings in which:
FIG. 1 is a block diagram schematically showing a
speech recognition system implemented with a speech recognition
apparatus embodying the present invention;
FIG. 2 is a flowchart demonstrating a specific operation
of the system shown in FIG. l; and
FIGS. 3, 4 and 5A through 5G are flowcharts showing
specific procedures which the embodiment executes for automatic
s ~
-
2~4~
DESCRIPTION OF THE PREFERRED EMBODIMENT
Referring to FIG. 1 of the drawings, a speech
recognition system implemented with a speed recognition
apparatus embodying the present invention is shown. As
5 shown, a subscriber's telephone 6 is connected to a channel
control 1 via first and second exchanges 7 and 8. The channel
control 1 sends an input signal to a speech recognition apparatus
10 under the control of a business application 2. The speech
recognition apparatus 10 has a speech recognition unit (S~U~ 3,
10 a PB signal or dial tone recognition unit (PBR), and a control
unit 5.
The operation of the system will be described with
reference also made to FIG. 2. When a person using the system
originates a call on the telephone 6, the call is sent to the
channel control 1 via the exchanges 7 and 8 (step 2 01 ) . In
response, the channel control 1 informs the business application
2 of the arrival of the call. On receiving a call termination
command from the business application 2, the channel control 1
connects the channel to the speech recognition apparatus 10
2 0 (step 2 0 2 ) and then informs the business application 2 of the
end of call termination. In response, the business application 2
notifies the control unit 5 of the apparatus 10 of the number of
words of an input signal to be recognized (step 2 0 3 ) . The
number of kinds of input signals to be recognized at the
2 5 same time are equal to the number of kinds of PB dials
of the telephone, and
~'
~5~ 2045959
most of them are numerals. Hence, in the illustrati~e
embodiment, let the number of words be treated as figures or
digits hereinafter.
On recei~ing the digit command from the business
application, the control unit 5 enables the SRT 3 and PBR 4 (step
204) so as to recognize input signals from the channel control 1
at the same time, thereby automatically discriminating the input
signals (step 2 0 5 ) . When the predetermined number of figures
have been recognized, the control unit 5 delivers the results of
recognition to the business application 2 (step 206).
The automatic discrimination of input signals which is
the characteristic feature of the present invention will be
described in detail. Preconditions for the automatic
identification are as follows:
(1 ) The detection rate of the PBR 4 is substantially
100 % while the recognition rate of the SRU 3 is less than 100 %;
and
(2) There is no user who uses speech and PB together.
FIG. 3 shows the details of the automatic discrimination
step 205 of FIG. 2 which the control unit 5 of the apparatus 10
executes. Implemented by a microProcessor, for example, the
control unit 5 enables the SRU 3 and PBR 4 for the first digit in
order to effect simultaneous recognition (step 204, FIG. 2).
Then, the control unit 5 sets a predetermined time (Tl ) in a
timer built therein. If the result of recognition of the first digit
~'~
.
-6- 2 0 4 5 9 5 9
is returned from the PBR 4 first (step 302), the control unit 5
immediately determines that it is the result of simultaneous
recognition of the first digit (step 311) since the recognition rate
of the PBR 4 is considered to be 100 %. At this instant, the
5 control unit 5 disenables the SRU 3. In the event of multi-figure
input, the control unit 5 determines that the second and
successive digits are not a speech on the basis of the previously
stated precondition (2) and, therefore, executes the processing
only with the PBR 4, i. e., without simultaneous recognition
10 (step 312). When the result of recognition of the first digit is
returned from the SRU 3 first (step 303), the control unit 5
waits a predetermined period of time (T2) to see if a result from
the PBR 4 is not really returned. For this purpose, the control
unit 5 sets the time T2 in a timer independent of the timer
15 assigned to the time Tl.
The time T2 should not be longer than about 1. 5 seconds
at most since it delays the processing time. If the PBR 4 returns
an answer to the control unit 5 within the time T2 (step 308),
the control unit 5 determines that the SRU 3 has misrecognized
20 due to noise or similar cause. Then, the control unit 5 regards
the result from the PBR 4 as the result of simultaneous
recognition of the first digit and disenables the SRU 3 (step
311). In the case of multi-figure input, the control unit 5 uses
only the PBR 4 in effecting recognition (step 312). If an answer
25 from the BPR 4 is not returned within the time T2 as determined
B
~7~ 204~959
in the step 3 0 8, the control unit 5 determines that the input
signal is a speech (step 3 0 9 ) and recognizes the second and
successive digits only by the SRU 3 (step 310). On completing
the recognition of the predetermined number of figures, the
5 control unit 5 sends the results to the business application 2
(step 2 0 6 ) . If the SRU 3 does not return an answer as
determined in the step 3 0 3, the control unit 5 determines
whether the time T1 has expired or not (step 304) and, if it has
expired, ends the processing while informing the business
10 application 2 of the expiration (step 306). If the time Tl has
not expired as determined in the step 304, the program returns
to the step 3 0 2 to see if the PBR 4 or the SRU 3 returns an
answer.
By the above procedure, input signals are automatically
1 5 discriminated.
As stated above, the illustrative embodiment recognizes
the first digit and, base on the result of this recognition,
recognizes the second and successive digits by either one of the
SRU 3 or the PBR 4. This is successful so long as only the
2 0 recognition unit associated with the input signal responds
correctly. In practice, however, it sometimes occurs that both
of the recognition units respond. Then, this embodiment
effecting simultaneous recognition would malfunction.
Generally, the two different recognition units may respond at the
25 same time under either one of the following two situations:
B~
-8- 2045959
(a) The PBR 4 also responds to a speech input; and
(b) The SRU 3 also responds to a PB input.
The above occurrence (a) is unavoidable although rare.
Hence, considering that the probability that the occurrence (a)
5 continues is low, it is determined that the input signal is PB if
the PBR 4 responds within a predetermined plurality of digits.
In the event of the occurrence (b), the input signal is determined
to be PB since the recognition rate of the PBR 4 is considered to
be 10 û %. However, when PB involving noise or speech is
10 recognized by the SRU 3, the PBR 4 is apt to return a result
after the SRU 3. To eliminate this problem, a result from the
PB~ 4 may be waited for after the return of a result from the
SRU 3. However, this implementation is not fully satisfactory
since the waiting time delays the response and, therefore,
15 cannot exceed a certain limit. A procedure which promotes more
accurate automatic identification consists in determining, every
time the SRU 3 returns a result, whether or not the PBR 4
returns an answer and regarding the input as PB if the PBR 3 has
returned an answer as to two or more digits. FIG. 4 shows a
20 sequence of steps for practicing such a procedure.
The procedure shown in FIG. 4 corresponds to the step
205 shown in FIG. 2. Specifically, both the SRU 3 and the P~3R 4
are enabled. First, the control UIlit 5 sets in a first timer an
input time T0 associated with the number of figures which is
25 instructed by the business application 2. If the PBR 4 returns a
~.
, .~
2045959
g
result of recognition of the first digit before the SRU 3 (step
402), the control unit 5 regards it as a result of simultaneous
recognition on the first digit by considering that the recognition
rate of the PBR 4 is 100 %. If it is the SRU 3 that has returned
5 a result first (step 403), the control unit 5 waits a
predetermined period of time (Tl) to see if the PBR 4 does not
really return an answer (step 406). Again, this waiting time T1
should not be longer than about 1. 5 second so as not to delay
the processing. If the PBR 4 returns an answer within the period
10 of time T1, the control unit 5 regards the result from the PBR as
a result of simultaneous recognition of the first digit by
determining that the SRU 3 has misrecognized due to noise or
similar cause (step 411).
If the first digit is PB as determined in the step 411, the
15 answer of a step 412 is NO without exception since the number of
times that the PBR 4 answers is unconditionally once. Then, the
next digit is recognized (step 413) . At this time, the control
unit 5 does not enable the SRU 3 and waits for a result from the
PBR 4 for a given period of time (T2) (step 414). Assuming
20 that the user of the telephone 6 presses the keys on the telephone
6 slowly, the period of time T2 is the interval between the
successive operations of the keys, e. g. 1 second to 3 seconds.
If the PBR 4 returns a result within the period of time T2, the
control unit 5 regards it as a result of simultaneous recognition
25 of the second digit (step 417). If the result of recognition from
..t
2045959
--1 o--
the PBR is not on the last digit (step 418 ), the control unit 5
recognizes the succeeding digit or digits with the PBR 4 only,
i. e., it does not execute simultaneous recognition (step 419).
If the PBR 4 does not return a result within the period of time T2
as determined in the step 414 and if the SRU 3 has returned a
result on the first digit (step 415), the control unit 5 replaces
the result on the first digit with the result from the SRU (step
416) and enables the SRU 4 (step 410).
On the other hand, if the SRU 3 has returned a result on
the first digit first (step 403) and if the PBR 4 has not returned
a result within the waiting time T1 (step 406), the control unit
determines that the input is not PB since the PBR 4 is free from
misrecognition. Then, the control unit 5 regards the result from
the SRU 3 as a result of simultaneous recognition of the first
digit (step 407).
In the illustrative embodiment, when the answers from
the SRU 3 and PBR 4 exist together, the results of recognition
are replaced with each other, depending on the situation, as
follows:
(i) First replacement: When the PBR 4 returns an
answer as to two or more digits during the recognition of a
plurality of digits, the results having been returned from the PBR
4 are substituted for the results of recognition; and
(ii) Second replacement: Assume that after a result
from the PBR 4 on a given digit has been regarded as a result of
.
.. ~
204~959
recognition, the PBR 4 does not return an answer as to the next
digit within the period of time T2. Then, if all the results of
recognition up to the digit of interest are the results from the
SRU 3, the results from the SRU 3 are substituted for the results
of recognition.
A reference will be made to FIGS. 5A through 5G for
describing the answers from the SRU 3 and PBR 4 and the results
of recognition on the assumption that five digits are sequentially
recognized. As shown, when the SRU 3 answers first as to the
first digit (step 403) and the PBR 4 does not answer within the
period of time T1 (step 406), the result S1 from the SRU 3 is
determined to be the result of recognition of the first digit (step
407 and FIG. 5A). As the SRU 3 answers first as to the second
digit also and the PBR 4 does not answer, the output S2 of the
SRU 3 is determined to be the result of recongnition of the second
digit (FIG. 5B). However, regarding the third digit, the PBR 4
answers before the SRU 3 (step 402), so that the output P1 of
the PBR 4 is determined to be the result of recognition of the
third digit (step 41 and FIG. 5C). Since the PBR 4 answers only
once, the step 412 is followed by the step 413 for recognizing
the next digit. Assume that thepBR 4has not returned an answer
within the period of time T2 (step 414). Then, the control unit
5 determines whether or not the SRU 3 has responded to the
third digit (step 415). If the answer of the step 415 is YES, the
above-mentioned replacement (i) is effected to substitute all of
. . .
--1 2--
2045959
the outputs of the SRU 3 having been returned from the results
of recognition up to third digits (step 416 and FIG 5D). Since
the SRU 3 has to be enabled digit by digit, the control unit 5
enables it to process the next digit (step 410). Assume that the
5 PBR 4 has responded as to the next digit also (step 4 0 2 ) .
Then, the result from the PBR 4 is selected as a result of
recongition of the digit of interest (step 411). Since the PBR has
responded twice as counted from the beginning of the
processing, the control unit 5 determines that the input signals
10 are PB and, therefore, effects the replacement (i) (steps 412
and 417 and FIG. 5E). After such a decision, the control unit 5
executes recognition with the succeeding digits by using only the
PBR 4, i. e., without enabling the SRU 3 (step 419 and FIG.
5F). On determining the results of recognition of five digits
(step 418 and FIG. 5G), the control unit 5 sends them to the
business application (step 2 0 6 ) . Such a procedure enhances
more accurate identification of input signals. It is to be noted
that the number of digits to which the PBR responds as
determined in the step 412 is not limited to two and may be three
20 or more.
In summary, it will be seen that the present invention
provieds a speech recognition apparatus which has a speech
recognition unit for identifying a speech from an input signal, a
PB recognition unit for detecting a PB signal from an input
2 5 signal, and a control unit capable of automatically determining
! B
-13- 2045~59
whether an input signal is a speech signal or whether it is a PB
signal. The apparatus, therefore, makes it needless for a
business application which controls it to discriminate a PB signal
and a speech signal. As a result, the customer intending to use
S an inquiry service, for example, does not have to discriminate
the telephone number for voice input and the telephone number
for PB input. In the case of information service, it is not
necessary for the customer to register at the system regarding
the PB/voice input. Hence, the apparatus enhances the
10 serviceability of the system. Further, the system sets up
efficient traffic since both of PB input processing and voice input
processing are implemented by a single telephone channel,
exhibiting the processing ability to the full extend.
Various modifications will become possible for those
15 skilled in the art after receiving the teachings of the present
disclosure without departing from the scope thereof.
....~
