TWI584139B - Speech recognition system and its information processing method applied to non - real - time signal source - Google Patents

Description

Voiceprint recognition system applied to non-instant signal source and information processing method thereof

The invention relates to a voiceprint recognition system, in particular to a voiceprint recognition system applied to a non-instant signal source and an information processing method thereof.

Due to the popularity of mobile devices and the rapid development of the Internet, people have a mobile device that can access the Internet, so that people can connect to the Internet to watch non-instant signal sources during commuting or rest, among which non-instant signals Source refers to media sources such as music videos, dramas, news or animations stored on various media and audio websites (such as youtube, potato or Youku, etc.), so that people are not bored when commuting or taking a break, so many operators will The source of your own media signal is placed on the above-mentioned media and audio-visual website for people to watch to assess whether their media source is popular.

In the prior art, there is a way of judging whether the user's favorite is loved by the click rate of the media signal source. However, sometimes people only turn off the media signal source and then turn it off, or open the media signal source and find that it is not their favorite. It is closed, otherwise it is after opening the media signal source, but it has not been watched because of the unexpected situation, which causes the click rate of the media signal source to be inaccurate. As a result, the operator evaluates the media signal source through the inaccurate click rate. Popularity may lead to erroneous assessments, so there is room for improvement in technology.

In view of the above problems in the prior art, the present invention provides a voiceprint recognition system applied to a non-instant signal source and an information processing method thereof, wherein a user transmits a set of current audio of a non-instant signal source viewed through a mobile device. The feature information is compared to a cloud server to generate a comparison result information, and the supplier refers to the non-instant signal source most frequently viewed by the user, thereby achieving better voiceprint recognition.

The technical means for achieving the above purpose is to apply the aforementioned voiceprint recognition system to a non-instant signal source, comprising: a remote cloud server, which is pre-built with a complex array of preset audio feature information; The device is connected to the cloud server via a network, and is configured to install an application; wherein the mobile device processes a non-instant signal source to obtain a current audio message, and executes an audio feature calculation program for the current The audio information is processed to generate a set of current audio feature information, and the mobile device transmits the current audio feature information to the cloud server via the network, and compares with the set of preset audio feature information to generate a Group comparison results information for reference.

According to the above configuration, the user installs the application on the mobile device to view the non-instant signal source, and the user processes the non-instant signal source through the application to obtain the current audio when viewing the non-instant source. Information, and executing the audio feature calculation program to process the current audio information to generate the current audio feature information, and transmitting the information to the cloud server for comparison with the set of preset audio feature information to generate the group comparison The result information, the supplier refers to the non-instant signal source that the user most often watches, thereby achieving the purpose of providing better voiceprint recognition.

Another technical means for achieving the above object is to apply the foregoing information processing method to a non-instant signal source, comprising a user end mode and a remote mode, the method comprising the following steps: the user end mode Obtaining a current audio information of a non-instant signal source; performing an audio feature calculation program to process the current audio information to generate a set of current audio feature information; transmitting the current audio feature information of the group to a remote end, and The remote preset audio features are compared; the remote mode: obtaining a preset audio information of the plurality of non-instant signal sources; performing the audio feature calculation program to process the preset audio information to respectively generate a group Presetting audio feature information; receiving the current audio feature information of the group, and comparing with the group of preset audio features to generate a set of comparison result information for reference.

According to the foregoing method, after the user terminal mode is executed, the current audio feature information of the group is transmitted to the remote end, and the remote mode is executed to receive the current audio feature information and the group preset. The audio feature information is compared to generate the comparison result information, and the supplier refers to the non-instant signal source most frequently viewed by the user, thereby achieving better sound pattern recognition.

For a preferred embodiment of the voiceprint recognition system of the present invention applied to a non-instant signal source, please refer to FIG. 1 , which includes a mobile device 10 and a remote cloud server 20 . The mobile device 10 transmits through the network. The mobile device 10 is connected to the cloud server 20; in this embodiment, the mobile device 10 can be a smart phone or a tablet computer; the cloud server 20 can be a computer.

The mobile device 10 has a microprocessor 11 , a first communication unit 12 , a touch display 13 and a first memory unit 14 . The microprocessor 11 and the first communication unit 12 and the touch display 13 respectively The first memory unit 14 is configured to execute an application provided by the operator. The first communication unit 12 is connected to the mobile device 10, and the touch display 13 is operated by a user. The application is connected to one or more non-instant signal sources stored on the audio and video media website by the network viewer. The first memory unit 14 stores related information and programs of the application.

When the user views the non-instant signal source through the mobile device 10, the touch display 13 operates the application to process the non-instant signal source viewed by the user according to a specific sampling frequency and audio parameters to obtain the non-instant signal. a current audio information of the source, the microprocessor 11 performs an audio feature calculation program to process the current audio information to generate a set of current audio feature information, and the microprocessor 11 passes the current audio feature information through the first The communication unit 12 is transmitted to the cloud server 20 through the network. In this embodiment, the current audio information may be a piece of audio or complete audio of the non-instant source.

When the microprocessor 11 executes the audio feature calculation program, the current audio information is first subjected to audio downsampling processing, and corresponding to the plurality of sound box information, wherein the adjacent sound box information has overlapping information.

The audio frame information is processed to generate an audio frequency information. In this embodiment, the audio frame information is frequency-converted by a Fourier transform to generate corresponding audio frequency information.

Filtering the audio frequency information to generate a plurality of corresponding filtered values. In this embodiment, the audio frequency information is filtered by a mel filter of different frequency bands to make each audio frequency The information corresponding to the different frequency bands of the Meyer filter produces corresponding filter values.

The filter values of the audio frequency information are subjected to a matrix processing, so that the filter values of the audio frequency information are sequentially arranged into a corresponding matrix, so that each audio frequency information corresponds to a matrix.

The microprocessor 11 convolves a matrix of each audio frequency information through a mask to respectively generate a corresponding convolution value, and the microprocessor 11 sequentially performs a binarization judgment on each convolution value. To note a convolution value greater than zero as "1", a convolution value less than or equal to zero is annotated as "0", and the decodated convolution values are sequentially arranged to produce the current set of audio feature information for the group.

The cloud server 20 has a processor 21, a second communication unit 22, and a second memory unit 23. The processor 21 is connected to the second communication unit 22 and the second memory unit 23, respectively. The unit 22 is connected to the network server 20 and connected to the first communication unit 12 of the mobile device 10. The second memory unit 23 pre-assembles the complex array of preset audio feature information.

The second memory unit 23 of the cloud server 20 pre-establishes the group of preset audio feature information, which is processed in advance according to a non-instant signal source provided by the operator, to respectively generate corresponding preset audio feature information, and is stored in the corresponding preset audio feature information. In the second memory unit 23, in the embodiment, the non-instant signal source provided by the operator corresponds to a set of preset audio feature information.

The processor 21 performs specific audio specification processing according to the non-instant signal source provided by the operator to obtain a corresponding preset audio information, and the processor 21 executes the audio feature calculation program to process the preset audio information. Corresponding preset audio feature information is generated respectively. In this embodiment, the preset audio information is a complete audio of the corresponding non-instant signal source; the audio feature calculation program executed by the processor 21 of the cloud server 20, The same is true for the audio feature calculation program executed by the microprocessor 11 of the mobile device 10.

The processor 21 of the cloud server 20 executes the audio feature calculation program to respectively perform audio down sampling processing on the preset audio information, and respectively generate corresponding complex sound box information, wherein the adjacent sound box information overlaps The information is frequency-converted by the Fourier transform to generate corresponding audio frequency information, and the audio frequency information of the preset audio information is filtered by the Meyer filter of different frequency bands. And respectively generating a corresponding plurality of filter values, and performing a matrix processing on the filter values of the preset audio information, so that the filter values of the preset audio information are sequentially arranged into corresponding matrices, and the processor 21 The matrix of each preset audio information is convoluted by the mask to generate a corresponding convolution value, and the processor 21 sequentially performs a binarization judgment on the convolution value of each preset audio information. To note a convolution value greater than zero as "1", a convolution value less than or equal to zero is annotated as "0", and the decodated convolution values are sequentially arranged to generate such Group preset audio feature information.

The processor 21 receives the current audio feature information transmitted by the mobile device 10, and compares the received current audio feature information with the set of preset audio feature information one by one to generate a complex error rate, wherein the lower the error rate is, the lower the error rate is. The closer the current audio feature information is to the preset audio feature information, the higher the error rate is, the more the current audio feature information is different from the preset audio feature information, and the processor 21 generates a set of comparison result information according to the lowest error rate, the supplier. Refer to the non-instant source of the most frequently viewed by the user.

In this embodiment, the second communication unit 22 of the cloud server 20 transmits the group comparison result information to the first communication unit 12 of the mobile device 10 via the network, and the microprocessor 11 of the mobile device 10 The comparison result information is transmitted to the touch display 13 for the user to refer to whether the non-instant signal source viewed by the user is the same as the non-instant signal source that is compared with the cloud server 20, for the user to refer to the favorite non-instant source. The click of the signal source supports the result.

According to the above content, an information processing method applied to a non-instant signal source is further summarized, which includes a user end mode and a remote mode, and the mobile device 10 executes the user end mode, and the cloud server 20 Executing the remote mode, wherein the user mode includes the following steps: obtaining a current audio information of a non-instant signal source (S11); in this embodiment, the mobile device 10 is to perform the specific sampling by the non-instant signal source. The frequency and audio parameters are processed to obtain the current audio information, and the current audio information may be a segment audio or a complete audio of the non-instant signal source; the current audio information is processed by an audio feature calculation program to generate a set of current The audio feature information (S12); transmitting the current audio feature information of the group to the cloud server 20 (S13).

The remote mode includes the following steps: obtaining a preset audio information of a plurality of non-instant signal sources (S21); in this embodiment, the non-instant signal sources are provided in advance by the operator, and each preset audio information is provided. Corresponding non-instant signal source complete audio; performing the audio feature calculation program to process the preset audio information to generate a set of preset audio feature information respectively (S22); receiving the group transmitted by the mobile device 10 The current audio feature information is compared with the group of preset audio feature information to generate a set of comparison result information for reference (S23); in this embodiment, the group comparison result information is received The current audio feature information is compared with the preset audio feature information to generate a complex error rate. The lower the error rate, the closer the current audio feature information is to the preset audio feature information, and the higher the error rate is, the current audio feature is represented. The more different the information is from the preset audio feature information, the cloud server 20 generates the set of comparison result information according to the lowest error rate.

In this embodiment, the audio feature calculation program executed by the mobile device 10 and the cloud server 20 includes the following steps: processing the audio information and obtaining the complex sound box information (S31); in this embodiment, the audio information is The information is subjected to downsampling processing to obtain the sound box information, and the adjacent sound box information has overlapping information; the mobile device 10 executes the audio feature calculation program and is the same as the audio feature calculation program executed by the cloud server 20. The mobile device 10 executes the audio feature calculation process to process the current audio information, and the cloud server 20 performs the audio feature calculation process to process the preset audio information, and the preset audio information is processed. Each of the sound box information is processed to generate corresponding audio frequency information (S32); in this embodiment, the frequency information is converted by using a Fourier transform; Filtering each audio frequency information to generate corresponding multiple filter values (S33); in this embodiment, the system transmits different The first Meyer filter is used to filter the audio frequency information, so that each of the audio frequency information corresponds to a different frequency band of the Meyer filter to generate a corresponding filter value; and the filtered values of the audio frequency information are sequentially arranged into corresponding ones. a matrix (S34); in this embodiment, the filtering values of the audio frequency information are matrixed, so that each audio frequency information corresponds to a matrix; and the matrix of each audio frequency information is volumed through a mask. a product operation to respectively generate a convolution value (S35); determine whether the generated convolution value is greater than zero (S36); if so, the convolution value is denoted as "1" (S37); in this embodiment, Each convolution value is binarized; if not, the convolution value is denoted as "0" (S38); all the convolved convolution values are sequentially arranged to generate corresponding audio feature information (S39); In an embodiment, the mobile device 10 generates corresponding current audio feature information, and the cloud server 20 generates a corresponding complex array of preset audio feature information.

As can be seen from the above, the current audio information of the non-instant signal source that is viewed by the mobile device 10 is processed by the audio feature calculation program to generate corresponding current audio feature information, and transmitted to the cloud server 20 and the preset audio feature. The information is compared to generate corresponding comparison result information, and the supplier refers to the non-instant source of the most frequently viewed by the user, or provides the user with reference to the click support result of the favorite non-instant signal source, thereby providing more information. The purpose of good sound pattern recognition.

10‧‧‧Mobile devices

11‧‧‧Microprocessor

12‧‧‧First communication unit

13‧‧‧Touch display

14‧‧‧First memory unit

20‧‧‧Cloud Server

21‧‧‧ Processor

22‧‧‧Second communication unit

23‧‧‧Second memory unit

1 is a block diagram of a system architecture in accordance with a preferred embodiment of the present invention. 2 is a flow chart of a user end mode in accordance with a preferred embodiment of the present invention. 3 is a flow chart of a remote mode in accordance with a preferred embodiment of the present invention. 4 is a flow chart of an audio feature calculation program in accordance with a preferred embodiment of the present invention.

10‧‧‧Mobile devices

11‧‧‧Microprocessor

12‧‧‧First communication unit

13‧‧‧Touch display

14‧‧‧First memory unit

20‧‧‧Cloud Server

21‧‧‧ Processor

22‧‧‧Second communication unit

23‧‧‧Second memory unit

Claims (7)

A voiceprint recognition system applied to a non-instant signal source, comprising: a remote cloud server pre-built with a complex array of preset audio feature information; and a mobile device connected to the cloud server via a network And installing an application; wherein the mobile device processes a non-instant signal source to obtain a current audio message, and performs an audio feature calculation program to process the current audio information to generate a current set of audio feature information. The mobile device transmits the current audio feature information to the cloud server via the network, and compares with the set of preset audio feature information to generate a set of comparison result information for reference; wherein the audio device The feature calculation program processes the corresponding audio information to generate complex sound box information, processes each sound box information to generate corresponding audio frequency information, and then filters each audio frequency information to generate each. Corresponding multiple filtering values, the filter values of each audio frequency information are sequentially arranged into a corresponding matrix, And convolving each matrix through a mask to respectively generate a convolution value, and determining that the convolution value greater than zero is denoted as 1, the convolution value less than or equal to zero is denoted as 0, and according to the convolution after the annotation The value produces corresponding audio feature information. The voiceprint recognition system of claim 1, wherein the mobile device has a microprocessor, a first communication unit, a touch display, and a first memory unit, the microprocessor The first communication unit is connected to the first communication unit, the touch display and the first memory unit, and the first communication unit is configured to connect to the network, and the touch display is configured to operate the application. The first memory unit stores the audio feature calculation program. The voiceprint recognition system of claim 2, wherein the cloud server has a processor, a second communication unit, and a second memory unit, and the processor is respectively associated with the second communication The second communication unit is connected to the network and is connected to the first communication unit of the mobile device. The second memory unit stores the set of preset audio feature information. The voiceprint recognition system of claim 3, wherein the processor of the cloud server processes a plurality of non-instant signal sources to obtain a corresponding preset audio information, the processor Performing the audio feature calculation program to process the preset audio information to generate the preset audio feature information. The voiceprint recognition system applied to the non-instant signal source according to claim 4, wherein each of the sound frames is frequency-converted by a Fourier transform to generate the audio frequency information; and a Meyer filter is transmitted through different frequency bands. The device filters each audio frequency information to generate the above-mentioned filtered values. An information processing method for a non-instant signal source, wherein a mobile device performs a user terminal mode, the user terminal mode comprising the steps of: obtaining a current audio information of a non-instant signal source; and executing an audio feature calculation program Processing the current audio information to generate a set of current audio feature information; wherein the audio feature calculation program comprises the steps of: processing the current audio information and obtaining the plurality of audio frame information; processing each of the audio frame information to Corresponding audio frequency information is generated separately; each audio frequency information is filtered to generate corresponding multiple filter values; the filtered values of the audio frequency information are sequentially arranged into a corresponding matrix; Convoluting a matrix of each audio frequency information to generate a convolution value; determining whether the generated convolution value is greater than zero; if so, the convolution value is noted as 1; if not, convolving the convolution value 0; sort all the convolved values of the annotations to generate the current audio feature information of the group; Transmitting the current audio feature information of the group to a remote end and comparing with the remote preset audio features. An information processing method for a non-instant signal source, wherein a remote mode is executed by a cloud server, the remote mode comprising the steps of: obtaining a preset audio information of a plurality of non-instant signal sources; and executing the audio feature calculation program Processing the preset audio information to generate a set of preset audio feature information respectively; wherein the audio feature calculation program comprises the steps of: processing the preset audio information and separately obtaining the plurality of audio frame information; Processing each of the sound box information to generate corresponding audio frequency information separately; filtering each audio frequency information to generate corresponding multiple filter values; and sequentially arranging the filtered values of the audio frequency information into Corresponding a matrix; convolving a matrix of each audio frequency information through a mask to respectively generate a convolution value; determining whether the generated convolution value is greater than zero; if so, the convolution value is denoted as 1; If not, the convolution value is annotated as 0; all the convolved convolution values are sequentially arranged to generate a complex array of preset audio feature information; Audio feature set of current information, and wherein the audio preset for comparison with such groups, to produce a set of comparison result information by reference.