JPS58173797A - Voice recognition equipment for vehicle - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

[Detailed Description of the Invention] The present invention provides a jk-
When a command or information is input using the voice processing device EK16゜*a, it% is used to listen to the voice and determine the content of the command or information. 60 Figure 1 shows an example of a conventional IL Ryoro voice command device with a pull-out shaft load, and the voice commands to the microphone 4 phone 1 are noisy. and ji
IIIk and input as sound information, the 6 signal (10o) output from the microphone 1 is increased by -2 to an appropriate level, and the sound information signal @ (200) is sent to the level detector 3 and the audio. This will be done manually by WA Department 5. Level detection s3 is the sound ina III! (200) is detected and output as a level signal (aOO), this level tIs (300) is detected in audio detection*4.
Ill determine whether the manually inputted sound information is a voice command or not based on the microphone, and determine that it is a voice command.
When this occurs, a sound Fb koji noodle command signal (400) is output. As a result, the sound % noodle 8t15 performs the chest sound information signal (20G) to which this voice detailed command signal (40G) is output, that is, the voice command 1iist.

By the way, when the noise level is greater than l1EL, such as when a car passes through a tunnel, r( is the level difference between the voice command and the noise, which makes detection of the voice command difficult. It is necessary to speak quite loudly in front of the microphone (t6c), but when the voice level becomes this high, there is a problem in that saturation occurs and the dynamic range cannot be maintained.

Therefore, in order to solve this problem, we will introduce the second part! The vehicle chapter ALIL system shown in 6 has been considered. The same reference 1 character as 1- in 1 indicates the same configuration 1 minute, and in this hook, the level signal from the level generator 3 is connected to the field $18112 shown in 1- and the level detector 3. (3,0
0) is large, the sound information signal from gain 2 (200
), and conversely, reduce the gain of the level signal (30
0) is small, a gain controller 6 is provided to increase the gain of the sound information signal (200), and the rest is exactly the same as - in Fig. 1. Sound output from 2: N”6s (zoo)
When the gain is controlled by the gain controller 6 and input to the level detection section 3 and the voice weaving device 5, the level 41!
! The level signal -@ (30G) from the output section 3 is input to the audio detection section 4 and is also manually inputted to the gain controller 6 as a gain-vehicle signal. The gain controller 6 adjusts the gain of the sound information signal (200) output from the tombstone 2 to #11 based on this gain adjustment signal. voice relaxation lI
The sound of the child's voice in the capital 5 is performed in the same manner as in Figure 1. In this way, the gain of the sound information signal (200) is adjusted based on the level signal (30G) K from the level detection section 3. Therefore, even if the noise level becomes high, such as in the tunnel mentioned above, there is no need to make loud noises, and therefore, the dynamic range is maintained because static noise does not occur. However, the gain of the sound information signal (200) is adjusted even in the 6th section (hereinafter referred to as "sound section") where the sound%1 command is input, so the level difference between the sound information 1 bell and the noise level is still small. The single ms &! that constitutes the voice command becomes 81'
There is a risk that accent information and intonation information may be lost, and voice commands may not be recognized by the voice command.The present invention was made in consideration of the above point K. In the sound%SkojillllI4 device, which performs the voice line by adjusting the chest in response to the bell, voice commands are detected in order to prevent deterioration of voice introduction due to accents of voice commands or omission of intonation symbols 1.
K is set so that the gain of the sound information is kept constant during the interval.

The present invention will be explained below based on the drawings.

FIG. 3 shows an embodiment of a vehicle voice command device according to the present invention.
The same reference numerals as in FIG. 2 indicate the same components. In this embodiment, a hold circuit is provided to keep the gain of the gain controller 6 constant during the voice section.
This is exactly the same as the conventional example shown in the figure. In addition, level detection m
For sK, a rectifying and smoothing path with a time constant of 120 ms to 50 ms is used.

Next, let's join Tsuya 41 and perform this action ff1i!
Bright light.

Now, when *ti information is input to microphone 1, its output signal (100) is amplified by amplifier 2 and sound information tts@
(200), the waveform (
Section M is only noise, and section B is a combination of noise and voice commands) and is manually input to the gain controller 6. The sound information signal (200) is detected in the gain controller 6 in a section where its level is relatively small (lk4, Figure 0).
In the interval (m), the level signal (300) of the level generator 3 is
Gain mlk'rm @ (700) K-yori constant melody tl b'f4 >"C'tilllla h,
*4 Figure (0) K shows t yo 5 txf & 1ef) fi
It is output as l@ (600) and input to the level wanderer 3 and the speech recognition unit 5. Thus gain 1thiI
The emitted sound information htH is temporally averaged in the level detection 13 and a level signal (ao
O). The waveform of this level signal (300) becomes K as shown in FIG. 4(c)k. Level signal (30
0) is connected to the gain@adjustment Oi (7) via the hold path 7.
0G), the gain controller 6 manually inputs the gain of the sound information signal (200).
00) increases, but the level signal (300) becomes the quantity information signal (200) due to the time constant of the level detector 3.
cannot follow the rapid increase in the level of Therefore, the effect of the gain sea of the gain controller 6 can also be affected, and the level f(300) is increased by 6#t- and becomes VC. From this rising edge k, the voice detection 81 indicates that the sound %4 command was manually input! N disconnection, voice conclusion command signal (400
) is output 6゜This voice recognition command signal (400) is inputted with a voice introduction aMsecλ and voice recognition of the sound information signal (200) is performed 6o Meanwhile, at this time, the voice detection section 4 outputs λ and the hold signal (41G ) is output, and the gain leg rectification @ (700) in the hold path 7 is the level signal (30
It was held at the level of 0). Note that the waveform of the hold signal (4]0) output from the voice detection section 4 becomes a trap as shown in FIG. 4(e)k. The hold action by the hold path 7 is caused by the hold signal (410) being manually input (i.e., the hold section T) &<. At this time, hold i path 7
The waveform of the gain adjustment signal (700) output from is #3
It becomes a constant level as shown in Figure 4).

In this way, the sound information signal (200) is adjusted to a constant gain by the gain control (D-RA) 6 during the hold period T, so that the sound information signal (200) is adjusted to a constant gain K during the hold period T.
The level of the level signal (!100) does not become small when the number is opened, and therefore the level difference between the voice command level and the noise level does not become small. Therefore, there is no omission of the accent information or the initiation of the voice command that constitutes the voice command, and the voice command can be accurately performed. On the other hand, in the conventional voice recognition device, as shown in Figure 5, the level of the level signal (300) becomes small in the voice section T, so the level difference between the voice command level and the noise level becomes small. As a result, accurate speech recognition becomes impossible.

Fig. 6 shows another example of the vehicle voice control device according to the present invention.
This figure shows a block diagram of A, and the same reference numeral #1 as in FIG. 3 indicates the same component.

This actual jll1M is the gain controller 8 [several resistors R1+Rh...R, (however, R1-K・R1
歎K” ” Rs ” ””” K”-' R@s
K: Constant, B II! =1 + 2 m! ”')
The analog switch 80 that was not connected is used to adjust the gain 82 pictorially as shown in 6, and a single value T is connected between the level detection section 8 and the hold 1 path 7 roughly shown in No. 317.
Comparison of ringing RI-91 and -medullary TH2 (THI)
Comparator 92 with TI(2) and decision logic path 9
3, and the gain of the gain controller 8 is set to 1II.
The other configuration is exactly the same as the implementation shown in FIG. 3. Note that the gain controller 8 generates tkl#1. In order to prevent this, the stepwise increase in gain by the gain controller 8 is set to be smaller than the ratio of the two wa noodles.

Next, the operation of this example will be explained with reference to FIG.

Now, when sound information is input manually to microphone 1, its output @(100) is amplified by -42 and becomes sound information signal (20
0) is the waveform shown in Figure 7 H)K (section M is only noise, section @B is a combination of noise and voice command).
and is manually input to the gain controller 8. sound fIj
The lk signal (200) is input to the gain controller 8 when the level of the sound information signal is relatively small (fla (Fig. 7)).
In section m) of r), 1lIi is detected via level detection w63.
The gain l11 which is the control m@(9oo) of IIs9
m101 (700) K Yo') 111N2 with the determined gain, a signal with a waveform as shown in Figure 7 (B) K (
Boo) is output as level detection h3 and audio i [section 5
The level of the sound information signal which is inputted to 6o and thus has a gain of 1sila is detected by the level detection section 3,
Level Gaga! (300) is output. Level signal (
The waveform of step 30 (m) becomes as shown in FIG. In the control section 9, when this level signal (300) is about to exceed the threshold value TRI (1 point), it is output from the comparison 4191, and when the level signal (30G) is about to become smaller than -RETH2 (point b), it is output from the comparison 992. Output from. As a result, 1i
Comparison I from ulr Logistics Road-93! A 1-signal (SOO) having a waveform as shown in FIG. I111Im pieces @
When the signal (900) is manually inputted to the hold path 7, it is directly outputted as a gain adjustment signal (700), and is manually inputted to the analog switch 80 of the gain controller 8. The analog switch 80 is switched to thicken the gain of the sound information signal (200) when this gain-adjusted signal (700) is large, and conversely, the gain adjustment signal (700) is switched to thicken the gain of the sound information signal (200).
00) is small, the gain of the sound information signal (20G) is switched to be small. Therefore, the trc%11111 signal (SO
Since the gain is increased in the part where the level of O) is low, the level signal (30G) is 1.
Gradually increases between porcelain and bc until reaching the point kI
The Kfld @ Sane Otsu level signal (300
) becomes increasingly small at Marob @. Now, if h is the detection level of the voice command in the voice detection unit 4, then the level signal (30G) is 0 point as a result of the voice command or human input. This voice - Tashi Bell -
When the value exceeds 1, the voice detection *4 outputs a voice introduction command signal (400) in response to a voice command or a human input. This audio introduction 111J! The first order signal (400) is input to the voice signal generator 5, and voice recognition of the signal (SOO) based on the sound information signal (20G)K is performed. - At this time, the voice detection section 4-
The hold signal (41o) is output from *-k V
k; JJtm7 to 16-jEr'f in
The 0 signal is held at the level of the control signal (900) from the bass at the time the sound was detected. At this time, gain m* (700) output from hold path 7
The waveform of is as shown in FIG. 7(e)k. Therefore, even if the voice command is detected past the C point th, the analog switch s
O is the gain adjustment signal (W) immediately before EI when the voice command is detected.
OO), and the sound information signal (20G) is gain-adjusted by a constant gain determined by this switched state.

Considering the case where a voice command that starts up relatively slowly is input in the above voice introduction device 'IN,
At this time, the sound information signal (2
0G) (Section M is only noise, Section B is noise and sound % command, etc.) or manually. As can be seen from (7) and (E), the gain piece m signal (7
0G) is held from point C where the voice command is detected, increasing the gain of section A, so the gain controller 8 is adjusted to increase the gain of the 1m signal (200). 6. Therefore, gain control 9-
The waveform of the output signal (SOO) of La 8 is (b) in Figure 8 (a).
) The waveform of the level signal (300) becomes as shown in 8th (2) (a) (c) K. As can be seen from this figure, the level signal (300) exceeds the voice detection level b after point C where the voice command is detected, so that the voice command can be detected.

On the other hand, if we consider the case where six voice commands that rise relatively slowly are given manually in the embodiment shown in Fig. 3, the sound information signal shown in (1> in Fig. 8←) If a sound information signal (200) as shown in (a) of Fig. 8 (mouth), which is the same as (200), is input, level detection a3 and hold 1gl path 7 follow the slow rise of the voice command. , the gain controller 6 gradually decreases the gain after detecting the voice command.
118. Therefore, the output signal (600) of the gain controller 6 does not become large even after the voice command is detected, as shown in FIG. As a result, the waveform of the level signal (300) is
As shown in (c) of Figure (b), the level signal (3
00) does not exceed the voice detection level b even after the sound μ command is input, so the voice command cannot be detected even though the voice command is inputted. The gain controller' is 11R so as to change the gain of the sound information signal in stages.
*8 Since the sound command can be reliably detected even when a voice command with a relatively slow rise is input manually, the present invention is capable of adjusting the gain of the sound fascination according to the level of the sound information. 11 scales and i voice 1IllllIII,
In the voice s1 setting spear that performs this, the gain of the voice was kept constant for 6 minutes after detecting the voice command, so
While the voice command is being performed manually, the level difference between the level of the voice command and the logical level will not be 2 small, so the accent information and intonation of the voice command will be the same. is never missing, so Ken Koji can do all the voice commands. By changing the gain of the sound information step by step, it is possible to reliably detect even a voice command with a slow rise. .

[Brief explanation of the drawing]

Figures 1 and 2 show the conventional vehicle river audio details & blocks &lIi! 2% Figure 3 shows car audio B according to the present invention.
IIIk Koji Kazumi IIh Examples of the Puschik line (2) and the 4 (2) shown in the block diagram shown in the mS diagram, No. 6 f II shadow diagram in each 1111 of the block diagram, Fig. 5 0) are according to the present invention. The waveform of the level signal (300)-1 is the waveform (2) of the level signal (30G) according to the conventional example rC shown in the second example. The block lines □□□ and 餉7 of other implementations are the blocks shown in Figure 6.
The signal waveform diagrams, l1181V&) and 枦] are the signal waveform diagrams at each part of the block IIi diagram when a voice command with a relatively slow rise is input manually using the voice planing device II/r-1 for six car hooks according to the present invention. 6 in the waveform diagram
゜1...Microphone %2-'' amplifier, 3...Level detection section, 4...Audio detection section, 5...Sound%am section, 6
゜8...Gain controller, 7...Hold-path, 9...Il! l1lIIl Miyako, 80...Anna Prize Guswitch. Patent Input/Issuance Japan Automatic Co., Ltd. Patent Attorney Hiroo Suzuki

Claims (1)

[Claims] 1i1 information manually and outputs a sound information signal, a gain adjustment means for adjusting the gain of sound information No. 01 according to the level of the sound information signal, and a gain adjusted by the gain adjustment means. Sound information (the level of No. 1 exceeds the reference level) A sound detection means that outputs a 6tkl sound detection signal, and a sound information signal that is output by the sound detection signal and whose gain is adjusted. In the voice introduction WL having a cajoling means, the front voice detection H5 signal is outputted, and a gain holding means for holding the gain of the 6-related variable gain-vertical means is provided.
J'H11%1illli outfit.