Speech recognition intelligent self-generation ticket system and method based on section decentration multi-wave matching

Abstract

The invention relates to a voice recognition intelligent self-ticket-forming system and method based on stripe decentration multi-wave matching, and belongs to the field of power grid dispatching. The system comprises a radio module, a forward filtering module, a neural network audio preprocessing module, a backward decentration module, a multi-wave matching module, a machine room, a dispatching operation platform, an on-duty telephone, a signal optical cable and a signal cable row; the name of a caller is recognized through a telephone number, voiceprint characteristics and report content through segment decentralized multi-wave matching, then local dialects are recognized according to dialect pronunciation characteristics, matching comparison is conducted on the incoming call content to generate an operation request, and a machine can recognize the dialects, generate characters and present the characters to answers. And the risk of misjudgment of the answering person due to the communication state is avoided.

Description

Technical Field

[0001] The present invention relates to a voice recognition intelligent self-ticketing system and method based on segment-based decentralized multi-wave matching, belonging to the field of power grid dispatching. Background Art

[0002] Traditional power grid dispatching completes work applications and approvals through staff making and receiving calls, and combines network diagram boards to detect power grid power flow changes. However, with the continuous expansion of the power grid scale and the increasing complexity of the power grid structure, the daily work content and work mode of the power system are extremely complex. The workload of the power grid dispatching department is constantly increasing. While staff make and receive a large number of calls, it is difficult to quickly process complex work information. During the work stage with too many maintenance tasks and frequent faults, there is a high risk of delayed judgment and misjudgment. Due to the work characteristics of the power grid dispatching department involving a large power supply range, delayed judgment and misjudgment of work applications will, at the least, lead to an extended power outage time, a decline in work efficiency, and damage to power grid equipment. At the worst, it will lead to personal safety accidents and power grid collapse. Therefore, there is an urgent need for an operation assistance device for dispatching staff that can intelligently identify operation requests and applicants, facilitating dispatching staff to record work, perform power flow calculations among a large number of incoming calls, which is of great significance to the safe and stable operation of large power grids. Summary of the Invention

[0003] According to the problems described in the background, the problem to be solved by the present invention is: to provide a voice recognition intelligent self-ticketing system and method based on segment-based decentralized multi-wave matching, which uses segment-based decentralized multi-wave matching to identify the caller's name from the phone number, voiceprint features, and report content, then identifies local dialects according to the pronunciation characteristics of dialects, and matches and compares the incoming call content to generate an operation request, so as to solve the problems existing above.

[0004] To achieve the above object, the present invention provides the following technical solutions: provide a voice recognition intelligent self-ticketing system based on segment-based decentralized multi-wave matching, including a sound collection module, a forward filtering module, a neural network audio preprocessing module, a backward wave division module, a multi-wave matching module, a computer room, a dispatching operation platform, a duty phone, a signal optical cable, and a signal cable row;

[0005] The sound collection module is electrically connected to the forward filtering module;

[0006] The forward filtering module is signal-connected to the neural network audio preprocessing module through a signal optical cable;

[0007] The neural network audio preprocessing module is signal-connected to the backward wave division module through a signal optical cable;

[0008] The backward wave division module is signal-connected to the multi-wave matching module through a signal cable row;

[0009] The output end of the multi-wave matching module is connected to the computer room;

[0010] The computer room data is connected to the scheduling operation platform.

[0011] Preferably, the duty phone input terminal is connected to a radio module and a forward filtering module.

[0012] The present invention also includes a voice recognition intelligent self-ticketing method based on segment weighted multi-wave matching, which includes the following steps:

[0013] (1) Use telephone records as local dialect samples to train the neural network and embed the trained neural network into the neural network audio preprocessing module;

[0014] (2) The audio receiving module receives the incoming call signal, activates the working state, and inputs the audio signal into the forward filtering module;

[0015] (3) The forward filtering module feeds back the filtered audio signal to the on-call telephone and transmits the audio data to the neural network audio preprocessing module;

[0016] (4) The neural network audio preprocessing module performs dialect recognition on the audio and performs segment weighted correction on the audio band;

[0017] (5) transmitting the corrected audio band to the backward demultiplexing module;

[0018] (6) The backward wave splitting module splits the voice signal into harmonic signals and transmits them to the multi-wave matching module in sections through the signal cable row;

[0019] (7) The multi-wave matching module identifies the operation request by scaling the standard operation request to match each voice band, and generates an operation request to be transmitted to the computer room;

[0020] (8) The computer room stores the generated operation request in the database, and the staff can perform further operations on the dispatch operation platform.

[0021] Preferably, in step (4), the preprocessing calculation method is as follows:

[0022]

[0023] In the above formula, A is the input signal matrix, T is the time vector, t is the time series, ρ is the recognition resolution, O is the sequence vector, o is the sequence number, s is the hidden layer state vector, h is the eigenvalue vector, N is the output segment weight vector, and the vector dot multiplication is input into the audio vector I for correction:

[0024] J=N*I

[0025] In the above formula, J is the corrected audio vector.

[0026] In step (7), the specific matching process is as follows:

[0027]

[0028] In the above formula, P is the matching evaluation factor, a is the amplification coefficient, b and c are the offset coefficients, and y is the input audio vector.

[0029] The stretching matching is performed on the audio using an optimization algorithm. The specific algorithm steps are as follows:

[0030] (7.1) Construct an optimization matrix X = [a, b, c];

[0031] (7.2) Substitute the matching formula in step (7) as the objective function into the algorithm for solution

[0032] (7.3) Determine whether the solved matching evaluation factor P value meets the error requirement, generally 0.8. If it meets, execute (7.4); if not, return to (7.2);

[0033] (7.4) Determine the credibility of the evaluation factor. The interval (0, 0.35] is strong credibility, (0.35, 0.55] is general credibility, and (0.55, 0.8] is weak credibility.

[0034] The confidence factor M is obtained by weighting the matching factors of each band:

[0035]

[0036] In the above formula, Q is the weighting matrix, d is the sub-wave sequence, and D is the number of sub-waves.

[0037] The beneficial effects of the present invention are:

[0038] 1. For telephone applications with local dialects, it can effectively identify dialect content, generate operation instructions, and reduce the risk of misjudgment by staff;

[0039] 2. It can quickly identify operation requests in multiple channels, prevent chaos in operations caused by staff answering and making multiple calls, and avoid misjudgment by staff through reference to operation instructions.

[0040] 3. It determines the identity of the caller through the telephone number, voiceprint recognition, and speech recognition, prevents misissuance and misjudgment of instructions, facilitates effective traceability of the work process, and prevents risky operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 is a schematic diagram of the system of the present invention;

[0042] Figure 2 is a schematic diagram of the method of the present invention; DETAILED DESCRIPTION OF THE INVENTION

[0043] The embodiments of the present invention are further described below in conjunction with the accompanying drawings:

[0044] Example 1

[0045] like Figure 1 As shown, the present invention provides a voice recognition intelligent self-ticketing system based on segment weighted multi-wave matching, including a sound receiving module, a forward filtering module, a neural network audio preprocessing module, a backward wave splitting module, a multi-wave matching module, a machine room, a dispatching operation platform, a duty telephone, a signal optical cable, and a signal cable row;

[0046] The sound receiving module is electrically connected to the forward filtering module; the electrical connection uses a low-impedance shielded cable to ensure that the original voice signal is transmitted to the forward filtering module in an analog form without distortion.

[0047] The forward filtering module is connected to the neural network audio preprocessing module through a signal optical cable; the forward filtering module uses an adaptive notch filter to eliminate power frequency interference and transmits the digital signal to the neural network module through the signal optical cable.

[0048] The neural network audio preprocessing module is connected to the backward wave splitting module signal through a signal optical cable;

[0049] The backward wave splitting module is connected to the multi-wave matching module signal through a signal cable row;

[0050] The output end of the multi-wave matching module is connected to the machine room;

[0051] The computer room data is connected to the scheduling operation platform.

[0052] The duty telephone input terminal is connected to the radio module and the forward filtering module.

[0053] By connecting this series of devices, the sound signal can be effectively sent to the recognition module with the greatest possible fidelity, while filtering out the noise and white noise that affect recognition.

[0054] Example 2

[0055] like Figure 2 As shown, the present invention also includes a voice recognition intelligent self-ticketing method based on segment weighted multi-wave matching, comprising the following steps

[0056] (1) Use telephone records as local dialect samples to train the neural network and embed the trained neural network into the neural network audio preprocessing module;

[0057] (2) The audio receiving module receives the incoming call signal, activates the working state, and inputs the audio signal into the forward filtering module;

[0058] (3) The forward filtering module feeds the filtered audio signal back to the duty phone and simultaneously transmits the audio data to the neural network audio preprocessing module;

[0059] (4) The neural network audio preprocessing module performs dialect recognition on the audio and corrects the audio band through segment weight correction;

[0060] (5) The corrected audio band is transmitted to the backward wave splitting module;

[0061] (6) The backward wave splitting module splits the speech signal into harmonic signals and transmits them to the multi-wave matching module in segments through the signal cable row;

[0062] (7) The multi-wave matching module identifies the operation request by stretching and matching the standard operation request with each band of the speech, and generates an operation request to be transmitted to the computer room;

[0063] (8) The computer room stores the generated operation request in the database, and the staff can perform further operations on the dispatching operation platform.

[0064] Since the dialects in each place are different, it is necessary to train the device with the local telephone recording data. Through the above steps, the trained device can recognize the content spoken by the caller and push it to the operator in text form for reference.

[0065] In step (4) above, the calculation method of preprocessing is as follows:

[0066]

[0067] In the above formula, A is the input signal matrix, T is the time vector, t is the time series, ρ is the recognition resolution, O is the sequence vector, o is the serial number, s is the hidden layer state vector, h is the eigenvalue vector, N is the output segment weight vector, and the vector dot product is used to correct the input audio vector I:

[0068] J = N * I

[0069] In the above formula, J is the corrected audio vector.

[0070] In step (7) above, the specific matching process is as follows:

[0071]

[0072] In the above formula, P is the matching evaluation factor, a is the amplification coefficient, b and c are the offset coefficients, and y is the input audio vector.

[0073] This process improves a type of differential evolution algorithm for finding the minimum value, and obtains a mathematical model suitable for acoustic wave recognition through data fitting. The magnitude of the output value of this mathematical model represents the similarity between this pronunciation and the correct pronunciation. The algorithm adjusts the coefficient to determine whether this pronunciation is a pitch and frequency variation of the correct pronunciation, so as to achieve the purpose of judging dialect content.

[0074] The stretching matching is performed on the audio using an optimization algorithm. The specific algorithm steps are as follows:

[0075] (7.1) Construct an optimization matrix X = [a, b, c];

[0076] (7.2) Substitute the matching formula in step (7) as the objective function into the algorithm for solution

[0077] (7.3) Judge whether the solved matching evaluation factor P value meets the error requirement, generally 0.8. If it meets, execute (7.4); if not, return to (7.2);

[0078] (7.4) Judge the credibility of the evaluation factor. The interval (0, 0.35] is strong credibility, (0.35, 0.55] is general credibility, and (0.55, 0.8] is weak credibility.

[0079] The evaluation factor is a fitted value obtained from a large number of data experiments and simulation training.

[0080] The confidence factor M is obtained by weighting the matching factors of each waveband:

[0081]

[0082] In the above formula, Q is the weighting matrix, d is the sub-wave sequence, and D is the number of sub-waves.

[0083] In order to flexibly incorporate various factors into consideration and enhance the universality through weighting, due to the different dialect natures and equipment characteristics in different regions, the sound quality noise and the degree of acoustic wave matching cannot match the empirical values mentioned in this article. Therefore, setting weights can adjust the weights when configuring personnel in different regions to achieve the maximum recognition accuracy.

[0084] During the communication process, there are situations such as non-standard language use by people, mishearing due to surrounding noise, misremembering due to the inattentiveness of the answering personnel, and misunderstanding due to unclear dialect expression. Machine recognition can avoid the above problems. The machine can recognize the dialect and generate text to present to the answering personnel, eliminating the risk of misjudgment of the answering personnel due to the communication state.

Claims

1. A voice recognition intelligent self-ticketing system based on segment weighted multi-wave matching, characterized in that: It includes radio module, forward filtering module, neural network audio preprocessing module, backward wave splitting module, multi-wave matching module, computer room, dispatching operation platform, duty telephone, signal optical cable, and signal cable row; The sound receiving module is electrically connected to the forward filtering module; The forward filtering module is connected to the neural network audio preprocessing module signal via a signal optical cable; The neural network audio preprocessing module is connected to the backward wave splitting module signal through a signal optical cable; The backward wave splitting module is connected to the multi-wave matching module signal through a signal cable row; The output end of the multi-wave matching module is connected to the machine room; The computer room data is connected to the scheduling operation platform.

2. According to claim 1, a voice recognition intelligent self-ticketing system based on segment weighted multi-wave matching is characterized in that: The duty telephone input terminal is connected to the radio module and the forward filtering module.

3. A method for intelligent self-generating tickets for speech recognition based on segment weighted multi-wave matching, characterized in that: The following steps are included (1) Use telephone records as local dialect samples to train the neural network and embed the trained neural network into the neural network audio preprocessing module; (2) The audio receiving module receives the incoming call signal, activates the working state, and inputs the audio signal into the forward filtering module; (3) The forward filtering module feeds back the filtered audio signal to the on-call telephone and transmits the audio data to the neural network audio preprocessing module; (4) The neural network audio preprocessing module performs dialect recognition on the audio and performs segment weighted correction on the audio band; (5) transmitting the corrected audio band to the backward demultiplexing module; (6) The backward wave splitting module splits the voice signal into harmonic signals and transmits them to the multi-wave matching module in sections through the signal cable row; (7) The multi-wave matching module identifies the operation request by scaling the standard operation request to match each voice band, and generates an operation request to be transmitted to the computer room; (8) The computer room stores the generated operation request in the database, and the staff can perform further operations on the dispatch operation platform.

4. According to claim 1, a method for intelligent self-generating tickets based on voice recognition based on segment weighted multi-wave matching, characterized in that: In step (4), the preprocessing calculation method is as follows: In the above formula, A is the input signal matrix, T is the time vector, t is the time series, ρ is the recognition resolution, O is the sequence vector, o is the sequence number, s is the hidden layer state vector, h is the eigenvalue vector, N is the output segment weight vector, and the vector dot multiplication is input into the audio vector I for correction: J=N*I In the above formula, J is the corrected audio vector.

5. According to claim 1, a method for intelligent self-generating tickets based on voice recognition based on segment weighted multi-wave matching, characterized in that: In step (7), the specific matching process is as follows: In the above formula, P is the matching evaluation factor, a is the multiplication coefficient, b and c are the offset coefficients, and y is the input audio vector.

6. According to claim 5, a method for intelligent self-ticketing based on voice recognition and weighted multi-wave matching is characterized in that: Use the optimization algorithm to scale and match the audio. The specific algorithm steps are as follows: (7.1) Construct the optimization matrix X = [a, b, c]; (7.2) Take the matching formula in step (7) as the objective function and bring it into the algorithm to solve (7.3) Determine whether the solution matching evaluation factor P value meets the error requirement, which is generally 0.

8. If it does, execute (7.4). If not, return to (7.2); (7.4) Determine the credibility of the evaluation factor. The interval (0, 0.35] is strong credibility, (0.35, 0.55] is general credibility, and (0.55, 0.8] is weak credibility.

7. According to claim 6, a method for intelligent self-ticketing based on voice recognition and weighted multi-wave matching is characterized in that: The confidence factor M is obtained by weighting the matching factors of each band: In the above formula, Q is the weighting matrix, d is the wavelet sequence, and D is the wavelet number.

Images