JP2023001754A - Wireless communication device and wireless communication system - Google Patents

Abstract

A wireless communication device that performs wireless communication can perform noise cancellation so as to effectively output human uttered voices as compared with the prior art.
A radio communication apparatus according to the present invention includes a modulation transmission section that modulates a radio carrier wave according to an input audio signal and transmits a radio signal. The wireless communication device includes a first noise canceller that performs audio signal processing so as to cancel noise from the input audio signal and outputs the signal to the modulation transmitter. Noise cancellation processing is performed using a deep learning model unit that is learned using the feature parameters of the speech and determines whether or not it is a non-speech period containing noise from the demodulated speech signal.
[Selection drawing] Fig. 1

Description

The present invention relates to a radio communication device for, for example, a specified low-power radio communication radio station and a radio communication system including a plurality of radio communication devices.

For example, FIGS. 1 and 4 of Patent Document 1 disclose use of a noise canceling circuit that performs noise canceling processing on an audio signal from a microphone of a wireless communication device.

Japanese Patent Application Laid-Open No. 2004-165962 (Figs. 1 and 4)

The noise canceling circuit disclosed in Patent Document 1 only discloses that "noise is removed from the audio signal input from the microphone", but in the conventional technology, the noise signal outside the microphone is inverted and input to the microphone. A so-called phase-reversal type noise cancellation circuit that is added to an audio signal is widely used. In the phase inversion type noise canceling circuit, when the noise is loud even when a person is speaking, the level of the audio signal is suppressed so that the human speech can be output effectively. There was a problem that noise cancellation could not be performed.

SUMMARY OF THE INVENTION An object of the present invention is to solve the above problems, and to provide a wireless communication device for performing wireless communication that can perform noise cancellation so as to effectively output human utterance voices compared to the conventional technology. An object of the present invention is to provide a wireless communication system including a device and the plurality of wireless communication devices.

A wireless communication device according to an aspect of the present invention includes:
In a wireless communication device comprising a modulation transmitter that modulates a wireless carrier wave according to an input audio signal and transmits a wireless signal,
A first noise cancellation unit that performs audio signal processing so as to cancel noise from the input audio signal and outputs it to the modulation transmission unit;
The first noise cancellation unit learns using a feature parameter of human speech, and determines whether it is a non-speech period containing noise from the demodulated speech signal Using a deep learning model unit, Perform noise cancellation processing.

Therefore, according to the wireless communication device according to the present invention, the noise cancellation circuit using the deep learning model unit is provided instead of the phase inversion type noise cancellation circuit, so that the wireless communication device that performs wireless communication is improved. , noise cancellation can be performed to effectively output human speech.

1 is a block diagram showing a configuration example of a wireless device 1 according to an embodiment; FIG. 2 is a block diagram showing a configuration example of a noise cancellation unit 13 of FIG. 1; FIG. 3 is a block diagram showing a configuration example of a deep learning model unit 35 of FIG. 2; FIG.

Hereinafter, embodiments and modifications according to the present invention will be described with reference to the drawings. In addition, the same code|symbol is attached|subjected about the same or similar component.

FIG. 1 is a block diagram showing a configuration example of a wireless device 1 according to an embodiment. In FIG. 1, a wireless device 1 is an example of a wireless communication device, and includes a receiving antenna 11, a receiving demodulation unit 12, a noise canceling unit 13, an audio signal amplifier 14, a speaker 15, a control unit 20, a PTT It comprises an operation unit 21 including a (Push To Talk) key 21A, a microphone 22, an audio signal amplifier 23, a noise cancellation unit 24, a modulation transmission unit 25, and a transmission antenna 26.

Here, the wireless device 1 according to the embodiment is an example of a wireless communication device of a specified low-power wireless station for a specified low-power wireless communication system, for example. In this embodiment, in the transmitter of the wireless device 1, in place of the phase-reversal type noise cancellation circuit, for example, a noise cancellation circuit that is particularly effective in noise reduction in FM (frequency modulation) demodulation or PM (phase modulation) demodulation A part 24 is provided. In addition, in the present embodiment, the wireless device 1 further includes a noise canceller 13, which is particularly effective in reducing noise, in its receiver. Also, a plurality of wireless devices 1 constitute a wireless communication system.

In FIG. 1, a radio signal received by a receiving antenna 11 is input to a receiving demodulator 12 . The reception demodulation unit 12 performs low-noise amplification, low-frequency conversion, intermediate frequency amplification, etc. on the received radio signal, and then performs a predetermined demodulation method such as FM (frequency modulation) demodulation or PM (phase modulation) demodulation. demodulates it into an audio signal and outputs it to the noise canceling unit 13 . The noise canceling unit 13 uses a deep learning model unit 35 (FIG. 3) that has been deep-learned using human speech to allow the demodulated audio signal to pass through only during the audio signal period, thereby canceling noise. After the audio signal is processed, the processed audio signal is output to the speaker 15 via the audio signal amplifier 14 .

The microphone 22 converts an input voice into a voice signal and outputs the voice signal to a modulation transmission section 25 via a voice signal amplifier 23 and a noise cancellation section 24 having the same configuration as the noise cancellation section 13 . When the PTT key 21A is turned on, the control unit 20 operates the modulation transmission unit 25. The modulation transmission unit 25 modulates the radio carrier wave according to a predetermined modulation method according to the input audio signal, and then modulates the modulated radio wave. A radio signal, which is a carrier wave, is transmitted from the transmitting antenna 26 after high frequency conversion and power amplification.

In this embodiment, the wireless device 1 operates in a simultaneous call system in which the transmission frequency and the reception frequency are different, but the present invention is not limited to this, and the wireless device 1 uses the same transmission frequency and reception frequency. frequency is used, the controller 20 stops the operation of the reception demodulator 12 when the PTT key 21A is turned on. That is, in the wireless device 1 that does not simultaneously perform the transmission operation and the reception operation, during the reception operation, the noise cancellation unit 24 may be operated instead of the noise cancellation unit 13 to perform noise cancellation processing during reception.

Next, with reference to FIG. 2, the configuration and operation of the noise cancellers 13 and 24 of FIG. 1 using the deep learning model section 35 will be described below.

FIG. 2 is a block diagram showing a configuration example of the noise cancellers 13 and 24 in FIG.

Here, the term "phoneme" refers to the unit of sound that distinguishes one word from another in a particular language, and the term "vibration rate" refers to the zero point of the digitized vibration data at each second. and 1, and the term "vibration count (VC)" refers to the sum of the values of the digitized vibration data within each frame. Further, the "vibration pattern" means the data distribution of the sum of the vibration frequencies calculated for each predetermined number of frames along the time axis. In the deep learning model unit 35, noise cancellation processing is performed in consideration of different vibration patterns, that is, differences in the data distribution of the sum of different vibration count values (VS values), and the vibration rate is similar to the vibration count value. However, the higher the vibration rate, the higher the vibration count.

Both the amplitude and vibration rate of the speech signal are observable. A feature of the noise cancellers 13 and 24 is to detect audio events according to the amplitude and vibration rate of the audio signal. Another feature is to distinguish between speech and non-speech/silence by measuring the sum of the vibration count values of the digitized vibration data for a predefined number of frames. Another feature is to classify the input audio signal data stream into different phonemes according to their vibration patterns. Another feature is the correct discrimination of the first activation phoneme from the incoming audio signal data stream to trigger downstream processing units, thereby calculating power consumption, etc. of the computing system containing the processing units. It is to save costs.

In FIG. 2, the noise cancellation units 13 and 24 perform noise cancellation processing using audio event detection, and include an audio signal preprocessing unit 38, an AD converter 39, and an audio signal processing unit 30. consists of Here, the audio signal preprocessing unit 38 performs audio signal preprocessing, including high-pass filtering, low-pass filtering, amplification, or a combination thereof, on the analog audio signal, and outputs the processed analog audio signal. Output to AD converter 39 . That is, the audio signal pre-processing unit 38 allows the audio signal from the microphone 22 to pass only within a predetermined level range of human audio signals and in a predetermined bandwidth. Next, the AD converter 39 AD-converts the analog audio signal into a digital audio signal according to a predetermined reference voltage Vref and an allowable voltage Vadm (<Vref), and outputs the digital audio signal to the input interface 36 of the audio signal processing section 30 .

In the present embodiment, in the AD converter 39, the allowable voltage Vadm smaller than the reference voltage Vref is combined with the reference voltage Vref to obtain the first threshold voltage Vth1 (=Vref+Vadm)) and the second threshold voltage Vth2 (=Vref−Vadm), and the AD converter 39 converts the first threshold voltage Vth1 based on the first threshold voltage Vth1 and the second threshold voltage Vth2. Noise and interference in the input analog audio signal are removed by not performing AD conversion on noise above or below the second threshold voltage Vth2, but by performing AD conversion on the audio signal between them. can do. Here, if Vref=1.0 V and Vadm=0.01 V, for example, it can be understood that the frequency of vibration data is low in a quiet environment, and the frequency of vibration data is high in a voice environment. In this embodiment, the “frame size” means the number of sampling points corresponding to the digitized vibration data in each frame, and the “phoneme window Tw” means the speech feature amount of each phoneme. Means time to collect. In a preferred embodiment, the duration Tf of each frame is eg 0.1-1 milliseconds (ms) and the phoneme window Tw is eg about 0.3 seconds. In a further preferred embodiment, the number of sampling points corresponding to the digitized vibration data within each frame ranges, for example, from 1-16.

When analyzing speech signals, the method of short-term analysis is usually adopted since most speech signals are stable in a short period of time. For example, assuming that the sampling frequency fs used by the AD converter 39 is 16000 and the duration Tf of each frame is 1 ms, the frame size is fs×1/1000=16 sample points.

In FIG. 2, the audio signal processing unit 30 is configured by, for example, a computer device,
(1) A CPU (Central Processing Unit) 31 that executes predetermined audio signal processing such as noise cancellation;
(2) a ROM (Read Only Memory) 32 that stores an operating system that executes the basic processing of the CPU 31, the audio signal processing program, and data necessary for executing the program;
(3) a RAM (Read Access Memory) 33 for storing data being processed when the operating system for executing the basic processing of the CPU 31 and the program for the audio signal processing are executed;
(4) a non-volatile EEPROM (Electrically Erasable Programmable Memory) 34 for storing later-described setting data necessary for executing the audio signal processing;
(5) For example, it is composed of a neural network, etc., and removes noise from voice signal data that is input after deep learning based on human voice signal data, and extracts and outputs substantially only the voice signal. a deep learning model unit 35;
(6) an input interface 36 that performs a predetermined signal conversion process for converting audio signal data input from the AD converter 39 into a signal specification value in the subsequent stage and outputs the result to the CPU 31;
(7) The audio signal data from which noise has been removed by the deep learning model unit 35 is subjected to a predetermined signal conversion process for converting it into a signal specification value in the subsequent stage, and the wireless device 1 is transmitted through the terminal T12, the audio line L2, etc. an output interface 37 that outputs to
configured with

Here, the EEPROM 34 stores, for example, a series of vibration count values VC, a sum of vibration count values VS, a sum of vibration count values VSf, a sum of vibration count values VSp (to be described later), and sound feature values of all feature vectors. do. Note that the EEPROM 34 may be a storage device such as an external memory. The audio event detection method applied to the audio signal processor 30 is executed during runtime by the CPU 31 to capture audio events. Audio event detection is performed assuming fs=16000, Tf=1 ms, and Tw=0.3 s.

Specifically, the CPU 31 calculates the sum of the vibration data values of the current frame (that is, within 1 ms) to be processed, acquires the vibration count value VC, and then obtains the VC of the current frame at time Tj. Store the value in EEPROM 34 . Here, the vibration count values VC of x frames are added to obtain the sum of the vibration count values VS of the current frame at time Tj. x frames includes the current frame. In one embodiment, the CPU 31 adds the vibration count value VC of the current frame at the time Tj and the sum VSp of the vibration count values of the (x−1) frames immediately before that frame to obtain the x number of vibration count values at the time Tj. Obtain the sum VS (=VC+VSp) of the vibration count values of the frame.

In the modified example, the CPU 31 outputs the vibration count value VC of the current frame at the time Tj, the total sum VSf of the vibration count values of the y frames immediately after that, and the (xy-1) frames immediately before that. to obtain the sum of vibration counts VS (=VC+VSf+VSp) of x frames at time Tj, where y is greater than or equal to zero. The CPU 31 stores the values of VS, VSf and VSp in the EEPROM 34 . In a preferred embodiment, the duration of x frames (phoneme window Tw) (x×Tf) is approximately 0.3 seconds. In a further preferred embodiment, the number of sampling points corresponding to x frames of digitized vibration data is in the range of x to 16x.

In general, with respect to speech signal data, the vibration patterns of the vibration count values VC are similar for the same phoneme, but the vibration patterns of the VS values are completely different for different phonemes. Therefore, the vibration pattern of the vibration count value VC can be used to distinguish between phonemes. In particular, for example, the barking of a chicken or a cat and human speech are completely different in terms of the frequency distribution of the vibration coefficient VC, and it is known that most of the vibration coefficient VC of human speech is distributed below 40. .

In the learning phase, the CPU 31 of the speech signal processing unit 30 first executes a predetermined speech signal data collection method a plurality of times to collect a plurality of feature vectors corresponding to a plurality of phonemes, and labels corresponding to the plurality of feature vectors. Append to form multiple labeled training examples. A plurality of labeled training examples for different phonemes, including the starting phoneme, are then applied to training the deep learning model unit 35 . Finally, a trained deep learning model unit 35 (constituting a predictive model for the speech signal data) is created to classify whether the incoming speech signal data stream contains the activation phoneme. When a predetermined phoneme is specified as a starting phoneme for the audio signal processing unit 30, the deep learning model unit 35 is trained with a plurality of labeled training examples for different phonemes including at least the specified phoneme.

That is, in the training phase, the deep learning model portion 35 is trained using a set of labeled training examples, whereby the deep learning model portion 35 learns the three audio features of each frame of the labeled training examples. Based on the quantity (eg, (VSj, TDj, TGj)), we try to recognize a given activation phoneme between j=0-299. At the end of the learning phase, the trained deep learning model unit 35 provides a learned score corresponding to the activation phoneme, which is then converted at runtime to the incoming audio signal data stream. Used as criteria for classification. VSj, TDj, and TGj are defined as follows.
(1) VSj: sum of vibration count values (VS value) of frame j;
(2) TDj: the time period of the sum of non-zero vibration counts (VS values) at frame j; and (3) TGj: the time between the sums of non-zero vibration counts (VS values) at frame j. Gap (time gap).

Various machine learning techniques related to supervised learning can be used to train the deep learning model portion 35, such as support vector machine (SVM) methods, random forest methods, convolutional neural network methods, etc. can. In supervised learning, a function calculator (i.e., deep learning model portion 35) is created using a plurality of labeled training examples, each of which consists of an input feature vector and a labeled output. When trained, deep learning model portion 35 can be applied to new unlabeled examples to generate corresponding scores or predictions.

FIG. 3 is a block diagram showing a detailed configuration example of the deep learning model unit 35 of FIG.

The deep learning model unit 35 is implemented using a neural network, as shown in FIG. 3, for example. Here, the neural network comprises one input layer 41 , at least one and preferably several intermediate layers 42 and one output layer 43 . The input layer 41 has three input neurons 51, 52, 53, each input neuron 51, 52, 53 corresponding to three audio feature values (i.e., VSj, TDj, TGj) of each frame of the feature vector. . The intermediate layer 42 is also composed of neurons 61-74 having weight coefficients associated with each input neuron 51, 52, 53 and bias coefficients for each neuron. By changing the weighting and biasing factors of each neuron 61-74 of the hidden layer 42 through cycles of the learning phase, the neural network can be trained to report a predicted value for a given type of input. . In addition, the output layer 43 includes one output neuron 81 that provides one prediction value corresponding to a phoneme (specifically indicating whether it is a speech period or a non-speech period containing noise).

As described above, in the noise cancellation units 13 and 24, the deep learning model unit 35 learns using the feature parameters of human speech, and determines whether or not the input speech signal is a non-speech period containing noise. judge. Then, the CPU 31 of the audio signal processing unit 30 performs noise cancellation processing so as not to pass the non-speech period containing noise from the input audio signal based on the determination of the deep learning model unit 35, and performs the noise cancellation. Outputs the processed audio signal. Here, the deep learning model unit 35 receives as input the characteristic parameters of human speech, and outputs the determination result of determining whether or not the input speech signal is a non-speech period containing noise. It consists of a neural network.

As described above, in the present embodiment, the radio device 1 in its transmission section, instead of the phase inversion type noise cancellation circuit, for example, in FM (Frequency Modulation) demodulation or PM (Phase Modulation) demodulation, especially noise reduction A noise canceling section 24 is provided, which is effective. As a result, in a wireless communication device that performs wireless communication, noise cancellation can be performed so as to effectively output a human uttered voice, compared to the conventional technology. By providing the noise canceling unit 24 on the transmission side, noise can be effectively removed from the audio signal in circuits and devices (for example, wireless relay devices) on and after the transmission side.

In addition, in the present embodiment, the wireless device 1 further includes a noise canceller 13, which is particularly effective in reducing noise, in its receiver. As a result, in a wireless communication device that performs wireless communication, noise cancellation can be performed so as to effectively output a human uttered voice, compared to the conventional technology. When FM or PM is used as the modulation method, by passing the demodulated audio signal through the noise canceling unit 13 before outputting it to the speaker 15, white noise is eliminated when there is almost no signal, and the call limit is reached. Even if the distance approaches, the received voice continues to keep clear voice. In addition, instead of the conventional noise squelch circuit, by canceling noise using the noise canceling unit 13 that uses the deep learning model unit 35, the output of noise in the no-signal state is stopped (or reduced). The output of the audio signal that can be originally received is not stopped. Therefore, it is possible to output the voice signal received up to the very limit of the communication distance, and it is possible to extend the communication distance compared to the noise squelch circuit.

(Modification)
Although the wireless device 1 includes the noise canceller 24 in the above embodiment, the present invention is not limited to this, and the noise canceller 24 may not be provided.

As described in detail above, according to the wireless communication device of the present invention, the deep learning model unit 35 is used on the transmission side of the wireless device 1, or further on the reception side, instead of the phase inversion type noise cancellation circuit. By performing noise cancellation using the noise canceling unit 13, a wireless communication device that performs wireless communication can perform noise cancellation so as to effectively output human uttered voices as compared to the conventional technology.

1 Wireless device 11 Receiving antenna 12 Receiving demodulator 13 Noise canceling unit 14 Audio signal amplifier 15 Speaker 20 Control unit 21 Operation unit 21A PTT key 22 Microphone 23 Audio signal amplifier 24 Noise canceling unit 25 Modulation transmitting unit 26 Transmitting antenna 30 Audio signal processing Part 31 CPU
32 ROMs
33 RAM
34 EEPROMs
35 deep learning model unit 36 input interface 37 output interface 38 audio signal preprocessing unit 39 AD converter 41 input layer 42 intermediate layer 43 output layers 51 to 81 neurons

Claims (13)

In a wireless communication device comprising a modulation transmitter that modulates a wireless carrier wave according to an input audio signal and transmits a wireless signal,
A first noise cancellation unit that performs audio signal processing so as to cancel noise from the input audio signal and outputs it to the modulation transmission unit;
The first noise cancellation unit learns using a feature parameter of human speech, and determines whether it is a non-speech period containing noise from the demodulated speech signal Using a deep learning model unit, A wireless communication device that performs noise cancellation processing. 2. The radio communication apparatus according to claim 1, wherein said modulating transmission unit modulates a radio carrier wave according to a frequency modulation method or a phase modulation method according to an input audio signal. The first noise cancellation unit performs noise cancellation processing so as not to pass a non-speech period containing noise from the input speech signal based on the determination of the deep learning model unit, After the noise cancellation processing Equipped with an audio signal processing unit that outputs an audio signal of
The radio communication device according to claim 1 or 2. The first noise cancellation unit
Further comprising an audio signal pre-processing unit provided in the preceding stage of the audio signal processing unit for passing only an input audio signal within a predetermined level range of a human audio signal and a predetermined bandwidth. 4. The wireless communication device of claim 3. The deep learning model unit is composed of a predetermined neural network that receives characteristic parameters of human speech as an input and outputs a determination result that determines whether or not the input speech signal is a non-speech period containing noise. The radio communication device according to any one of claims 1 to 4, wherein the radio communication device is The wireless communication device further includes a reception demodulator that demodulates a received wireless signal into an audio signal,
a second noise cancellation unit that performs audio signal processing so as to cancel noise from the demodulated audio signal and outputs it to the modulation transmission unit;
The second noise cancellation unit learns using a feature parameter of human speech, and determines whether it is a non-speech period containing noise from the demodulated speech signal Using a deep learning model unit, 3. The wireless communication device according to claim 1, which performs noise cancellation processing. 7. The radio communication apparatus according to claim 6, wherein said reception demodulation unit demodulates a received radio signal using a frequency modulation method or a phase modulation method. The second noise cancellation unit performs noise cancellation processing so as not to pass a non-speech period containing noise from the input speech signal based on the determination of the deep learning model unit, After the noise cancellation processing Equipped with an audio signal processing unit that outputs an audio signal of
The radio communication device according to claim 6 or 7. The second noise cancellation unit
Further comprising an audio signal pre-processing unit provided in the preceding stage of the audio signal processing unit for passing only an input audio signal within a predetermined level range of a human audio signal and a predetermined bandwidth. 9. A wireless communication device according to claim 8. The deep learning model unit is composed of a predetermined neural network that receives characteristic parameters of human speech as an input and outputs a determination result that determines whether or not the input speech signal is a non-speech period containing noise. A radio communication device according to any one of claims 6 to 9, wherein A wireless communication device that does not simultaneously perform a transmission operation and a reception operation, and performs noise cancellation processing during reception by operating the first noise cancellation unit instead of the second noise cancellation unit during the reception operation. The wireless communication device according to any one of claims 6 to 10. The wireless communication device according to any one of claims 1 to 11, wherein said wireless communication device is a specified low power radio station which is a wireless communication device for a specified low power wireless communication system. A wireless communication system comprising a plurality of wireless communication devices according to any one of claims 1-12.

Images