Intelligent transcription method, system and program product for financial large model

Through wavelet transformation and Mel frequency cepspectral technology, the financial speech signal is processed, combined with speech recognition and keyword extraction, and the training corpus of financial big models is automatically constructed, solving the problem of low efficiency of traditional manual corpus collection and realizing the financial big model efficient automated training.

CN119091903AActive Publication Date: 2024-12-06HEFEI ENTROPY TECHNOLOGY CO LTD

Patent Information

Application Number
CN202411570054.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-11-06
Publication Date
2024-12-06
Estimated Expiration
2044-11-06

AI Technical Summary

Technical Problem

The acquisition method of traditional financial big model training corpus relies on manual collection and labeling, which consumes manpower and is inefficient and needs improvement.

Method used

Wavelet transformation processing and threshold filtering technology are used to reduce the noise of financial speech signals, combined with the Mel frequency cepspectral feature extraction and speech recognition model, financial speech is automatically transcribed into training corpus, and a financial training corpus is constructed through keyword extraction and syntactic analysis.

Benefits of technology

It realizes intelligent transcription from pronunciation to financial big model training corpus, significantly improves the processing efficiency of training corpus, replaces traditional manual collection methods, and supports automated corpus training of financial big model.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119091903A_ABST
    Figure CN119091903A_ABST
Patent Text Reader

Abstract

The invention belongs to the technical field of machine learning, and particularly discloses an intelligent transcription method, system and program product for a large financial model, and the method comprises the steps: collecting a financial voice signal for making a training corpus of the large financial model, and carrying out the corresponding wavelet decomposition, improved wavelet threshold filtering and wavelet reconstruction processing, thereby obtaining a noise reduction signal; performing corresponding feature extraction and voice recognition on the noise reduction signal to obtain a corresponding transcriptional text, performing financial keyword extraction and syntactic analysis processing on the transcriptional text, constructing a corresponding financial training corpus, and compiling the corpus into a corpus for performing corpus training on the financial large model; and intelligent transcription from voice to financial large model training corpora can be realized. According to the method, the financial voice can be quickly, reliably and automatically transcribed into the corresponding financial large model training corpus, the processing efficiency of the financial large model training corpus is remarkably improved, and automatic corpus training of the financial large model is facilitated.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of machine learning, and in particular relates to an intelligent transcription method, system and program product for a large financial model. Background Art

[0002] The financial big model is a financial service platform based on big data, cloud computing and artificial intelligence. It combines traditional financial services with emerging technologies, and responds quickly to customer needs by deeply mining massive data. This model mainly relies on big data technology and makes fine adjustments on the basic model to meet the specific needs of the financial industry. The development of the financial big model requires a large amount of training corpus. The traditional way to obtain the training corpus for the financial big model is to manually collect a large amount of financial texts and perform word and sentence decomposition and annotation. This is very labor-intensive and inefficient, and needs to be improved urgently. Summary of the invention

[0003] The purpose of the present invention is to provide an intelligent transcription method, system and program product for financial large models to solve the above-mentioned problems existing in the prior art.

[0004] In order to achieve the above object, the present invention adopts the following technical solutions: In a first aspect, an intelligent transcription method for a large financial model is provided, comprising: Collect financial speech for making financial model training corpus, and convert financial speech into corresponding financial speech signals; Perform wavelet transform processing on the financial speech signal with a set number of decomposition layers to obtain the wavelet transform coefficients corresponding to each layer; Perform threshold filtering on each wavelet transform coefficient based on the set wavelet threshold to obtain the improvement coefficient corresponding to each wavelet transform coefficient; Perform wavelet reconstruction based on the improved coefficients corresponding to each wavelet transform coefficient to obtain the financial speech signal after noise reduction; Extract features from the de-noised financial speech signal to obtain corresponding speech signal features, and input the speech signal features into a preset speech recognition model for speech recognition processing to obtain corresponding transcription text; Extract keywords from the transcribed text to obtain several financial keywords, and perform syntactic analysis on the transcribed text to obtain context dependencies corresponding to each financial keyword; The corresponding financial training corpus is composed of each financial keyword and its context dependency, and the corresponding financial training corpus is compiled into a corpus, and the corpus is used to perform corpus training on the financial big model.

[0005] In a possible design, the financial speech signal is subjected to wavelet transform processing with a set number of decomposition layers to obtain wavelet transform coefficients corresponding to each layer, including: The DB4 wavelet basis function is used to perform 4-layer wavelet decomposition on the financial speech signal to obtain the wavelet transform coefficients corresponding to each layer.

[0006] In a possible design, the threshold filtering process is performed on each wavelet transform coefficient based on a set wavelet threshold to obtain an improved coefficient corresponding to each wavelet transform coefficient, including: Substitute the set wavelet threshold and the corresponding wavelet transform coefficient into the preset wavelet threshold function for threshold filtering to obtain the improved coefficient corresponding to the wavelet transform coefficient. The wavelet threshold function is:

[0007] Wherein, x represents the wavelet transform coefficient, y represents the improved coefficient corresponding to the wavelet transform coefficient x, sgn represents the step function, λ is the set wavelet threshold, δ is the set first coefficient, and δ∈(0,1), ω is the set second coefficient, and ω>1; the wavelet threshold λ is expressed as

[0008] Among them, j is the decomposition level of wavelet transform, σ is the set noise variance, and N represents the signal length of the financial speech signal.

[0009] In one possible design, the financial speech signal is represented as , where t is the time parameter, is a financial voice signal, For financial voice signals The signal component without noise in For financial voice signals The noise signal component in .

[0010] In a possible design, the feature extraction of the de-noised financial speech signal to obtain the corresponding speech signal features includes: The Mel-frequency cepstrum coefficient algorithm is used to extract the Mel-frequency cepstrum features of the denoised financial speech signal to obtain the Mel-frequency cepstrum features of the financial speech signal.

[0011] In a possible design, the step of inputting the speech signal features into a preset speech recognition model for speech recognition processing to obtain a corresponding transcription text includes: The Mel-frequency cepstrum features of the financial speech signal are input into a preset speech recognition model for speech recognition processing to obtain a corresponding transcription text, wherein the speech recognition model adopts a trained Whisper speech recognition model.

[0012] In a possible design, the transcribed text is subjected to keyword extraction to obtain a number of financial keywords, and the transcribed text is subjected to syntactic analysis to obtain context dependencies corresponding to each financial keyword, including: The keyword extraction method based on Word2Vec word clustering is used to extract keywords from the transcribed text to obtain several financial keywords, and the Arc-standard method is used to perform dependency analysis on the transcribed text to obtain the context dependency corresponding to each financial keyword, wherein the context dependency includes context dependency words and corresponding dependency types.

[0013] In a second aspect, an intelligent transcription system for a large financial model is provided, comprising a speech collection unit, a signal decomposition unit, a threshold filtering unit, a signal reconstruction unit, a speech recognition unit, a text parsing unit and a corpus construction unit, wherein: A speech collection unit, used to collect financial speech for making financial model training corpus, and convert the financial speech into corresponding financial speech signals; A signal decomposition unit, used to perform wavelet transform processing of a set number of decomposition layers on the financial speech signal to obtain wavelet transform coefficients corresponding to each layer; A threshold filtering unit is used to perform threshold filtering on each wavelet transform coefficient based on a set wavelet threshold to obtain an improved coefficient corresponding to each wavelet transform coefficient; A signal reconstruction unit, used for performing wavelet reconstruction processing based on the improved coefficients corresponding to each wavelet transform coefficient to obtain a financial speech signal after noise reduction; A speech recognition unit is used to extract features from the financial speech signal after noise reduction to obtain corresponding speech signal features, and input the speech signal features into a preset speech recognition model for speech recognition processing to obtain corresponding transcription text; A text parsing unit is used to extract keywords from the transcribed text to obtain a number of financial keywords, and to perform syntactic analysis on the transcribed text to obtain context dependencies corresponding to each financial keyword; The corpus construction unit is used to use various financial keywords and their contextual dependencies to form corresponding financial training corpus, and compile the corresponding financial training corpus into a corpus, which is used to perform corpus training on the financial big model.

[0014] In a third aspect, an intelligent transcription system for a large financial model is provided, comprising: A memory for storing instructions; A processor is used to read the instructions stored in the memory and execute any one of the methods described in the first aspect according to the instructions.

[0015] In a fourth aspect, a computer-readable storage medium is provided, wherein instructions are stored on the computer-readable storage medium, and when the instructions are executed on a computer, the computer executes any one of the methods described in the first aspect. In addition, a computer program product is provided, and when the computer program product is executed on a computer, the computer executes any one of the methods described in the first aspect.

[0016] Beneficial effect: The present invention collects financial speech signals for making financial large model training corpus, performs corresponding wavelet decomposition, improved wavelet threshold filtering and wavelet reconstruction processing, obtains a noise reduction signal, then performs corresponding feature extraction and speech recognition on the noise reduction signal to obtain a corresponding transcribed text, then performs financial keyword extraction and syntactic analysis on the transcribed text, constructs corresponding financial training corpus and incorporates it into the corpus for corpus training of financial large model, and can realize intelligent transcription from speech to financial large model training corpus. The present invention can quickly and reliably automatically transcribe financial speech into corresponding financial large model training corpus through corresponding speech signal processing recognition and transcription text analysis, so as to replace the traditional method of manually collecting text and annotating financial large model training corpus, significantly improve the processing efficiency of financial large model training corpus, and help to realize automatic corpus training of financial large model. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 This is a schematic diagram of the steps of the method in Example 1 of the present invention; Figure 2 This is a schematic diagram of the structure of the system in Example 2 of the present invention; Figure 3 Schematic diagram of the system structure in Example 3 of the present invention. DETAILED DESCRIPTION

[0019] It should be noted that the description of these embodiments is used to help understand the present invention, but does not constitute a limitation of the present invention. The specific structures and functional details disclosed herein are only used to describe the exemplary embodiments of the present invention. However, the present invention can be embodied in many alternative forms, and it should not be understood that the present invention is limited to the embodiments set forth herein.

[0020] It should be understood that, unless otherwise clearly specified and limited, the term "connection" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the embodiments can be understood according to specific circumstances.

[0021] In the following description, certain details are provided to facilitate a complete understanding of the example embodiments. However, it will be appreciated by those of ordinary skill in the art that the example embodiments may be implemented without these certain details. For example, devices may be shown in block diagrams to avoid obscuring the examples with unnecessary details. In other embodiments, well-known processes, structures, and techniques may not be shown in unnecessary detail to avoid obscuring the embodiments.

[0022] Embodiment 1: This embodiment provides an intelligent transcription method for a large financial model, which can be applied to a corresponding large financial model training terminal, such as Figure 1 As shown, the method comprises the following steps: S1. Collect financial speech for making financial model training corpus, and convert the financial speech into corresponding financial speech signals.

[0023] In specific implementation, financial speech for making financial model training corpus can be collected in advance, such as by crawling the corresponding financial speech on the relevant network platform or collecting the financial speech from the personnel on site through the speech collector. Then, the financial speech is converted into the corresponding financial speech signal through the corresponding speech conversion processing, so as to perform subsequent speech signal processing.

[0024] S2. Perform wavelet transform processing on the financial speech signal with a set number of decomposition layers to obtain the wavelet transform coefficients corresponding to each layer.

[0025] In specific implementation, the financial speech signal obtained by converting financial speech will contain corresponding noise signals. Therefore, the financial speech signal can be expressed as , where t is the time parameter, is a financial voice signal, For financial voice signals The signal component without noise in For financial voice signals The noise signal component in , N is the final time point of the financial voice signal, which can characterize the signal length of the financial voice signal. In order to improve the subsequent recognition accuracy of the financial voice signal, it is necessary to effectively eliminate the noise part in the financial voice signal. Therefore, the financial voice signal can be suppressed by wavelet transform, but the traditional wavelet transform method is not effective in suppressing noise. In order to improve the denoising effect, this embodiment designs an improved wavelet transform denoising method, which can first use the DB4 wavelet basis function to perform 4-layer wavelet decomposition on the financial voice signal to obtain the wavelet transform coefficients corresponding to each layer.

[0026] S3. Perform threshold filtering on each wavelet transform coefficient based on the set wavelet threshold to obtain an improved coefficient corresponding to each wavelet transform coefficient.

[0027] In specific implementation, after decomposing and obtaining the wavelet transform coefficients corresponding to each layer, the corresponding wavelet transform coefficients and the set wavelet thresholds can be substituted into the preset wavelet threshold function for threshold filtering to obtain the improved coefficients corresponding to the wavelet transform coefficients. The wavelet threshold function is

[0028] Wherein, x represents the wavelet transform coefficient, y represents the improved coefficient corresponding to the wavelet transform coefficient x, sgn represents the step function, λ is the set wavelet threshold, δ is the set first coefficient, and δ∈(0,1), ω is the set second coefficient, and ω>1; the wavelet threshold λ is expressed as

[0029] Among them, j is the decomposition level of wavelet transform, σ is the set noise variance, and N represents the signal length of the financial speech signal.

[0030] S4. Perform wavelet reconstruction processing based on the improved coefficients corresponding to each wavelet transform coefficient to obtain the financial speech signal after noise reduction.

[0031] In specific implementation, after calculating the improvement coefficients corresponding to the wavelet transform coefficients of each layer, the improvement coefficients corresponding to the wavelet transform coefficients of each layer can be used to perform wavelet reconstruction processing, that is, reverse signal reconstruction processing, to obtain the financial speech signal after noise reduction.

[0032] S5. Perform feature extraction on the financial speech signal after noise reduction to obtain corresponding speech signal features, and input the speech signal features into a preset speech recognition model for speech recognition processing to obtain corresponding transcription text.

[0033] In specific implementation, after obtaining the denoised financial speech signal, the denoised financial speech signal can be used for speech recognition. To ensure the accuracy of speech recognition, the speech recognition can be completed by extracting Mel frequency cepstrum features for neural network recognition. Other speech recognition methods can also be used according to actual conditions. This method first uses the Mel frequency cepstrum coefficient algorithm to extract the Mel frequency cepstrum features of the denoised financial speech signal to obtain the Mel frequency cepstrum features of the financial speech signal. The Mel frequency cepstrum feature extraction process includes: Pre-emphasis: The speech signal is processed through a high-pass filter to enhance the high-frequency part to compensate for the high-frequency part of the speech signal suppressed by the pronunciation system.

[0034] Framing: Divide the speech signal into several short time frames, each frame is 20-40 milliseconds.

[0035] Windowing: Use the corresponding window function to window each frame of signal. The window function can be a Hamming window.

[0036] Fast Fourier transform processing: Perform fast Fourier transform processing on each frame of the windowed signal to convert the time domain signal into a frequency domain signal.

[0037] Mel filter filtering: The frequency domain signal is filtered through a Mel filter bank, and the signal is logarithmically operated through the Mel filter to convert the amplitude spectrum into a logarithmic energy spectrum.

[0038] Discrete cosine transform: Perform discrete cosine transform on the logarithmic energy spectrum of each Mel filter and convert it into cepstral coefficients.

[0039] Normalization: The obtained cepstral coefficients are normalized to obtain the Mel-frequency cepstral features of the financial speech signal.

[0040] After extracting the Mel-frequency cepstrum features of the financial speech signal, the Mel-frequency cepstrum features of the financial speech signal are input into a preset speech recognition model for speech recognition processing to obtain a corresponding transcription text, and the speech recognition model adopts a trained Whisper speech recognition model.

[0041] S6. Perform keyword extraction on the transcribed text to obtain a number of financial keywords, and perform syntactic analysis on the transcribed text to obtain context dependencies corresponding to each financial keyword.

[0042] In specific implementation, a keyword extraction method based on Word2Vec word clustering can be used to extract keywords from the transcribed text, or other supervised or unsupervised methods can be used to extract keywords from the transcribed text according to actual conditions to obtain several financial keywords. Then, the Arc-standard method is used to perform dependency analysis on the transcribed text to obtain the context dependency corresponding to each financial keyword, or other dependency analysis methods are used to perform dependency analysis on the transcribed text according to actual conditions, and the context dependency includes context dependency words and corresponding dependency types.

[0043] S7. Utilize each financial keyword and its context dependency to form a corresponding financial training corpus, and compile the corresponding financial training corpus into a corpus, which is used to perform corpus training on a large financial model.

[0044] In specific implementation, after extracting each financial keyword and the corresponding context dependency, each financial keyword and its context dependency can be used to form a corresponding financial training corpus, and the corresponding financial training corpus can be compiled into a corpus so that the corpus can be used to perform corpus training on the financial big model later.

[0045] The method of this embodiment can realize intelligent transcription from speech to financial big model training corpus, replacing the traditional method of manually collecting text and annotating financial big model training corpus, and quickly and reliably automatically transcribe financial speech into corresponding financial big model training corpus, which can significantly improve the processing efficiency of financial big model training corpus and help realize automated corpus training of financial big models.

[0046] Embodiment 2: This embodiment provides an intelligent transcription system for a large financial model, such as Figure 2 As shown, it includes a speech acquisition unit, a signal decomposition unit, a threshold filtering unit, a signal reconstruction unit, a speech recognition unit, a text parsing unit and a corpus construction unit, wherein: A speech collection unit, used to collect financial speech for making financial model training corpus, and convert the financial speech into corresponding financial speech signals; A signal decomposition unit, used to perform wavelet transform processing of a set number of decomposition layers on the financial speech signal to obtain wavelet transform coefficients corresponding to each layer; A threshold filtering unit is used to perform threshold filtering on each wavelet transform coefficient based on a set wavelet threshold to obtain an improved coefficient corresponding to each wavelet transform coefficient; A signal reconstruction unit, used for performing wavelet reconstruction processing based on the improved coefficients corresponding to each wavelet transform coefficient to obtain a financial speech signal after noise reduction; A speech recognition unit is used to extract features from the financial speech signal after noise reduction to obtain corresponding speech signal features, and input the speech signal features into a preset speech recognition model for speech recognition processing to obtain corresponding transcription text; A text parsing unit is used to extract keywords from the transcribed text to obtain a number of financial keywords, and to perform syntactic analysis on the transcribed text to obtain context dependencies corresponding to each financial keyword; The corpus construction unit is used to use various financial keywords and their contextual dependencies to form corresponding financial training corpus, and compile the corresponding financial training corpus into a corpus, which is used to perform corpus training on the financial big model.

[0047] Embodiment 3: This embodiment provides an intelligent transcription system for a large financial model, such as Figure 3 As shown, at the hardware level, it includes: Data interface, used to establish data connection between the processor and the external data terminal; A memory for storing instructions; The processor is used to read the instructions stored in the memory and execute the intelligent transcription method for financial large models in Example 1 according to the instructions.

[0048] Optionally, the system further includes an internal bus, through which the processor, the memory and the data interface can be interconnected, and the internal bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc.

[0049] The memory may include, but is not limited to, random access memory (RAM), read only memory (ROM), flash memory, first input first output (FIFO) and / or first in last out (FILO), etc. The processor may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components.

[0050] Embodiment 4: This embodiment provides a computer-readable storage medium, on which instructions are stored. When the instructions are executed on a computer, the computer executes the intelligent transcription method for a large financial model in Embodiment 1. The computer-readable storage medium refers to a carrier for storing data, which may include but is not limited to a floppy disk, an optical disk, a hard disk, a flash memory, a USB flash drive, and / or a memory stick, etc. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0051] This embodiment also provides a computer program product, which, when running on a computer, executes the intelligent transcription method for a large financial model in Embodiment 1. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0052] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An intelligent transcription method for a large financial model, characterized in that: include: Collect financial speech for making financial model training corpus, and convert financial speech into corresponding financial speech signals; Perform wavelet transform processing on the financial speech signal with a set number of decomposition layers to obtain the wavelet transform coefficients corresponding to each layer; Perform threshold filtering on each wavelet transform coefficient based on the set wavelet threshold to obtain the improvement coefficient corresponding to each wavelet transform coefficient; Perform wavelet reconstruction based on the improved coefficients corresponding to each wavelet transform coefficient to obtain the financial speech signal after noise reduction; Extract features from the de-noised financial speech signal to obtain corresponding speech signal features, and input the speech signal features into a preset speech recognition model for speech recognition processing to obtain corresponding transcription text; Extract keywords from the transcribed text to obtain several financial keywords, and perform syntactic analysis on the transcribed text to obtain context dependencies corresponding to each financial keyword; The corresponding financial training corpus is composed of each financial keyword and its context dependency, and the corresponding financial training corpus is compiled into a corpus, and the corpus is used to perform corpus training on the financial big model.

2. According to claim 1, an intelligent transcription method for a large financial model is characterized in that: The wavelet transform processing of the financial speech signal with a set number of decomposition layers is performed to obtain the wavelet transform coefficients corresponding to each layer, including: The DB4 wavelet basis function is used to perform 4-layer wavelet decomposition on the financial speech signal to obtain the wavelet transform coefficients corresponding to each layer.

3. The intelligent transcription method for a large financial model according to claim 1, characterized in that: The threshold filtering process is performed on each wavelet transform coefficient based on the set wavelet threshold to obtain the improved coefficient corresponding to each wavelet transform coefficient, including: Substitute the set wavelet threshold and the corresponding wavelet transform coefficient into the preset wavelet threshold function for threshold filtering to obtain the improved coefficient corresponding to the wavelet transform coefficient. The wavelet threshold function is: Wherein, x represents the wavelet transform coefficient, y represents the improved coefficient corresponding to the wavelet transform coefficient x, sgn represents the step function, λ is the set wavelet threshold, δ is the set first coefficient, and δ∈(0,1), ω is the set second coefficient, and ω>1; the wavelet threshold λ is expressed as Among them, j is the decomposition level of wavelet transform, σ is the set noise variance, and N represents the signal length of the financial speech signal.

4. The intelligent transcription method for a large financial model according to claim 3, characterized in that: The financial speech signal is expressed as , where t is the time parameter, is a financial voice signal, For financial voice signals The signal component without noise in For financial voice signals The noise signal component in .

5. The intelligent transcription method for a large financial model according to claim 1, characterized in that: The feature extraction of the de-noised financial speech signal to obtain the corresponding speech signal features includes: The Mel-frequency cepstrum coefficient algorithm is used to extract the Mel-frequency cepstrum features of the denoised financial speech signal to obtain the Mel-frequency cepstrum features of the financial speech signal.

6. The intelligent transcription method for a large financial model according to claim 5, characterized in that: The step of inputting the speech signal features into a preset speech recognition model for speech recognition processing to obtain a corresponding transcription text includes: The Mel-frequency cepstrum features of the financial speech signal are input into a preset speech recognition model for speech recognition processing to obtain a corresponding transcription text, wherein the speech recognition model adopts a trained Whisper speech recognition model.

7. The intelligent transcription method for a large financial model according to claim 1, characterized in that: The keyword extraction is performed on the transcribed text to obtain a number of financial keywords, and the syntactic analysis is performed on the transcribed text to obtain the context dependency relationship corresponding to each financial keyword, including: The keyword extraction method based on Word2Vec word clustering is used to extract keywords from the transcribed text to obtain several financial keywords, and the Arc-standard method is used to perform dependency analysis on the transcribed text to obtain the context dependency corresponding to each financial keyword, wherein the context dependency includes context dependency words and corresponding dependency types.

8. An intelligent transcription system for large financial models, characterized in that: It includes a speech acquisition unit, a signal decomposition unit, a threshold filtering unit, a signal reconstruction unit, a speech recognition unit, a text parsing unit and a corpus construction unit, wherein: A speech collection unit, used to collect financial speech for making financial model training corpus, and convert the financial speech into corresponding financial speech signals; A signal decomposition unit, used to perform wavelet transform processing of a set number of decomposition layers on the financial speech signal to obtain wavelet transform coefficients corresponding to each layer; A threshold filtering unit is used to perform threshold filtering on each wavelet transform coefficient based on a set wavelet threshold to obtain an improved coefficient corresponding to each wavelet transform coefficient; A signal reconstruction unit, used for performing wavelet reconstruction processing based on the improved coefficients corresponding to each wavelet transform coefficient to obtain a financial speech signal after noise reduction; A speech recognition unit is used to extract features from the financial speech signal after noise reduction to obtain corresponding speech signal features, and input the speech signal features into a preset speech recognition model for speech recognition processing to obtain corresponding transcription text; A text parsing unit is used to extract keywords from the transcribed text to obtain a number of financial keywords, and to perform syntactic analysis on the transcribed text to obtain context dependencies corresponding to each financial keyword; The corpus construction unit is used to use various financial keywords and their contextual dependencies to form corresponding financial training corpus, and compile the corresponding financial training corpus into a corpus, which is used to perform corpus training on the financial big model.

9. An intelligent transcription system for large financial models, characterized in that: include: A memory for storing instructions; A processor is used to read the instructions stored in the memory and execute the intelligent transcription method for financial large models described in any one of claims 1-7 according to the instructions.

10. A computer program product, characterized in that When the computer program product is run on a computer, the intelligent transcription method for a large financial model described in any one of claims 1 to 7 is executed.

Citation Information

Patent Citations

  • Model training method, device and equipment

    CN110443291A

  • Wavelet-based wireless voice noise reduction device and method

    CN110931039A

  • Call service processing method, system and device, storage medium and program product

    CN118450053A

Cited By

  • Terahertz transmission-based filter stick bead blasting state discrimination method

    CN122346676A