JP7724678B2 - Howling prevention circuit, microphone device and electronic device - Google Patents

Description

The present invention relates to an anti-feedback circuit for preventing feedback, a microphone device equipped with the anti-feedback circuit, and an electronic device equipped with the anti-feedback circuit.

Figure 8 is a block diagram showing an example configuration and problems with a conventional public address system 110A.

As shown in Figure 8, when a microphone and speaker are combined in a public address system (or conference device, communication device, etc.) 110A, and the audio signal of the user speaking into the microphone is amplified and output from the speaker (in the case of a conference device or communication device, howling occurs due to partial leakage in the transmission/reception separation hybrid circuit (two-wire/four-wire converter), or howling occurs when output from the speaker of the receiver of another communication device), a loop circuit is formed between the microphone, audio amplifier, and speaker, and the amplification of the audio signal of the return sound is repeated within the loop circuit, causing howling.

For example, Patent Document 1 discloses a loudspeaker telephone that allows communication using a speaker and microphone without using a handset (transmitter/receiver). This conventional loudspeaker telephone is characterized by the inclusion of an echo cancellation circuit to prevent feedback, which occurs when sound emitted from the speaker reflects off the walls of a room and enters the microphone.

Japanese Patent Application Publication No. 198155

To prevent the above-mentioned feedback, the input audio signal is passed through a limiter that restricts the gain of the incoming audio signal to a certain level or below. However, this limiter causes changes in sound quality or volume, which can cause changes in the sound quality of the original audio that is not feedback.

Furthermore, there is the following method for eliminating feedback, which uses a first audio signal input from a microphone and a second audio signal output to a speaker.
(1) For example, canceling the first audio signal from a microphone by adding an inverted signal to the second audio signal.
(2) Converting the first audio signal into digital data and removing only the first audio signal from the microphone from the second audio signal.

These methods require feedback cancellation control at both the audio input and audio output sections, which complicates the system configuration and makes it difficult to miniaturize the product.

The object of the present invention is to solve the above problems and provide a feedback prevention circuit that can prevent feedback with high accuracy using a relatively simple configuration compared to conventional examples, a microphone device equipped with said feedback prevention circuit, and electronic equipment equipped with said feedback prevention circuit.

A howling prevention circuit according to one aspect of the present invention comprises:
1. A howling prevention circuit for preventing howling that occurs when at least a part of a sound of an audio signal from a microphone that converts input audio into an audio signal is input to the microphone,
a noise cancellation unit that removes noise from the audio signal from the microphone and outputs only the audio signal;
the noise cancellation unit performs noise cancellation processing using a deep learning model unit that is trained using feature parameters of human voice and determines whether an input voice signal contains noise or not;
The noise cancellation unit includes an audio signal processing unit that performs noise cancellation processing based on the judgment of the deep learning model unit to prevent non-audio periods containing noise from passing through the input audio signal, and outputs the audio signal after the noise cancellation processing.

Therefore, the anti-feedback circuit according to the present invention can prevent feedback with a relatively simple configuration compared to conventional examples, and with high accuracy.

1 is a block diagram showing an example of the configuration of a loudspeaker 110 according to a first embodiment. FIG. 2 is a block diagram showing a detailed configuration example of a noise cancellation unit 102 in FIG. 1 . FIG. 3 is a block diagram showing a detailed configuration example of the deep learning model unit 35 in FIG. 2 . FIG. 10 is a block diagram showing an example of the configuration of a loudspeaker system 113 according to a second embodiment. FIG. 10 is a block diagram illustrating an example of the configuration of a conference device 120 according to a third embodiment. FIG. 10 is a block diagram showing an example of the configuration of a wireless communication device 130 according to a fourth embodiment. 2 is a block diagram showing an example of the configuration and operation of the loudspeaker 110 of FIG. 1. FIG. 1 is a block diagram showing an example of the configuration and problems of a conventional loudspeaker 110A.

Embodiments and variations of the present invention will be described below with reference to the drawings. Note that the same or similar components are designated by the same reference numerals.

(Embodiment 1)
FIG. 1 is a block diagram showing an example of the configuration of a loudspeaker 110 according to the first embodiment.

In FIG. 1, the public address system 110 is composed of a microphone 101, a noise cancellation unit 102, an audio signal amplification unit 103, and a speaker 104.

In the public address system 110, audio input to the microphone 101 is converted into an electrical signal and then input to the noise cancellation unit 102. The noise cancellation unit 102 uses the deep learning model unit 35 (Figures 2 and 3), described below, to distinguish between audio periods and silent periods containing noise, and performs noise cancellation processing to prevent the silent periods from passing through, removing noise other than audio, and then outputs the processed audio signal to the audio signal amplification unit 103. The audio signal amplification unit 103 amplifies the input audio signal and outputs it to the speaker 104, which then converts the input audio signal into audio and outputs it.

Figure 2 is a block diagram showing an example of the detailed configuration of the noise cancellation unit 102 in Figure 1.

With reference to Figure 2, the configuration and operation of the noise cancellation unit 102 in Figure 1 using the deep learning model unit 35 will be described below.

Here, the term "phoneme" refers to a unit of sound that distinguishes one word from another in a particular language, the term "vibration rate" refers to the number of shifts between 0 and 1 in digitized vibration data per second, and the term "vibration count (VC)" refers to the sum of the values of digitized vibration data in each frame. Furthermore, the term "vibration pattern" refers to the data distribution of the sum of vibration frequencies calculated for each predetermined number of frames along the time axis. The deep learning model unit 35 performs noise cancellation processing by taking into account the differences in the data distribution of different vibration patterns, i.e., the sum of different vibration counts (VS values). The vibration rate is similar to the vibration count, but the higher the vibration rate, the larger the vibration count.

Both the amplitude and vibration rate of the audio signal can be observed. A feature of the noise cancellation unit 102 is to detect audio events according to the amplitude and vibration rate of the audio signal. Another feature is to distinguish between voice and non-voice/silence by measuring the sum of vibration count values of digitized vibration data for a predefined number of frames. Another feature is to classify the input audio signal data stream into different phonemes based on their vibration patterns. Another feature is to correctly identify the first activation phoneme from the input audio signal data stream so as to trigger downstream processing units, thereby saving computational costs such as power consumption of the computing system including the processing units.

In FIG. 2, the noise cancellation unit 102 performs noise cancellation processing using audio event detection and is composed of an audio pre-processing unit 38, an AD converter 39, and an audio signal processing unit 30. Here, the audio pre-processing unit 38 performs audio signal pre-processing on the analog audio signal, including high-pass filtering, low-pass filtering, amplification, or a combination thereof, and outputs the processed analog audio signal to the AD converter 39. In other words, the audio pre-processing unit 38 passes only the audio signal from the microphone 101 within a predetermined level range of human audio signals and a predetermined bandwidth. Next, the AD converter 39 AD-converts the analog audio signal to a digital audio signal in accordance with 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 unit 30.

In this embodiment, the AD converter 39 uses an allowable voltage Vadm, which is smaller than the reference voltage Vref, in combination with the reference voltage Vref to form a first threshold voltage Vth1 (= Vref + Vadm) and a second threshold voltage Vth2 (= Vref - Vadm). Based on the first and second threshold voltages Vth1 and Vth2, the AD converter 39 does not perform AD conversion on noise above the first threshold voltage Vth1 or below the second threshold voltage Vth2, but instead performs AD conversion on the audio signal in between, thereby eliminating noise and interference from the input analog audio signal. Here, for example, if Vref = 1.0 V and Vadm = 0.01 V, it can be understood that the frequency of vibration data is low in a quiet environment and high in a voice environment. In this embodiment, "frame size" refers to the number of sampling points corresponding to the digitized vibration data within each frame, and "phoneme window Tw" refers to the time required to collect audio features for each phoneme. In a preferred embodiment, the duration Tf of each frame is, for example, 0.1 to 1 millisecond (ms), and the phoneme window Tw is, for example, approximately 0.3 seconds. In a further preferred embodiment, the number of sampling points corresponding to the digitized vibration data within each frame is, for example, in the range of 1 to 16.

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

In FIG. 2, the audio signal processing unit 30 is configured, for example, by 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, a program for the audio signal processing, and data necessary for executing the program;
(3) RAM (Read Access Memory) 33 for storing data being processed when the CPU 31 executes an operating system that executes basic processing and the audio signal processing program;
(4) a non-volatile EEPROM (Electrically Erasable Programmable Memory) 34 for storing setting data, etc., which will be described later, necessary for executing the audio signal processing;
(5) A deep learning model unit 35, which is configured, for example, by a neural network, removes noise from input voice signal data that has been deep learned based on human voice signal data, and extracts and outputs substantially only the voice signal;
(6) an input interface 36 that performs a predetermined signal conversion process on the audio signal data input from the AD converter 39 to convert it into a signal specification value for the subsequent stage and outputs it to the CPU 31;
(7) An output interface 37 that performs a predetermined signal conversion process on the audio signal data from which noise has been removed by the deep learning model unit 35 to convert it into a signal specification value at a subsequent stage and outputs it to the audio signal amplifier unit 103;
The device is configured to include:

Here, EEPROM 34 stores, for example, a series of vibration count values VC, a vibration count sum VS, a vibration count sum VSf, a vibration count sum VSp (described later), and audio feature values of all feature vectors. EEPROM 34 may also be a storage device such as an external memory. Here, the vibration count values VC of x frames are added together to obtain the vibration count sum VS of the current frame at time Tj. The x frames include the current frame. In one embodiment, CPU 31 adds the vibration count value VC of the current frame at time Tj to the vibration count sum VSp of the immediately preceding (x-1) frames to obtain the vibration count sum VS (= VC + VSp) of the x frames at time Tj.

In a modified example, the CPU 31 adds the vibration count value VC of the current frame at time Tj, the sum of the vibration count values VSf of the immediately succeeding y frames, and the sum of the vibration count values VSp of the immediately preceding (x-y-1) frames to obtain the sum of the vibration count values VS (=VC+VSf+VSp) of the 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 (x×Tf) of x frames (phoneme window Tw) is approximately 0.3 seconds. In a more preferred embodiment, the number of sampling points corresponding to the digitized vibration data of x frames is in the range of x to 16x.

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

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

That is, in the training phase, the set of labeled training examples is used to train the deep learning model unit 35 so that the deep learning model unit 35 recognizes a given activated phoneme, where j=0 to 299, based on three speech features (e.g., (VSj, TDj, TGj)) of each frame of the labeled training examples. At the end of the training phase, the trained deep learning model unit 35 provides a trained score corresponding to the activated phoneme, which is then used as a criterion for classifying an input stream of speech signal data at runtime, where VSj, TDj, and TGj are defined as follows:
(1) VSj: the sum of vibration count values of frame j (VS value);
(2) TDj: the time period of non-zero sum of vibration counts (VS values) in frame j; and (3) TGj: the time gap (time gap) between non-zero sum of vibration counts (VS values) in frame j.

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

Figure 3 is a block diagram showing an example of the detailed configuration of the deep learning model unit 35 in Figure 2.

The deep learning model unit 35 is implemented using a neural network, as shown in FIG. 3, for example. Here, the neural network includes an input layer 41, at least one, but preferably multiple, hidden layers 42, and an output layer 43. The input layer 41 includes three input neurons 51, 52, and 53, each corresponding to one of three audio feature values (i.e., VSj, TDj, and TGj) for each frame of the feature vector. The hidden layer 42 includes neurons 61-74, each with a weight coefficient associated with each input neuron 51, 52, and 53, and a bias coefficient for each neuron. By changing the weight coefficients and bias coefficients of the neurons 61-74 in the hidden layer 42 through a training phase cycle, the neural network can be trained to report a prediction for a given type of input. The output layer 43 includes an output neuron 81 that provides a prediction corresponding to a phoneme (specifically, indicating whether the input is a speech period or a noisy non-speech period).

As explained above, in the noise cancellation unit, the deep learning model unit 35 learns using human voice characteristic parameters and determines whether the input audio signal contains noise or not. Then, based on the determination by the deep learning model unit 35, the CPU 31 of the audio signal processing unit 30 performs noise cancellation processing to prevent noise or not from passing through the input audio signal, and outputs the audio signal after the noise cancellation processing. Here, the deep learning model unit 35 is configured by the neural network of Figure 3, which receives human voice characteristic parameters as input and outputs the determination result of whether the input audio signal contains noise or not.

An example of the operation of the loudspeaker 110 configured as described above will be described below with reference to Figures 1 and 7.

Figure 7 is a block diagram showing an example configuration and operation of the loudspeaker 110 in Figure 1.

In Figure 7, the audio signal input from microphone 101 is passed through noise cancellation unit 102, which uses deep learning model unit 35 (Figure 3). The original purpose is to reduce the non-audio signals of ambient noise input to microphone 101 and extract the audio signal of the desired audio. However, it is also possible to similarly reduce the non-audio signals of feedback noise, thereby avoiding repeated amplification of the feedback audio signal. This makes it possible to extract only the audio signal of the desired audio from microphone 101. Even during feedback, there is no change in the sound quality of the audio signal output from speaker 104, including the reduction of ambient noise input to microphone 101. Furthermore, a small-scale system configuration makes it possible to eliminate feedback in small products.

The inventors conducted an experiment in which they prototyped the public address system 110 shown in Figure 1 and generated feedback. As a result of the experiment, they confirmed that the noise cancellation unit 102 of the public address system 110 according to this embodiment can effectively prevent feedback with high accuracy.

As described above, the noise cancellation unit 102 using the deep learning model unit 35 can provide a howling prevention circuit that can prevent howling with high accuracy, with a relatively simple configuration compared to conventional examples. Furthermore, by providing the noise cancellation unit 102 in the public address system 110, it is possible to realize a public address system that can effectively prevent howling with high accuracy.

(Embodiment 2)
FIG. 4 is a block diagram showing an example of the configuration of a loudspeaker system 113 according to the second embodiment.

In Figure 4, the public address system 113 is composed of a microphone device 111 and a public address device 112 connected using an audio signal cable 105.

The microphone device 111 is configured to include a microphone 101 and a noise cancellation unit 102 that is powered by a DC power supply 102B, which is a secondary battery such as a lithium battery. The configuration and operation of the noise cancellation unit 102 are similar to the noise cancellation unit 102 according to the first embodiment. Power supply to the noise cancellation unit 102 is not limited to the DC power supply 102B, but may also be provided by a so-called AC adapter that rectifies and smooths AC voltage, or by DC voltage power supply from the loudspeaker 112 itself.

The loudspeaker 112 is also configured with an audio signal amplifier 103 and a speaker 104, and their operation is the same as in embodiment 1 in Figure 1.

As described above, the noise cancellation unit 102 using the deep learning model unit 35 can prevent feedback with a relatively simple configuration compared to conventional examples, and with high accuracy. Furthermore, by providing the noise cancellation unit 102 in the microphone device 111, a public address system 113 can be realized that can effectively prevent feedback with high accuracy.

In FIG. 4, the microphone device 111 may be referred to as, for example, an "optional microphone" or an "external microphone." Furthermore, the loudspeaker device 112 may be a wireless communication device or a wired communication device.

(Embodiment 3)
FIG. 5 is a block diagram illustrating an example of the configuration of the conferencing device 120 according to the third embodiment.

In FIG. 5, in the conferencing device 120, the voice input to microphone 101-1 is converted into an electrical signal and then input to noise cancellation unit 102-1. Similarly, the voice input to microphone 101-2 is converted into an electrical signal and then input to noise cancellation unit 102-2. Each noise cancellation unit 102-1, 102-2 uses the deep learning model unit 35 (FIGS. 2 and 3) to distinguish between voice periods and silent periods containing noise, performs noise cancellation processing to prevent the silent periods from passing through, and performs processing to remove noise other than voice. The processed voice signal is then output to adder 121. Adder 121 adds the two input voice signals and then outputs the resulting synthesized voice signal to communication interface 123 via transmission/reception separation hybrid circuit (two-wire/four-wire converter) 122.

The communication interface 123 is, for example, a USB (Universal Serial Bus) interface, and is connected to, for example, a personal computer (PC) 125 via a communication cable 124 to send and receive USB interface signals. In this embodiment, the communication interface 123 transmits a synthesized voice signal acquired by the conferencing device 120 to a remote personal computer (not shown) connected to the personal computer 125 via a predetermined network such as the Internet, and also receives the remote voice signal. The received remote voice signal is passed through the hybrid circuit 122 and the voice signal amplifier 103, and the voice of the voice signal is output from the speaker 104.

In a conference system using the conference device 120 configured as above, for example, the following howling paths are conceivable.
(1) Due to partial leakage in hybrid circuit 122, the voice signal of the user of conference device 120 input to microphones 101-1 and 101-2 is passed from adder 121 through hybrid circuit 122 and voice signal amplifier 103, and the voice output from speaker 104 is passed through to microphones 101-1 and 101-2.
(2) The voice signal of the user of the conference device 120 input to the microphones 101-1 and 101-2 is output from the speaker of the other party via the communication interface 123, the personal computer 125, and the other party's personal computer. This voice is picked up by the microphone of the other party and output from the speaker 104 in the reverse direction via the personal computer 125, the communication interface 123, the hybrid circuit 122, and the voice signal amplifier 103, and then returns to the microphones 101-1 and 101-2. Alternatively, the user's voice signal may return due to partial leakage in the other party's hybrid circuit.

However, in this embodiment, the noise cancellation units 102-1 and 102-2 using the deep learning model unit 35 can prevent feedback with a relatively simple configuration compared to conventional examples, and with high accuracy. Furthermore, by providing the noise cancellation units 102-1 and 102-2 in the conference device 120, a conference system can be realized that can effectively prevent feedback with high accuracy.

In the above embodiment, two microphones 101-1 and 101-2 and two noise cancellation units 102-1 and 102-2 are provided, but the present disclosure is not limited to this and multiple microphones 101 and multiple noise cancellation units 102 may be provided.

In the above embodiment, one noise cancellation unit 102-1, 102-2 is provided for each microphone 101-1, 101-2, but the present invention is not limited to this. After adding two audio signals from the two microphones 101-1, 101-2, noise cancellation processing using the deep learning model unit 35 may be performed by a single noise cancellation unit 102.

(Embodiment 4)
FIG. 6 is a block diagram showing an example of the configuration of a wireless communication device 130 according to the fourth embodiment.

In FIG. 6, the wireless communication device 130 is configured to include a microphone 101, a noise cancellation unit 102, an audio signal amplification unit 103A, a modulation transmission unit 131, a transmission antenna 132, a reception antenna 133, a reception demodulation unit 134, an audio signal amplification unit 103, and a speaker 104.

In the wireless communication device 130 of FIG. 6, voice input to the microphone 101 is converted into an electrical signal and then input to the noise cancellation unit 102. The noise cancellation unit 102 uses the deep learning model unit 35 (FIGS. 2 and 3) to distinguish between voice periods and silent periods containing noise, performs noise cancellation processing to prevent the silent periods from passing through, and performs processing to remove noise other than voice. The noise cancellation unit 102 then outputs the signal to the modulation/transmission unit 131 via the audio signal amplification unit 103A. The modulation/transmission unit 131 generates a modulated radio signal by modulating a carrier wave using a predetermined modulation method in accordance with the input voice signal, and transmits the modulated radio signal via the transmission antenna 132. Meanwhile, the reception/demodulation unit 134 receives the modulated radio signal from the other wireless communication device via the reception antenna 133, performs low-noise amplification, frequency conversion, intermediate frequency amplification, etc. on the received modulated radio signal, and then demodulates it into an audio signal using a predetermined demodulation method and outputs the audio signal to the speaker 104 via the audio signal amplification unit 103.

In a wireless communication system using the wireless communication device 130 configured as above, for example, the following howling paths are conceivable.
(1) A voice signal of the user of wireless communication device 130 input to microphone 101 is modulated and wirelessly transmitted by modulation/transmission unit 131, and the sound of the voice signal is output from the speaker of the other wireless communication device. This voice may be picked up by the other microphone, and output from speaker 104 in the opposite direction via reception/demodulation unit 134 and audio signal amplifier 103 of wireless communication device 130, and may then be sent back to microphone 101.

However, in this embodiment, the noise cancellation unit 102 using the deep learning model unit 35 can prevent feedback with a relatively simple configuration compared to conventional examples, and with high accuracy. Furthermore, by providing the noise cancellation unit 102 in the wireless communication device 130, a wireless communication system can be realized that can effectively prevent feedback with high accuracy.

In the above embodiment, the system is equipped with a modulation/transmission unit 131 and a reception/demodulation unit 134, but the present invention is not limited to this. The reception/demodulation unit 134 may be a separate device, and at least the modulation/transmission unit 131 may be included.

In the above embodiment, a wireless communication device 130 has been described, but the present invention is not limited to this and can also be applied to communication devices such as wired communication devices, telephones, and smartphones instead of wireless communication devices 130.

As described above in detail, the anti-feedback circuit of the present invention uses a noise cancellation unit that uses a deep learning model, making it possible to prevent feedback with a relatively simple configuration compared to conventional examples and with high accuracy. Furthermore, by incorporating a noise cancellation unit into a public address system, communication device, conference device, telephone, smartphone, or computer, it is possible to realize an audio processing system that can effectively prevent feedback with high accuracy.

30 Audio signal processing unit 31 CPU
32 ROM
33 RAM
34 EEPROM
35 Deep learning model unit 36 Input interface 37 Output interface 38 Audio signal pre-processing unit 39 AD converter 41 Input layer 42 Intermediate layer 43 Output layer 51 to 81 Neurons 101, 101-1, 101-2 Microphones 102, 102-1, 102-2 Noise cancellation unit 102B DC power supply 103, 103A Audio signal amplifier unit 104 Speaker 105 Audio signal cable 110, 110A Public address device 111 Microphone device 112 Public address device 113 Public address system 120 Conference device 121 Adder 122 Hybrid circuit 123 Communication interface 124 Communication cable 125 Personal computer 130 Wireless communication device 131 Modulation transmission unit 132 Transmission antenna 133 Reception antenna 134 Reception demodulation unit

Claims (6)

1. A howling prevention circuit for preventing howling that occurs when at least a part of a sound of an audio signal from a microphone that converts input audio into an audio signal is input to the microphone,
a noise cancellation unit that removes noise from the audio signal from the microphone and outputs only the audio signal;
the noise cancellation unit uses a deep learning model unit that determines whether an input audio signal is a non-audio period including noise, and performs noise cancellation processing based on the result of the determination ;
During learning, the deep learning model unit uses, as a learning input, a feature parameter of human voice, which is a data distribution of a sum of vibration frequencies calculated for each predetermined number of frames along a time axis in the voice signal, and learns a result of a determination as to whether or not the voice signal is the non-voice period containing noise, based on a vibration pattern representing the data distribution of the sum of vibration frequencies;
the deep learning model unit, during operation after learning, outputs the result of the determination when a feature parameter of a human voice related to a voice signal converted from an input voice is input;
The noise cancellation unit includes an audio signal processing unit that, when the result of the determination by the deep learning model unit is the silent period, performs noise cancellation processing on an audio signal converted from input audio so as not to pass the silent period containing noise, and outputs the audio signal after the noise cancellation processing.
Anti-feedback circuit. The deep learning model unit is configured by a predetermined neural network.
2. The howling prevention circuit according to claim 1. The noise cancellation unit
an audio signal pre-processing unit provided in a stage preceding the audio signal processing unit, for passing only audio signals from the microphone that are within a predetermined level range of human audio signals and have a predetermined bandwidth;
3. The howling prevention circuit according to claim 1 or 2. The howling prevention circuit according to any one of claims 1 to 3 is provided.
Microphone device. The howling prevention circuit according to any one of claims 1 to 3 is provided.
electronic equipment. The electronic device of claim 5, wherein the electronic device is a public address system, a communication device, a conference device, a telephone, a smartphone, or a computer.