JPH0138320B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a voice recognition device for a vehicle that accurately detects the end of voice.

2. Description of the Related Art Recently, voice recognition devices that recognize commands or information input by voice to determine the content of the commands or information have begun to be used in various fields, and are also being put into practical use in vehicles.

FIG. 1 shows a block diagram of a conventional vehicle voice recognition device. In the figure, 1 is a microphone that inputs voice commands as sound information by superimposing them with noise; 2
3 is an amplifier that amplifies the signal 100 output from the microphone 1 to an appropriate level and outputs the sound information signal 200, and 3 is a rectifier circuit that rectifies the sound information signal 200, but a square circuit may also be used. 4 is rectifier circuit 3
5 is a smoothing circuit (with a time constant of about 20 to 30 ms) that smooths the rectified signal 300 and outputs it as a smoothed signal 400 using, for example, a low-pass filter, in order to temporally average the rectified signal 300. a threshold value setter for setting a threshold value as a criterion for determination;
6 is a smoothed signal 400 from the smoothing circuit 4 and a threshold value 50
A comparator 7 outputs a voice detection signal 600 when the signal 400 is larger than a threshold value 500, and 7 determines whether the sound information input to the microphone 1 is voice based on the output of the comparator 6. A determining unit 8 analyzes the sound information signal 200 using a filter or the like based on the voice recognition command signal 700 output when the determining unit 7 determines that it is a voice. , a voice recognition unit that performs voice recognition. Incidentally, signal waveforms at each point in the block diagram shown in FIG. 1 are shown in FIG.

In the vehicle voice recognition device, sound information is input to the microphone 1, and a signal 100 output from the microphone 1 is amplified by the amplifier 2 and input to the rectifier circuit 3 and the voice recognition section 7 as a sound information signal 200. The waveform of the sound information signal 200 becomes as shown in FIG. This smoothed signal 400 is input to the comparator 6 together with a threshold value 500, and their signal waveforms become as shown in FIG. The waveform of the voice detection signal 600 output from the comparator 6 is as shown in FIG. 2C. As can be seen from the figure, when the voice detection signal 600 is output for a predetermined time t ₁ (for example, about 150 ms) or more, the time point beyond which the voice detection signal 600 is output is detected as the start of the voice. After that, when the voice command is interrupted momentarily and the sound information signal 200 decreases, the smooth signal 400
may once become smaller than the threshold value 500,
Since the time T in between is less than t ₂ (for example, about 300ms), it is determined that the voice command continues, and the determination unit 7
The voice recognition command signal 700 remains outputted, and the voice recognition section 8 continues to perform voice recognition.

When the voice command ends and the voice detection signal 600 from the comparator 6 is not output for a predetermined time _t2 or more, that time is detected as the end of the voice. In this way, the determining unit 7 determines whether or not it is a voice based on the voice detection signal 600 from the comparator 6, and outputs the voice recognition command signal 700 as a result.
The waveform of is as shown in Figure 2 D. In this way, the detection of the beginning of the voice is delayed by time t ₁ , and the detection of the end is delayed by time
However, the speech recognition unit ₈ is provided with a memory that constantly inputs and stores the signal 200 from the amplifier 2, and the above-mentioned delay time t ₁ or t ₂ is stored in this memory.
Since the audio information input during this period is stored, there is no loss of audio information due to the time delay as described above.

In such a speech recognition device, a fixed threshold value is provided to detect whether or not the sound information input to the microphone 1 is voice, and the level of the sound information signal is compared with the threshold value to detect the sound information. Because it was determined that audio was input when the signal level exceeded a threshold, when the noise level is extremely high, such as when a car passes through a tunnel, the sound information is Since the signal level increases and exceeds the threshold value, it is erroneously detected that audio has been input, making it impossible to detect the beginning and end of the audio. To solve this problem, if the above fixed threshold is made variable according to the sound information signal level, the level of the sound information signal will not exceed the threshold even if the noise increases, and the sound information signal level will not exceed the threshold. Although it is possible to detect the beginning of the voice because the threshold value is sometimes exceeded, the threshold value gradually increases depending on the level of the sound information signal in order to smooth the sound information signal with a large time constant. However, the sound level changes rapidly because the sound information signal is smoothed with a time constant smaller than the threshold, and for this reason, there is a risk that the judgment unit alone may detect the end of the sound even though it has not yet ended. Ta.

The present invention has been made in view of the above-mentioned points, and includes a voice recognition device for a vehicle that determines that voice has been input while the sound information level in the vehicle exceeds a reference signal that changes according to the level. In order to accurately detect the end of the reference signal, the reference signal is held at at least one of the time when the smoothed signal exceeds the reference signal or the time when the start end is detected.

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

FIG. 3 is a block diagram showing one embodiment of the vehicle voice recognition device according to the present invention. In the figure, the same reference numerals as in FIG. 1 indicate the same components, and the difference between this embodiment and the conventional example shown in FIG. 1 is that the smoothing circuit 9 has a time constant of 100 to 2000 ms,
An amplifier 1 that multiplies the output signal 900 from the smoothing circuit 9 to an appropriate level and outputs a reference signal 1000.
0 and the reference signal 100 output from the amplifier 10
A hold circuit 11 that passes 0 as is or holds the current level by a hold signal to be described later and outputs it as a signal 1100.
The point is that

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

Sound information is input to the microphone 1, and a signal 100 output from the microphone 1 is input to the rectifier circuit 3 and the voice recognition section 8 via the amplifier 2. Signal 3 rectified by rectifier circuit 3 and output
00 is input to the smoothing circuit 4 and the smoothing circuit 9.
The signal 300 input to the smoothing circuit 4 is averaged with a small time constant, outputted as a smoothed signal 400, and inputted to the comparator 6. On the other hand, the signal 300 input to the smoothing circuit 9 is averaged with a large time constant to become a signal 900, which is amplified by the amplifier 10 and output as a reference signal 1000, and then input to the comparator 6 as it is via the hold circuit 11.

By the way, when audio is input and the sound information signal level increases, the smoothed signal 4 output from the smoothing circuit 4
Although the level of 00 immediately rises, the output signal 1100 of the hold circuit 11 cannot follow the rapid increase in the level of the sound information signal because the time constant of the smoothing circuit 9 is considerably larger than the time constant of the smoothing circuit 4. Therefore, the smoothed signal 40 is larger than the output signal 1100.
When 0 is input, the comparator 6 outputs the voice detection signal 60.
0 is output, and as a result, it is determined that voice has been input, and the voice recognition command signal 700 is output from the determination unit 7. This voice recognition command signal 700 is input to the voice recognition section 8 and performs voice recognition of the sound information signal 200. In this way, the beginning of the voice can be reliably detected. On the other hand, the voice recognition command signal 700 from the judgment unit 7 is also input as a hold signal to the hold circuit 11, and due to the operation of the hold circuit 11, the reference signal 1000 from the amplifier 10 is output at a time point _t1 after the voice is input. The signal 1100 at that time becomes the threshold for detecting the end of the voice. Smoothed signal 400 input to comparator 6
The threshold value 1100 and the waveform of the reference signal 1000 output from the amplifier 10 are as shown in FIG. 4A. As can be seen from the figure, when the voice command is interrupted momentarily and the sound information signal 100 decreases, the smoothed signal 400 may temporarily become smaller than the threshold value 1100, but if the time T during that time is less than _t2 , the judgment unit 7 From then on, the voice recognition command signal 700 continues to be output, and the voice recognition section 8 continues to perform voice recognition. When the voice command ends and the signal 400 becomes smaller than the held threshold 1100 and this state is maintained for a time _t2 or more, the time
Detect the end of the audio after _t2 . However, without the hold circuit 11, the level of the threshold value 1100 would rise as shown by the dashed line in Figure 4A, and the smoothed signal 4 would rise before the audio ends.
Since 00 becomes smaller than the threshold value 1100, that point will be mistakenly detected as the end of the audio. Furthermore, since the threshold value rises at the level of the voice itself, accurate termination detection cannot be performed. Note that the waveform of the voice detection signal 600 output from the comparator 6 is as shown in FIG. 4B.

Next, based on the voice detection signal 600 from the comparator 6, the determination unit 7 determines whether or not it is voice, and when it is determined that it is voice, the waveform of the voice recognition command signal 700 is output. As shown in FIG. 4C, there is a time delay at the beginning of the voice by time _t1 and at the end by time _t2 , but as in the conventional example, the voice recognition unit 8 prevents the loss of voice due to such a time delay. It can be easily solved.

FIG. 5 is a block diagram showing another embodiment of the vehicle voice recognition device according to the present invention. In the figure, the same reference numerals as in FIG. 3 indicate the same components, and the difference between this embodiment and the embodiment shown in FIG. OR circuit 1 which receives the recognition command signal 700 and outputs it as a hold signal 1200
2. The operation of this embodiment is similar to that of the embodiment shown in FIG. 3 from when sound information is input to the microphone 1 until it is input to the comparator 6. When the voice detection signal 600 is output by the comparator 6, it is immediately inputted to the hold circuit 11 as a hold signal 1200 via the OR circuit 12, so that the reference signal 1000 output from the amplifier 10
is held at the level at the time when the hold signal 1200 was input, that is, when the audio was input, and output as the threshold value 1100. Smoothed signal 400 and threshold value 1 input to comparator 6 at this time
100 and the reference signal 10 output from the amplifier 10
The waveform of 00 is as shown in FIG. 6A. The waveform of the voice detection signal 600 output from the comparator 6 is as shown in FIG. 6B. On the other hand, comparator 6
Based on the voice detection signal 600 output from the voice detection signal 600, the determination unit 7 determines whether or not it is voice, and when it is determined that it is voice, the waveform of the voice recognition command signal 700 output is shown in FIG. 6C. As shown, the hold circuit 11 is connected via the OR circuit 12.
However, at this time, the hold circuit 11 is already held by the voice detection signal 600 from the comparator 6, so even if the voice recognition command signal 700 from the judgment unit 7 is input to the hold circuit 11, it will not be in the hold state. is maintained as it is, and detection of the start and end of the voice is performed in the same manner as in the embodiment shown in FIG. As described above, in this embodiment, the threshold level is held at the time when the voice detection signal 600 from the comparator 6 is output, that is, at the time when voice is input, and there is no increase in the threshold level due to voice. Therefore, the held threshold level is the level of the voice that exceeds the noise, and if the smoothed signal 400 falls below this level, it can be reliably detected as the end of the voice. Thereafter, the voice recognition command signal 700 is not output from the determining section 7, so the hold signal 1200 is not input to the hold circuit 11, and as a result, the hold operation is canceled.

Next, as can be seen from FIG. 6C, there is a time delay at the beginning of the audio by time t ₁ and at the end by time t ₂ , but as in the first embodiment and in the prior art, no loss of audio due to such a time delay occurs. It can be easily solved.

As explained above, in the present invention, the reference signal is smoothed in the speech recognition device that determines that voice has been input while the sound information level in the vehicle exceeds the reference signal that changes depending on the level, that is, the threshold value. Since the reference signal is held at least at the time the signal exceeds the threshold or the start point is detected, the sound information level will be lower than the threshold while the audio is being input due to the threshold being raised by the audio itself. Therefore, not only the starting end but also the ending end can be detected accurately.

[Brief explanation of drawings]

FIG. 1 is a block diagram of a conventional vehicle voice recognition device, FIG. 2 is a signal waveform diagram at each part of the block diagram shown in FIG. 1, and FIG. 3 is a block diagram of a vehicle voice recognition device according to the present invention. Diagram, Figure 4 is the 3rd
FIG. 5 is a block diagram of another embodiment of the vehicle speech recognition device according to the present invention, and FIG. 6 is a signal waveform diagram at each part of the block diagram shown in the figure. It is a signal wave diagram in each part. DESCRIPTION OF SYMBOLS 1... Microphone, 2... Amplifier, 3... Rectifier circuit, 4, 9... Smoothing circuit, 5... Threshold value setting device, 6... Comparator, 7... Judgment part, 8... Speech recognition part, 10... Amplifier, 11... Hold circuit , 12...OR circuit.

Claims (1)

[Claims] 1 Sound information input means for inputting sound information in the vehicle interior and outputting a sound information signal; rectifying means for rectifying the sound information signal; and outputting a smooth signal by smoothing the sound information signal rectified by the rectifying means. smoothing means to
smoothing/amplifying means for smoothing and amplifying the sound information signal rectified by the rectifying means with a time constant larger than a time constant of the smoothing means to output a reference signal; audio input determining means that determines the start of the sound information when the smoothed signal continues for a second predetermined time, and determines the end of the sound information when the smoothed signal continues below the reference signal and continues for a second predetermined time; recognition means for recognizing the sound information signal in the section; and holding means for holding the reference signal at at least one of a point in time when the smoothed signal exceeds the reference signal or a point in time when the starting edge is detected. A vehicle voice recognition device characterized by: