JPH03220959A - Voice guidance processing method - Google Patents

Abstract

PURPOSE:To prevent malfunction of a detector (PBR) by applying Fourier transformation to a voice guidance to be processed to be converted into a frequency region, attenuating a pushbutton signal band in the voice guidance along with a curve of a trigonometric function, applying inverse Fourier transformation to the signal so as to be restored into a time region. CONSTITUTION:A voice guidance 11 to be processed is Fourier-transformed by a Fourier transformation section 12 and the time region of the signal is transformed into a frequency region. Then power spectrum 13 of the voice guidance transformed into the frequency region passes through a comb-line trigonometric passing filter 18 attenuating the frequency of a PB in terms of Henning function. Finally, an output of the filter 18 is subject to inverse Fourier transformation by an inverse Fourier transformation section 16 to restore the frequency region into the time region thereby obtaining a processed voice guidance 21. Since the processed voice guidance 21 does not include the frequency used by the PB signal, even when the voice is inputted to a PBR, no malfunction is caused in the PBR.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a processing method for creating voice guidance used for network service guidance, etc.

[Conventional technology]

In order to use network services currently in use, such as message dialing, users usually dial a predetermined telephone number for that service, and then receive voice guidance prepared by the service side (center). The user listens to the message and presses a push button (hereinafter referred to as PB) to receive the desired service. For example, in the case of message dialing, the sequence is as shown in FIG.

In this way, after the audio guidance has been played, the system (center) detects the push button signal input by the user and performs the required service.

In such services, as users become accustomed to the service, they begin to press push buttons without listening to the voice guidance all the way through. To deal with this, it is necessary at the center to operate a push button signal detection device (hereinafter abbreviated as PBR) before the audio guidance ends playing.

[Problem to be solved by the invention]

However, current telephone networks always have echo back.

Therefore, if the PBR is operated before the end of audio guidance playback, this audio guidance will be input to the PBR by echo back, as shown in FIG. Voice has a wide frequency band, including the frequencies used for the PB double signal. Therefore, with a certain probability, while the audio guidance is being played, the PB signal frequency bone PBR is PB
This results in a malfunction such as double the signal and detection.

In order to avoid this, a conventional method has been to provide PBR with high accuracy. In other words, by incorporating a circuit that improves the distinguishability from the PB multiplied audio signal, the PB signal in the audio signal is prevented from being detected. However, since such PBR is *i, it tends to be expensive. In addition, there is a possibility that the opposite malfunction occurs, that is, a malfunction in which the signal is regarded as a PB multiplied audio signal and is not detected. Moreover, a large number of conventional inexpensive PBRs are already connected to the telephone network. It is practically impossible to replace these.

Furthermore, a method has been proposed in which the voice guidance is processed to attenuate the PB multiple signal frequency component. However, this method has the drawback of causing distortion in the voice guidance.

In other words, as shown in Fig. 4, the audio guidance (digitized) to be processed first is processed by the Fourier transform unit (
The signal is usually Fourier transformed using Fast Fourier Transform (FFT) 12 to transform from the time domain to the frequency domain. A comb-shaped pass filter 14 cuts only the frequencies used by the PB from the power spectrum 13 of the voice guidance. FIG. 5 shows the frequencies used by PB. PB is one frequency from the lower frequencies 697, 770, 852.941Hz and the upper frequencies 1209, 13
Each button is designated in 4×4=16 types by combination with one frequency from among 36, 1447, and 11633 Hz.

And each frequency band is 1.5%. The comb filter 14 attenuates this band.

However, although the frequency band of PB is ±1.5%, the center frequency of actual PBR is shifted because it includes analog system errors. Therefore, it is necessary to increase the bandwidth to some extent. In practice, it is necessary to attenuate ±3.0%. In other words, the frequencies to be attenuated are 676-718, 747-7
93, 826-877, 913-969゜1172
~1245.1296~1376, 1433~1521
．． It is 1584-1682Hz. Comb filter 14
The power spectrum 15 that has passed through the inverse Fourier transform section 1
6, the signal is inversely Fourier transformed from the frequency domain back to the time domain, and a processed audio guidance 17 is obtained.

The audio guidance processed by the above operations is PB
This results in audio that does not include the frequencies used by .

Therefore, even if this voice guidance is input to PBR, P
BR will not malfunction.

However, this processing causes deterioration in the quality of the voice guidance. In particular, a beeping sound that seems to contain a single component is generated.

This is thought to be caused by Fourier transform.

In other words, when FFT is used, the time domain is not completely transformed into the frequency domain, and the result is a convolution with a function obtained by Fourier transforming the wind function. This is usually 5 inches (
f)/f. Therefore, false sidelobe frequencies occur before and after the true frequency. When the attenuation curve is rectangular, the side lobe at the boundary frequency between the attenuated frequency and the non-attenuated frequency is considered to be particularly emphasized and audible because there are no surrounding frequencies.

[Means to solve the problem]

According to the invention, the audio guidance to be processed is Fourier transformed into the frequency domain, then the push button signal band in the audio guidance converted into the frequency domain is attenuated by a trigonometric curve, and finally , the frequency domain audio guidance whose push button signal band is attenuated is inversely Fourier transformed back to the time domain.

[For production]

As the curve of the trigonometric function, curves such as a Hamming function, a Hamming function, and a Blankman function pass through. Because the attenuation curve has such a curve, the cut-off characteristics are not sharp, but are sloped and gentle, and the side ropes do not bend or generate a beeping sound.

(Example) Next, the present invention will be explained in detail with reference to FIG.
In the figure, parts corresponding to those in FIG. 4 and FIG. 7 are given the same reference numerals.

First, the audio guidance 11 to be processed in the same way as in the conventional art is subjected to Fourier transform in the Fourier transform section 12 to transform from the time domain to the frequency domain. Next, in this embodiment, the power spectrum 13 of the voice guidance converted into a frequency range is passed through a comb-shaped trigonometric function pass filter 18 that attenuates the frequency used by the PB in a Hamming function manner. Finally, the output of this filter 18 is converted to an inverse Fourier transform unit 16.
Perform an inverse Fourier transform from the frequency domain to the time domain,
A processed audio guidance 21 is obtained.

The audio guidance 21 processed through the above operations becomes audio that does not include the frequency used by the PB. Therefore, even if this voice is input to the PBR, the PBR will not malfunction.

As shown in Figure 5, PB is the lower frequency 697°77
One frequency from 0.852.941Hz and upper frequencies 1209, 1336, 1447, 16
It is specified in 4 x 4 = 16 types by combination with one frequency from among 33 Hz. And each frequency band is 1.5%. The comb-shaped trigonometric filter 18 attenuates this band using a Hamming function (more precisely, a 1-Hamming function). In other words, this damping characteristic can be expressed by the following equation.

1(f)=(1,0+cos(2,Oz(f-fo
low)/fo range))/2.0However, fOl
ow <f<fo highfoIow =fOX(
1-range) fo high =rox (1+
range) fo range = fo high
-fo lowfO=697.770.852.941
．． 1209.1336°1447, 1633 Hz range=0.06 range is a value that determines the frequency band to be attenuated.

As can be seen from this function, this attenuation curve does not attenuate all of the frequencies used by the PR.

For example, when range=0.06 is selected, the level in the frequency band used by PB is as follows.

, fo*0.015 In other words, there are approximately 8 frequency bands used by PB.
% remains unblocked. However, with this 8% voice power, it is not recognized by PB in PBR. Therefore, by performing the processing shown in FIG. 1, it is possible to obtain voice guidance that is not recognized as a PB and has almost no quality deterioration.

Although this processing causes almost no deterioration in sound quality, the following may be used to further reduce the deterioration in sound quality.

(a) Change the value of r a n g e. The smaller this value is, the smaller the deterioration will be. However, if this value is too small, the cutoff will not be complete and the PBR will produce a sound that malfunctions, so cant-and-try is required in actual operation.

(b) Although the Hamming function was shown in the above example, a Blankman function (to be exact, 1 Blakman) is used instead of this function. It is known that the Blackman function has fewer sidelobes than the Hamming function. Therefore, it is estimated that this method causes less quality deterioration.

Blackman's function is as follows.

1(f)=(0,58+0.5cos(2,0x(f
-fo 1o@)/fo range)0.08cos
(2,0π2.o(f-fo low)/fo ra
(c) Cut only the lower four frequencies or the upper four frequencies used by the PB. PB is composed of a combination of one of the upper four frequencies and one of the lower four frequencies. Therefore, just by cutting either frequency, PBR will no longer operate.

(d) Cant only the frequencies of the PB types that are accepted by PBR. In reality, even if audio guidance is input to the PBR, not all PB frequencies will be recognized. Therefore, for example, if you actually input voice guidance into PBR and it recognizes, for example, °"*°゛, the frequencies 941Hz and 1336Hz in this voice guidance
Either or both can be cut.

Through the above operations, it is possible to create voice guidance with almost no deterioration in voice quality.

(Effects of the Invention) As described above, according to the present invention, it is possible to create voice guidance that does not include frequency components that cause PBR to operate and is free from distortion.As a result, even if this voice guidance is input to the PBH, the PBR will not be activated. No more malfunctions.

Therefore, the PB can be used even while this voice guidance is being played back. As an application field, it can be used to process voice guidance when the PB and voice guidance need to coexist in network services such as message dialing that use PB and voice guidance.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing an embodiment of the present invention, FIG. 2 is a diagram showing the sequence of message dials currently used,
Figure 3 is a diagram that simply shows the reason why voice guidance is input to PBR, Figure 4 is a block diagram that shows the proposed voice guidance processing method, and Figure 5 is a diagram that shows the frequencies used by PB. It is.

Claims (1)

[Claims] (1) In a telephone network service system in which a user controls a push button to reproduce voice guidance through a telephone line, and the user inputs a push button signal according to the reproduced voice guidance, a push button signal detector is connected to the telephone network. This is a method for processing voice guidance so as to avoid misrecognizing echoed voice guidance as a push button signal. First, the voice guidance to be processed is Fourier transformed to the frequency domain, and then the frequency The method is characterized in that the push button signal band in the audio guidance in the region is attenuated by a trigonometric function curve, and finally the audio guidance in the frequency domain in which the push button signal band has been attenuated is inversely Fourier transformed and returned to the time domain. Voice guidance processing method.