KR20190042928A - Semiconductor device processing sound signal and microphone including the same - Google Patents

Abstract

A semiconductor device processing a sound signal according to the present technique comprises: an analog-to-digital converter converting an input signal into a digital signal; a signal processing unit processing the digital signal under the influence of an operation parameter inputted from the outside; and an interface providing an output of the analog-to-digital converter or an output of the signal processing unit to the outside.

Description

TECHNICAL FIELD [0001] The present invention relates to a semiconductor device for processing a sound signal, and a microphone device including the semiconductor device. [0002]

The present invention relates to a semiconductor device for processing a sound signal and a microphone device including the same.

Fig. 1 shows an example of a conventional microphone device.

A conventional microphone device includes a substrate 30, a transducer 10 attached on the substrate, and a semiconductor device 20 and a case 40.

The transducer 10 and the semiconductor device 20 and the semiconductor device 20 and the substrate 30 are electrically connected through the conductors 21 and 22. [

The transducer 10 has a membrane or plate 11 and an internal space 12 is formed.

In the conventional microphone device, the passage (41) is formed in the case (40).

The conventional microphone device vibrates the membrane or plate 11 of the transducer 10 and converts the movement of the membrane or the plate into an electric signal in the passage 41 formed in the case 40.

The electrical signal is processed in the semiconductor device 20 and output to the outside.

Recently, various signal processing techniques such as a technique of recognizing a sound signal have been used.

For example, in a conventional microphone device, only an analog or digital signal processed in the semiconductor device 20 is output to the outside, so an external system is required to perform the recognition function.

Accordingly, there has been a problem that the size, power consumption, cost, and the like of the entire system for performing the signal processing function increase in the related art.

EP 2528297 A1

The present invention provides a semiconductor device for processing a sound signal and a microphone device including the semiconductor device.

A semiconductor device according to an embodiment of the present invention includes an analog-to-digital converter for converting an input signal into a digital signal; A signal processing unit for processing a digital signal under the influence of an externally input operation parameter; And an interface for providing the output of the analog-to-digital converter or the output of the signal processing unit to the outside.

According to an aspect of the present invention, there is provided a microphone device including: a transducer for generating a sound signal corresponding to a flow of air; A semiconductor device which converts a sound signal into a digital signal and processes the digital signal under the influence of an externally input operation parameter; A substrate on which a transducer and a semiconductor device are mounted; And a case mounted on the substrate and defining a space so that the transducer and the semiconductor device are contained therein.

In the present invention, the microphone device performs a signal processing function required externally, such as a recognition function, and provides the result to the outside, thereby simplifying the configuration of the entire system and reducing the burden of area, power consumption, and cost .

The present invention can preliminarily learn operation parameters necessary for signal processing from outside and prepare and use it by changing it as needed, so that the functions of the semiconductor device and the microphone can be easily changed.

1 is a sectional view of a conventional microphone device;
2 is a block diagram showing a semiconductor device according to an embodiment of the present invention;
3 is a block diagram showing a semiconductor device according to another embodiment of the present invention;
FIG. 4 is a detailed block diagram of the signal processing unit of FIG. 2;
FIG. 5 is a detailed block diagram of the feature extraction unit of FIG. 3;
FIG. 6 is an explanatory diagram of a neural network that implements the recognition unit of FIG. 4;
Figs. 7 to 9 are detailed block diagrams showing another embodiment of the signal processing unit of Fig. 2; Fig.

Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings.

2 is a block diagram showing a semiconductor device according to an embodiment of the present invention.

The semiconductor device 100 according to an embodiment of the present invention outputs digital data according to a signal output from the transducer 10.

In the present embodiment, the transducer 10 may be made of MEMS technology and converts the sound signal into an analog electric signal and outputs it.

The transducer 10 and the semiconductor device 100 may be included in a microphone device of the type shown in FIG.

However, the microphone device is not limited to the structure shown in FIG. 1, and it is sufficient that the microphone device includes a semiconductor device 100 that processes signals of the transducer 10 and the transducer 10 and outputs them.

A semiconductor device 100 according to an embodiment of the present invention includes an analog-to-digital converter 110. The analog-

In this embodiment, the analog-to-digital converter 110 includes a sigma delta modulator 111 and a decimator 112, but is not limited thereto.

The semiconductor device 100 may further include an amplifier 120 for amplifying the input signal and providing it to the sigma delta modulator 110. [

In another embodiment, the amplifier 120 may exist in a separate configuration outside the semiconductor device 100. [

The semiconductor device 100 according to an embodiment of the present invention includes a signal processing unit 200 for digital signal processing of the output of the analog-to-digital converter 110.

In the present invention, the signal processing unit 200 performs signal processing on an input signal using operation parameters.

The type of signal processing may vary depending on the embodiment.

For example, it can perform various functions such as a sound recognition function, a function to distinguish a kind of sound, a function to reduce a noise signal and a voice signal.

The operation parameters are not generated by the own learning of the signal processing unit 200 but received from the outside.

For example, when the signal processing unit 200 includes a neural network, which is one implementation method of machine learning, information to be generated through learning, such as a synapse weight and a bias of a neural network, .

In the present invention, the signal processing unit 200 is provided with operation parameters obtained by learning in advance from outside without proceeding with learning.

In this way, the configuration of the signal processing unit 200 can be further simplified, thereby reducing the area and power consumption of the circuit.

When the operation of the signal processing unit 200 is changed, the operation parameters of the signal processing unit 200 can be easily changed by receiving previously learned operation parameters as new inputs.

In another embodiment, a signal processing unit using a machine learning technique such as GMM (Gaussian Mixture Model) or SVM (Support Vector Machine) may be implemented in addition to the neural network.

The semiconductor device 100 according to an embodiment of the present invention may further include an interface 130 for outputting the output of the analog-to-digital converter 110 or the output of the signal processing unit 200 to the outside.

The interface 130 may output the output of the signal processing unit 200 and the output of the analog-digital converter 110 to the outside.

At this time, the interface 130 can mix and output the signals output from the analog-digital converter 110 and the signal processing unit 200 in various ways.

Mixing of signals according to a prescribed protocol can be easily changed by a person skilled in the art, so a detailed description thereof will be omitted.

3 is a block diagram showing a semiconductor device 100-1 according to another embodiment of the present invention.

The embodiment of FIG. 3 differs from the embodiment of FIG. 2 in that operation parameters are input to the signal processor 200-1 through the interface 130-1.

When it is desired to control operation parameters outside the semiconductor device 100, it may be desirable to adopt a structure capable of changing the operation parameters by a protocol determined through an interface as shown in FIG.

4 is a block diagram showing an example of the signal processing unit 200 of FIG.

In FIG. 4, the signal processing unit 200 performs a sound recognition function.

The signal processing unit 200 includes a feature extraction unit 210 for extracting a feature vector from a signal output from the A / D converter 110, a recognition unit 220 for receiving a feature vector as an input vector, And a decoder 230 for generating a final output signal using the output vector.

In this embodiment, the recognition unit 220 receives operation parameters prepared in advance from outside and prepared.

The recognition unit 220 can be implemented using a machine learning technique such as a neural network.

5 is a block diagram showing a feature extraction unit 210 according to an embodiment of the present invention.

The block diagram of FIG. 5 may be implemented using hardware, software, or a combination thereof, and is not limited to any one.

When the recognition unit performs speech recognition using the neural network, the feature vector is extracted from the digital signal and then input to the neural network.

At this time, the feature extraction unit 210 extracts a feature vector from the digital signal provided by the A / D converter 110.

5 is a block implementing a Mel Frequency Cepstral Coefficient (MFCC) algorithm widely used in the field of sound recognition.

The hamming window 211 cuts the input digital signal into a frame of a predetermined size.

At this time, each frame constitutes one sample, and a certain number of these samples are secured, and the next operation proceeds.

The FFT operation unit 212 performs an FFT operation on each frame to obtain a power spectrum for each frame.

The Melt filter bank 213 applies the power spectrum to each filter of the Melt filter bank and then calculates the filter bank energy including the output of each filter.

The log operation unit 214 performs a log operation on the output of the melfilter bank 213. [

The DCT operation unit 215 performs a DCT operation on the output of the log operation unit 214. [

Then, the feature vector is generated using the output of the DCT operation unit 215.

The feature vector can be generated using some of the coefficients obtained as a result of the DCT operation.

The feature extracting unit of FIG. 5 is only an embodiment, and a typical technician can design change by applying various techniques.

6 is an explanatory diagram showing a recognition unit 220 implemented using a neural network.

The neural network includes an input layer 221, a hidden layer 222, and an output layer 223.

The hidden layer 222 exists between the input layer 221 and the output layer 223 and may include a plurality of layers therein.

In this embodiment, it is assumed that there are two layers: a first hidden layer 222-1 and a second hidden layer 222-2.

Each layer contains one or more neurons, neurons in adjacent layers are connected via a synapse, and each synapse has a weight value.

The number of neurons in the input layer 221 is matched with the number of elements in the input vector, and a value corresponding to each neuron is input.

In this embodiment, the input vector is the same as the feature vector output from the feature extraction unit 210.

The values of each neuron in the first hidden layer 222-1 are multiplied and added using the neuron values of the input layer 221 and the weight values of the synapses connected thereto (Wij, i, j are indexes) .

At this time, each neuron may have a bias (Bij, i, j is an index) value. The bias value can be multiplied by a predetermined value through the above multiplication and addition to determine the final value of the neuron.

These operations are sequentially performed to determine the neuron values of the output layer 223, and an output vector having the neuron values as elements is output.

The value of a particular element in the output vector can be large and the remainder small.

The decoder 230 may identify the size difference and output the recognition result of the neural network as a digital signal.

The weight values of the synapses or the bias values of the neurons are values determined through learning and they can be included in the operation parameters.

In the present invention, since the operation parameters obtained through externally performed learning, that is, the weights of the synapses and the bias values of the neurons, are provided, it is possible to eliminate the complicated hardware / software configuration required for learning.

Also, it is possible to easily modify the recognition function by preparing various learned operation parameter sets in advance and receiving operating parameters suitable for the situation from the outside.

As described above, the present invention has an advantage in that the structure of the system can be simplified, and the expandability of the function can be achieved simultaneously with the area and the power consumption.

FIG. 7 is a detailed block diagram showing another embodiment of the signal processing unit of FIG. 2. FIG.

The embodiment of FIG. 7 further includes an endpoint detection unit 240 and an activation control unit 250 in the embodiment of FIG.

The end point detection unit 240 observes feature vectors provided by the feature extraction unit 210 and detects an end point.

The activation control unit 250 controls activation of the recognition unit 220 and the decoder 230 according to the end point detection result of the end point detection unit 240.

In an embodiment requiring a parameter for the operation of the activation control unit 250, the parameter may be included externally as part of the operation parameter.

For example, the activation control unit 250 may be implemented through a neural network. At this time, the weight and the bias used in the neural network are included as a part of the operation parameters and can be provided from the outside.

For example, if an end point is detected by the end point detection unit 240, the recognition unit 220 and the decoder 230 may be activated to perform a recognition function using the feature vector output before the end point.

It is possible to reduce the power consumption by outputting the signal obtained as a result of performing the recognition function to the interface 130 and deactivating the recognition unit 220 and the decoder 230. [

8 is a detailed block diagram showing another example of the signal processing unit of FIG.

The signal processing unit 200-3 of FIG. 8 classifies the sound signal into a certain reference.

The feature extracting unit 210 may have substantially the same configuration as that described above.

The classification unit 220-3 is operated under the influence of operation parameters previously learned and input from the outside.

In this embodiment, the classifying unit 220-3 outputs a classifying signal indicating the type of the acoustic signal from the feature vector output from the feature extracting unit.

In this embodiment, the classifying unit 220-3 may be referred to as a recognizing unit and may be implemented using a neural network.

The post-processing unit 230-3 post-processes the classification signal output from the classification unit 220-3 and outputs the classification result. In this embodiment, the post-processing unit 220-3 may be referred to as a decoder.

The signal processing unit 200-3 of FIG. 8 may classify the sound event occurring in the vicinity of the microphone device in the classifying unit 220-3 under the influence of the operation parameter, and output the result to the outside.

9 shows an example of a signal processing unit 200-4 for enhancing a voice signal in a sound signal.

In the present embodiment, voice enhancement means an operation of reducing the strength of a noise signal and increasing the strength of a voice signal when a noise signal and a voice signal are input together.

The configuration and operation of the feature extraction unit 210 in this embodiment are as described above.

The masking unit 220-4 masks and extracts the speech signal from the feature vector.

The masking unit 220-4 may be implemented using a non-negative matrix factorization (NMF) technique.

Since the NMF algorithm itself is a well-known technology, a detailed description is omitted.

In the case of applying the NMF algorithm, a parameter obtained by learning is needed. In the present invention, an operation parameter previously learned in advance is used.

The masking unit 220-4 outputs a feature vector distinguished by a speech signal and a noise signal.

The reconstructing unit 230-4 combines the feature vectors of the speech signal and the noise signal so that the strength of the speech signal is increased.

Accordingly, the signal output from the signal processing unit 200-4 corresponds to a signal in which the voice signal is enhanced.

Although the embodiments of the present invention have been disclosed above, the scope of the present invention is not limited by the foregoing disclosure.

The scope of the present invention should be construed in accordance with the scope of the claims and equivalents thereof.

10: Transducer
100: semiconductor device
110: analog-to-digital converter
111: sigma delta modulator
112: decimator
120: Amplifier
130: Interface
200: Signal processor
210: Feature extraction unit
220:
230: decoder
240: end point detection unit
250:

Claims (19)

An analog-to-digital converter for converting an input signal into a digital signal;
A signal processing unit for processing the digital signal under the influence of an externally inputted operation parameter; And
An interface for providing the output of the analog-to-digital converter or the output of the signal processing unit to the outside
. The semiconductor device of claim 1, further comprising an amplifier for amplifying the input signal and providing the amplified input signal to the analog-to-digital converter. The semiconductor device according to claim 1, wherein the interface provides an output of the analog-to-digital converter and an output of the signal processing unit to the outside. The semiconductor device according to claim 1, wherein the operation parameter is input to the signal processing section via the interface. The signal processing apparatus according to claim 1,
A feature extraction unit for generating a feature vector from the digital signal;
A recognition unit for performing a sound recognition function on the feature vector under the influence of the operation parameter to generate an output vector; And
A decoder for decoding the output vector and outputting a recognition result;
. The signal processing apparatus according to claim 1,
A feature extraction unit for generating a feature vector from the digital signal;
A classification unit for receiving a classification signal indicating the type of acoustic signal from the feature vector under the influence of the operation parameter; And
A post-processing unit for post-processing the classification signal and outputting a classification result;
. 6. The apparatus of claim 5, wherein the recognition unit comprises a neural network comprising a plurality of layers each comprising a plurality of neurons, the operating parameters including a weight of a plurality of synapses connecting the plurality of neurons, And a bias, wherein the recognition unit does not learn the weight and the bias itself. The semiconductor device according to claim 5, further comprising: an end point detection unit for detecting an end point of a sound signal from the feature vector; and an activation control unit for controlling activation of the recognition unit according to a detection result of the end point detection unit. The signal processing apparatus according to claim 1,
A feature extraction unit for generating a feature vector from the digital signal;
A masking unit for distinguishing between the noise signal and the speech signal with respect to the feature vector under the influence of the operation parameter; And
And a reconstruction unit for outputting a signal enhanced with a voice signal from the output of the masking unit,
. A transducer for generating a sound signal corresponding to the air flow; And
A semiconductor device for converting the sound signal into a digital signal and processing the digital signal under the influence of an externally inputted operation parameter;
A substrate on which the transducer and the semiconductor device are mounted; And
A case which is mounted on the substrate and forms a space so that the transducer and the semiconductor device are contained therein;
. 11. The semiconductor device according to claim 10,
An analog-to-digital converter for converting the sound signal into the digital signal;
A signal processing unit for processing the digital signal under the influence of the operation parameter input from the outside; And
An interface for providing the output of the analog-to-digital converter or the output of the signal processing unit to the outside
. 12. The microphone device according to claim 11, further comprising an amplifier for amplifying the input signal and providing the amplified signal to the analog-to-digital converter. The microphone device according to claim 11, wherein the interface provides an output of the analog-to-digital converter and an output of the signal processing unit to the outside. The microphone device according to claim 11, wherein the operation parameter is input to the signal processing section via the interface. 12. The apparatus of claim 11, wherein the signal processing unit comprises a neural network comprising a plurality of layers including a plurality of neurons, the operating parameters including a weight of a plurality of synapses connecting the plurality of neurons, And a bias, wherein the signal processing unit does not learn the weight and the bias itself. 12. The apparatus of claim 11, wherein the signal processing unit
A feature extraction unit for generating a feature vector from the digital signal;
A recognition unit for performing a sound recognition function on the feature vector under the influence of the operation parameter to generate an output vector; And
A decoder for decoding the output vector and outputting a recognition result;
. 12. The apparatus of claim 11, wherein the signal processing unit
A feature extraction unit for generating a feature vector from the digital signal;
A classification unit for receiving a classification signal indicating the type of acoustic signal from the feature vector under the influence of the operation parameter; And
A post-processing unit for post-processing the classification signal and outputting a classification result;
. 17. The microphone device according to claim 16, wherein the signal processing unit further comprises an end point detection unit for detecting an end point of the sound signal from the feature vector, and an activation control unit for controlling activation of the recognition unit according to a detection result of the end point detection unit. 12. The apparatus of claim 11, wherein the signal processing unit
A feature extraction unit for generating a feature vector from the digital signal;
A masking unit for distinguishing between the noise signal and the speech signal with respect to the feature vector under the influence of the operation parameter; And
And a reconstruction unit for outputting a signal enhanced with a voice signal from the output of the masking unit,
.

Images