JPH0443399A - On-vehicle voice recognition device - 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] Limb tips alternate The present invention relates to a vehicle-mounted voice recognition device.

2. Description of the Related Art Techniques for controlling devices inside a car every moment using voice commands are becoming commonplace. For example, JP-A-59-1090
In Publication No. 96, while the microphone switch for inputting voice data to the voice recognition device is turned on, or
After the microphone switch is turned ON, the device's output is stopped for a certain period of time to prevent malfunctions and improve recognition rates. In this case, the device to be compared is, for example, a radio, a cassette deck, an air conditioning device, a car phone call, etc., and the driver (user)
The voice uttered by the person is recognized by the voice recognition device, and the device that is roughly associated with the result is operated.

A speaker-specific system and a speaker-independent system are used as speech recognition devices. In the case of the speaker-specific system, the dictionary is registered by the user, but in the case of the speaker-independent system, it is built into the device.

In general, a speaker-independent dictionary is created from a voice database that records the voices of many people.

Voice databases are recorded in environments such as studios, and microphones such as vocal microphones are often placed close to the mouth. This is to record audio in a state close to the normal usage state (for example, using a headset microphone) and to improve the quality of the audio. However, when using a voice recognition device in a car,
Using a headset is unlikely. Therefore,
The microphone is used away from the mouth (20-30C!m), but the frequency characteristics of the microphone change as it moves away from the mouth.

Figure 4 shows the frequency characteristics near the mouth, and Figure 5 shows the frequency characteristics when the mouth is away from the mouth.As is clear from these figures, differences occur mainly at frequencies above several kHz; When the microphone moves away from the mouth, the gain of the high frequency response becomes higher than that of the low frequency range.

Furthermore, since it is inside a vehicle, the frequency characteristics change depending on the shape of the vehicle interior. In studios and other places, sound reflections from audio walls are prevented from entering the microphones, but the inside of a car is made up of a lot of glass and is small, so it is easy for sound reflections to enter the microphones.

Therefore, differences in high frequency characteristics tend to occur depending on the shape of the vehicle interior. As mentioned above, the dictionary used in the speaker-independent system is recorded in a studio, so the frequency characteristics differ from the actual voice inside the vehicle, and the pattern of input cover turns differs from the expected pattern. In particular, u F P” sound,
Since the 'H' sound does not have a local peak, the pattern is different and the recognition rate is lowered.

■--The present invention was made in view of the above-mentioned circumstances.
In particular, by correcting the frequency characteristics of a single uttered voice, the purpose of this technique is to obtain the expected input pattern and improve the recognition rate.

In order to achieve the above object, the present invention provides an in-vehicle speech recognition device that includes (1) a microphone, a microphone amplifier, an AGC, a band pass filter, a detector, a low pass filter, and an A/
In an in-vehicle speech recognition device having a D converter, a correction table, a multiplier, a speech interval detection section, a feature extraction section, a speech pattern memory, a recognition section, a speaker-independent dictionary, a recognition result output section, and a load active section, In order to approximate the frequency characteristics of the input voice uttered in the voice database to the characteristics recorded in the voice database used when creating the speaker-independent dictionary, the channel power data is multiplied by a constant recorded in the correction table. The feature is that recognition is performed using the power data of each channel obtained as a result.
Furthermore, (2) in (1) above, several types of correction tables are provided and can be switched by a switch for each vehicle type, or (3) in (1) above, correction tables are provided in a cart for each vehicle type. This is what I did.

Hereinafter, the present invention will be explained based on examples.

FIG. 1 is a configuration diagram for explaining one embodiment of the present invention. In the figure, 1 is a microphone, 2 is a microphone amplifier, and 3 is an AGC (
automatic gain adjustment device), 4□ to 4□ are band pass filters (BPF) and detectors (DET).

A power detector for each channel band consists of a low-pass filter (LPF), etc., and 5 is an A/D converter.

6□〜61. is channel power data, 71, 7□ is a multiplier, 8 is a correction table, 11 is a voice section detector,
12 is a feature extraction section, 13 is a voice input cover pattern section, 14 is a recognition section, 15 is a speaker-independent dictionary, 16 is a recognition result output section, 17 is a load drive section, 18 a - 18 d are loads;
8a is a radio, 18b is a cassette deck, 18c is an air conditioner, and 18d is a car phone. Voice input from a microphone 1 is amplified by a microphone amplifier 2, and input to an AGC 3 to normalize differences in volume between speakers. AGC may not be necessary depending on the recognition method. The audio signal is n (=15
) is input to power detectors 41 to 4□ consisting of channel bandpass filters (BPF), detectors (DET), low pass filters (LPF), etc., and the power in each channel band is detected. This power voltage is converted into a digital value for each channel by an A/D converter 5. Since the power (digital value) of each channel has the frequency characteristics of the input inside the vehicle, the frequency characteristics are corrected by multiplying this power by the inverse characteristic calculated in advance for the recording characteristics of the studio. Of course, simple multiplication cannot provide accurate correction, but it is effective enough once it becomes a power. The multipliers 7□, 7□ for frequency correction need only be used for high frequencies. Further, the correction value to be multiplied is recorded in the correction table 8, and the correction value is a value experimentally determined from the characteristics of the microphone and the characteristics of the vehicle interior. In the case shown, the audio band is divided into 15 channels, only the two high-frequency channels are corrected, and the power of the 14 channels is reduced to 0.
Multiply by 5, and the power of 15 channels is multiplied by 0.25. A voice section detector 11 detects a voice section from the channel power data corrected in this way, and a feature extraction section 12 detects a voice feature amount, and then the voice input cover pattern 1
Create 3. This pattern is compared with the dictionary pattern of the speaker-independent dictionary 15, and the recognition unit 14 outputs the recognition result from the result output unit 16, with the most similar word as the correct answer. In response to this result, the load control section 17 controls the equipment corresponding to the result. This may be, for example, tuning the radio 18a, fast forwarding the cassette deck 18b, or voice dialing 18d of a car telephone.

However, in the case of the embodiment shown in FIG.

Since the values in the correction table 8 are constant, it is not possible to correct the frequency characteristics depending on the type of vehicle. Changes in frequency characteristics due to changes in car models are mainly due to the shape of the passenger compartment.

FIG. 2 is a block diagram for explaining an embodiment of an in-vehicle voice recognition device suitable for the above-mentioned actual situation, and in the figure, 8a, 8
Reference numerals b and 8c are correction tables, and other parts having the same functions as in the embodiment shown in FIG. 1 are given the same reference numerals as in the case of FIG. Therefore, in this embodiment, correction values for typical car shapes are determined, separate correction tables are provided for each type, and the values are divided into, for example, 1-box cars, 2-box cars, sedans, trucks, etc. Typical correction values for each vehicle type are recorded in each correction table. During use, the user selects the vehicle type and inputs it using switches 98 to 9C. The power is corrected based on the value of the selected correction table.

FIG. 3 is a configuration diagram for explaining another embodiment of the present invention. This embodiment is different from the embodiment shown in FIG. 2 in which the user selects the car type with a switch. In the embodiment, a card 10 in which a correction table 8 is recorded is created, and the value to be corrected is changed by having the user insert a card suitable for the car type, such as a one-box car card or a sedan card. I try to let them do it.

As is clear from the above explanation, the in-vehicle speech recognition device according to the present invention distinguishes between the characteristics of the input voice inside the car and the characteristics of the voice database used to create the speaker-independent dictionary. By approximating, one expected input pattern can be obtained, and the recognition rate of speaker-independent recognition is improved. Furthermore, since it is possible to make more detailed corrections to the frequency characteristics for each vehicle model, it is possible to obtain the expected input cover pattern.
The recognition rate of speaker-independent recognition is improved. Furthermore, by creating a card for each vehicle type and having the user wear it, the user can use the card without having to make a selection by vehicle type, which has the advantage of being more convenient.

[Brief explanation of the drawing]

FIGS. 1 to 3 are block diagrams for explaining embodiments of a vehicle-mounted voice recognition device according to the present invention, and FIGS. 4 and 5 are frequency characteristic diagrams of a microphone. 1...Microphone, 2 Microphone amplifier, 3.AGC14□
~4□,・Power detector in each channel band. 5・A/D converter, 6□~6□,・channel power data, 71,7□・multiplier, 8.8a~8c...correction table, 98~9c...switch, 10...cart , 11... Voice section detection section, 12. Feature extraction section, 1
3... Voice input cover turn section, 14... Recognition section, 15.
・・Speaker-independent dictionary, 16・・Recognition result output unit, 17・
...Load drive unit, 18a...Radio, 18b...Cassette deck, 18c...Air conditioner, 18d...Car phone.

Claims (1)

[Claims] 1. A channel power data section for obtaining power data for each channel of input audio, a correction table, a multiplier, a speech interval detection section, a feature extraction section, a speech pattern memory, a recognition section, a speaker-independent dictionary, a recognition result output section, and a load drive. In an in-vehicle voice recognition device having a section, a constant in the correction table is recorded, the constant is multiplied by the channel power data, and a speaker-independent dictionary is created based on the frequency characteristics of the input voice uttered in the car. An in-vehicle speech recognition device characterized by approximating the recorded characteristics of the speech database used when the vehicle was used, and performing recognition using the power data of each channel obtained as a result.