Speech recognition method, speech recognition apparatus, electronic device, and storage medium

Abstract

This application provides a speech recognition method, speech recognition device, electronic device, and storage medium, belonging to the field of financial technology. The method includes: acquiring target speech data; extracting acoustic features from the target speech data to obtain target acoustic features; decoding the target speech data based on a decoding model to obtain a target word graph; the target word graph includes word nodes and a speech feature sequence, the speech feature sequence including at least two speech words; constructing multiple candidate sentence paths based on the word nodes and speech feature sequence; initially scoring the speech words of each candidate sentence path to obtain a preliminary word score; correcting the preliminary word score based on the target acoustic features to obtain a target word score; filtering the candidate sentence paths according to the target word score to obtain a target sentence path, and concatenating the speech words of the target sentence path to obtain target sentence data. This application can improve the accuracy of speech recognition.

Description

Technical Field

[0001] This application relates to the field of financial technology, and more particularly to a speech recognition method, speech recognition device, electronic device, and storage medium. Background Technology

[0002] With the development of network, communication, and computer technologies, enterprises are exhibiting characteristics of digitalization, remote operation, virtualization, and networking, leading to a surge in online businesses. Communication and dialogue between customers and enterprises have also evolved from face-to-face consultations and negotiations to remote exchanges and communication based on the internet, telephone, and other means. Against this backdrop, intelligent voice interaction is being widely applied in fields such as finance, logistics, and customer service.

[0003] For example, voice-interactive financial trading platforms handle a large volume of telephone calls daily, addressing diverse customer needs including pre-sales inquiries, purchases, after-sales support, and complaints. During these calls, intelligent customer service robots need to respond appropriately to different users. If the intelligent robot fails to accurately identify the user's needs expressed in the voice data, it can lead to inconsistencies between the voice-based responses and the user's requirements, impacting service quality and customer satisfaction.

[0004] Most current speech recognition methods rely on neural network models. However, the recognition process of commonly used neural network models is relatively complex and prone to semantic recognition errors, which affects the accuracy of speech recognition. Therefore, how to improve the accuracy of speech recognition has become an urgent technical problem to be solved. Summary of the Invention

[0005] The main objective of this application is to provide a speech recognition method, speech recognition device, electronic device, and storage medium, which aim to improve the accuracy of speech recognition.

[0006] To achieve the above objectives, a first aspect of this application proposes a speech recognition method, the method comprising:

[0007] Acquire target speech data;

[0008] Acoustic features are extracted from the target speech data to obtain the target acoustic features;

[0009] The target speech data is decoded based on a preset decoding model to obtain a target word map; wherein the target word map includes word nodes and speech feature sequences, the speech feature sequences include at least two speech words, and each speech word is connected to two adjacent word nodes;

[0010] Based on the word nodes of the target word map and the speech feature sequence, multiple candidate sentence paths are constructed;

[0011] A preliminary word score is obtained by scoring the phonetic words in each candidate sentence path.

[0012] The initial word scores are corrected based on the target acoustic features to obtain the target word scores;

[0013] The candidate sentence paths are filtered based on the target word scores to obtain the target sentence paths, and the speech words in the target sentence paths are concatenated to obtain the target sentence data.

[0014] In some embodiments, the step of extracting acoustic features from the target speech data to obtain target acoustic features includes:

[0015] The target speech data is input into a preset acoustic model, which includes a temporal convolutional layer and a fully connected layer.

[0016] Based on the temporal convolutional layer, feature extraction is performed on the target speech data to obtain preliminary acoustic features;

[0017] The target acoustic features are obtained by performing feature filtering on the preliminary acoustic features based on the fully connected layer.

[0018] In some embodiments, the word node includes a start node, an end node, and multiple intermediate nodes. The construction of multiple candidate sentence paths based on the word nodes of the target word graph and the speech feature sequence includes:

[0019] The first target weight between the starting node and the intermediate node is calculated according to a preset algorithm;

[0020] The second target weight between the end node and the intermediate node is calculated according to the preset algorithm;

[0021] The third target weight between each intermediate node and other intermediate nodes is calculated according to the preset algorithm.

[0022] Based on the first target weight, the second target weight, and the third target weight, a speech word traversal is performed on each intermediate node to obtain multiple candidate sentence paths.

[0023] In some embodiments, the step of refining the initial word score based on the target acoustic features to obtain the target word score includes:

[0024] The speech words in each candidate sentence path are scored again to obtain the intermediate word score;

[0025] The initial word score is scored and corrected based on the preset formula, the target acoustic features, and the intermediate word scores to obtain the target word score.

[0026] In some embodiments, the step of filtering the candidate sentence paths based on the target word scores to obtain the target sentence path, and concatenating the speech words in the target sentence path to obtain the target sentence data, includes:

[0027] The target word scores of each candidate sentence path are summed to obtain the candidate sentence score of the candidate sentence path;

[0028] The candidate sentence paths are filtered based on the candidate sentence scores to obtain the target sentence path;

[0029] The target sentence data is obtained by concatenating the phonetic words in the target sentence path according to the preset sentence template.

[0030] In some embodiments, the step of filtering the candidate sentence paths based on the candidate sentence scores to obtain the target sentence path includes:

[0031] Compare the candidate sentence scores for all the candidate sentence paths;

[0032] The candidate sentence path with the lowest candidate sentence score is taken as the target sentence path.

[0033] In some embodiments, acquiring the target speech data includes:

[0034] Acquire raw speech data;

[0035] The original speech data is processed by frame segmentation to obtain initial speech data;

[0036] The initial speech data is subjected to spectral transformation to obtain the target speech data, wherein the target speech data is a Mel cepstral plot.

[0037] To achieve the above objectives, a second aspect of this application provides a speech recognition device, the device comprising:

[0038] The data acquisition module is used to acquire target speech data;

[0039] The feature extraction module is used to extract acoustic features from the target speech data to obtain target acoustic features;

[0040] A decoding module is used to decode the target speech data based on a preset decoding model to obtain a target word map; wherein, the target word map includes word nodes and speech feature sequences, the speech feature sequences include at least two speech words, and each speech word connects two adjacent word nodes;

[0041] The path construction module is used to construct multiple candidate sentence paths based on the word nodes of the target word graph and the speech feature sequence;

[0042] The preliminary scoring module is used to give a preliminary score to the speech words in each of the candidate sentence paths and obtain a preliminary word score;

[0043] The scoring correction module is used to correct the preliminary word scores based on the target acoustic features to obtain the target word scores.

[0044] The filtering module is used to filter the candidate sentence paths according to the target word scores to obtain the target sentence paths, and to concatenate the speech words in the target sentence paths to obtain the target sentence data.

[0045] To achieve the above objectives, a third aspect of the present application provides an electronic device, the electronic device including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method described in the first aspect.

[0046] To achieve the above objectives, a fourth aspect of the present application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in the first aspect.

[0047] The speech recognition method, speech recognition device, electronic device, and storage medium proposed in this application acquire target speech data; extract acoustic features from the target speech data to obtain target acoustic features, which can easily identify the target acoustic features of the target speech data. Further, the target speech data is decoded based on a preset decoding model to obtain a target word graph; wherein the target word graph includes word nodes and a speech feature sequence, the speech feature sequence including at least two speech words, each speech word connecting two adjacent word nodes. This method can decode the target speech data into a graph form, intuitively displaying the decoded content of the target speech data, which is beneficial for subsequent speech recognition. Furthermore, based on the word nodes and speech feature sequence of the target word graph, multiple candidate sentence paths are constructed, which can combine multiple speech words based on the connection relationships between word nodes to form multiple candidate sentence paths. The system performs preliminary scoring on the speech words in each candidate sentence path to obtain preliminary word scores. Then, it corrects the preliminary word scores based on the target acoustic features to obtain target word scores. This method can score multiple speech words belonging to the same candidate sentence path in batches, which can significantly improve scoring efficiency and thus speech recognition efficiency. At the same time, the system corrects the preliminary word scores based on the aforementioned target acoustic features to obtain more accurate target word scores, which can significantly improve scoring accuracy. Finally, candidate sentence paths are filtered based on target word scores to obtain target sentence paths. The speech words within these target sentence paths are then concatenated to obtain target sentence data. This method allows for the quantification of candidate sentence paths by score, and the determination of the corresponding target sentence data based on the target speech data. This significantly improves the accuracy of speech recognition, enabling intelligent customer service robots to more accurately identify the needs expressed in the speech data of service recipients during dialogues. This leads to more targeted responses and service feedback, effectively improving the quality and effectiveness of dialogues in financial transactions. Ultimately, intelligent voice dialogue services enhance customer service quality and satisfaction, thereby increasing business conversion rates. Attached Figure Description

[0048] Figure 1 This is a flowchart of the speech recognition method provided in the embodiments of this application;

[0049] Figure 2 yes Figure 1 The flowchart of step S101 in the text;

[0050] Figure 3 yes Figure 1 The flowchart of step S102 in the document;

[0051] Figure 4 yes Figure 1The flowchart of step S104 in the process;

[0052] Figure 5 yes Figure 1 The flowchart of step S106 in the process;

[0053] Figure 6 yes Figure 1 The flowchart of step S107 in the process;

[0054] Figure 7 yes Figure 6 The flowchart of step S602 in the document;

[0055] Figure 8 This is a schematic diagram of the structure of the speech recognition device provided in the embodiments of this application;

[0056] Figure 9 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0058] It should be noted that although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, and the aforementioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0059] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0060] First, let's analyze some of the terms used in this application:

[0061] Artificial intelligence (AI) is a new branch of computer science that studies, develops, and applies theories, methods, technologies, and systems to simulate, extend, and expand human intelligence. It aims to understand the essence of intelligence and produce intelligent machines that can react in a way similar to human intelligence. Research in this field includes robotics, speech recognition, image recognition, natural language processing, and expert systems. AI can simulate the information processes of human consciousness and thought. Furthermore, AI utilizes digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceiving the environment, acquiring knowledge, and using that knowledge to achieve optimal results.

[0062] Natural Language Processing (NLP): NLP uses computers to process, understand, and utilize human language (such as Chinese and English). NLP is a branch of artificial intelligence and an interdisciplinary field of computer science and linguistics, often referred to as computational linguistics. NLP includes syntactic analysis, semantic analysis, and discourse understanding. It is commonly used in machine translation, handwritten and printed character recognition, speech recognition and text-to-speech conversion, intent recognition, information extraction and filtering, text classification and clustering, sentiment analysis, and opinion mining. It involves data mining, machine learning, knowledge acquisition, knowledge engineering, artificial intelligence research, and linguistic research related to language computation.

[0063] Information Extraction (NER) is a text processing technique that extracts factual information such as entities, relationships, and events from natural language text and outputs it as structured data. Information extraction is a technique for extracting specific information from text data. Text data is composed of specific units, such as sentences, paragraphs, and chapters. Text information is composed of smaller, specific units, such as characters, words, phrases, sentences, paragraphs, or combinations of these units. Extracting noun phrases, names of people, and place names from text data is an example of text information extraction. Of course, text information extraction techniques can extract information of various types.

[0064] Decoding: This involves transforming a previously generated fixed vector into an output sequence; the input sequence can be text, speech, image, or video; the output sequence can be text or image.

[0065] A Hidden Markov Model (HMM) is a statistical model used to describe a Markov process with hidden, unknown parameters. The challenge lies in determining these hidden parameters from the observable parameters. These parameters are then used for further analysis, such as pattern recognition. It is the simplest dynamic Bayesian network in structure, a directed graphical model primarily used for time series data modeling.

[0066] Fourier transform: Represents a function that satisfies certain conditions as a linear combination of trigonometric functions (sine and / or cosine functions) or integrals. In different research fields, the Fourier transform has various variants, such as the continuous Fourier transform and the discrete Fourier transform.

[0067] Mel-Frequency Cipstal Coefficients (MFCCs) are a set of key coefficients used to construct the Mel-Frequency Cipstal Spectrum. From a segment of a music signal, a set of cepstrum values ​​can be obtained that is sufficient to represent the music signal. The Mel-Frequency Cipstal Coefficients are the cepstrum values ​​derived from this cepstrum (i.e., the spectrum of the spectrum). Unlike a regular cepstrum, the most distinctive feature of the Mel-Frequency Cipstrum is that its frequency bands are uniformly distributed across the Mel scale. In other words, compared to a typical linear cepstrum representation, this frequency band is closer to the non-linear human auditory system. For example, Mel-Frequency Cipstal Spectrum is frequently used in audio compression techniques.

[0068] With the development of network, communication, and computer technologies, enterprises are exhibiting characteristics of digitalization, remote operation, virtualization, and networking, leading to a surge in online businesses. Communication and dialogue between customers and enterprises have also evolved from face-to-face consultations and negotiations to remote exchanges and communication based on the internet, telephone, and other means. Against this backdrop, intelligent voice interaction is being widely applied in fields such as finance, logistics, and customer service.

[0069] Voice-based financial trading platforms handle a large volume of telephone calls daily, addressing diverse customer needs including pre-sales inquiries, purchases, after-sales support, and complaints. During these calls, intelligent customer service robots need to respond appropriately to different users. If the AI ​​fails to accurately identify the user's needs expressed in the voice data, it can lead to inconsistencies between the voice-based responses and the user's requirements, impacting service quality and customer satisfaction.

[0070] For example, in the process of recommending products through virtual characters on a voice-interactive financial trading platform, when the service recipient has questions and needs consultation and communication, the virtual character can only find answers from pre-set options and cannot accurately identify the service recipient's needs expressed in the voice data, resulting in the phenomenon of "answering the wrong question" and low accuracy of the virtual character's voice interaction.

[0071] Most current speech recognition methods rely on neural network models. However, the recognition process of commonly used neural network models is relatively complex and prone to semantic recognition errors, which affects the accuracy of speech recognition. Therefore, how to improve the accuracy of speech recognition has become an urgent technical problem to be solved.

[0072] Based on this, embodiments of this application provide a speech recognition method, a speech recognition device, an electronic device, and a storage medium, aiming to improve the accuracy of speech recognition.

[0073] The speech recognition method, speech recognition device, electronic device and storage medium provided in the embodiments of this application are specifically described through the following embodiments. First, the speech recognition method in the embodiments of this application is described.

[0074] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0075] Foundational technologies for artificial intelligence generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies mainly encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0076] The speech recognition method provided in this application relates to the field of artificial intelligence technology. The speech recognition method provided in this application can be applied to a terminal, a server, or software running on either a terminal or a server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, etc.; the server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application implementing the speech recognition method, but is not limited to the above forms.

[0077] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0078] It should be noted that in all specific embodiments of this application, when processing data related to user identity or characteristics, such as user information, user behavior data, user voice data, user historical data, and user location information, user permission or consent is obtained first. Furthermore, the collection, use, and processing of this data comply with relevant laws, regulations, and standards. In addition, when embodiments of this application require access to sensitive personal information of users, separate permission or consent from the user is obtained through pop-ups or redirection to a confirmation page. Only after obtaining the user's separate permission or consent is the necessary user-related data required for the proper functioning of these embodiments acquired.

[0079] Figure 1 This is an optional flowchart of the speech recognition method provided in the embodiments of this application. Figure 1 The method may include, but is not limited to, steps S101 to S107.

[0080] Step S101: Obtain target speech data;

[0081] Step S102: Extract acoustic features from the target speech data to obtain the target acoustic features;

[0082] Step S103: Decode the target speech data based on the preset decoding model to obtain the target word map; wherein, the target word map includes word nodes and speech feature sequences, the speech feature sequences include at least two speech words, and each speech word connects two adjacent word nodes;

[0083] Step S104: Based on the word nodes and speech feature sequences of the target word graph, construct multiple candidate sentence paths;

[0084] Step S105: Perform preliminary scoring on the phonetic words in each candidate sentence path to obtain preliminary word scores;

[0085] Step S106: Based on the target acoustic features, the preliminary word scores are corrected to obtain the target word scores;

[0086] Step S107: The candidate sentence paths are filtered according to the target word scores to obtain the target sentence path, and the speech words in the target sentence path are concatenated to obtain the target sentence data.

[0087] Steps S101 to S107 of this embodiment involve acquiring target speech data and extracting acoustic features from the target speech data to obtain target acoustic features, which can facilitate the identification of target acoustic features of the target speech data. Further, the target speech data is decoded based on a preset decoding model to obtain a target word graph; wherein the target word graph includes word nodes and a speech feature sequence, the speech feature sequence including at least two speech words, each speech word connecting two adjacent word nodes. This method can decode the target speech data into a graph form, intuitively displaying the decoded content of the target speech data, which is beneficial for subsequent speech recognition. Further, based on the word nodes and speech feature sequence of the target word graph, multiple candidate sentence paths are constructed, which can combine multiple speech words based on the connection relationship between word nodes to form multiple candidate sentence paths. Preliminary word scores are awarded for each candidate sentence path's speech words. These scores are then refined based on target acoustic features to obtain target word scores. This approach, using candidate sentence paths as batch units, scores multiple speech words belonging to the same path, significantly improving scoring efficiency and thus speech recognition efficiency. Furthermore, refining the preliminary word scores based on the aforementioned target acoustic features yields more accurate target word scores, further enhancing scoring accuracy. Finally, candidate sentence paths are filtered based on target word scores to obtain target sentence paths. The speech words within these target sentence paths are then concatenated to obtain target sentence data. This score-based filtering of candidate sentence paths and the determination of target sentence data based on target speech paths significantly improves speech recognition accuracy.

[0088] Please see Figure 2 In some embodiments, step S101 may include, but is not limited to, steps S201 to S203:

[0089] Step S201: Obtain raw voice data;

[0090] Step S202: Perform frame segmentation on the original speech data to obtain initial speech data;

[0091] Step S203: Perform spectral transformation on the initial speech data to obtain the target speech data, wherein the target speech data is a Mel cepstral plot.

[0092] In step S201 of some embodiments, raw audio data can be obtained by writing a web crawler, setting up a data source, and then selectively crawling data. The data source can be various types of online platforms, social media, or specific audio databases. The raw audio data can be the speaker's music, a speech, a chat conversation, etc. Raw audio data can also be obtained through other methods, and is not limited to these.

[0093] For example, in financial transaction scenarios, raw voice data is audio data containing commonly used dialogue in the financial field; in insurance sales scenarios, raw voice data is audio data containing descriptions of a particular insurance product, such as its type, cost, and target audience.

[0094] In step S202 of some embodiments, the original speech data is processed by frame segmentation to obtain initial speech data. Specifically, the original speech data is processed by signal frame segmentation and windowing to obtain multiple speech segments. A short-time Fourier transform is performed on each speech segment to convert the time-domain features of the speech segment into frequency-domain features. Finally, the frequency-domain features of each frame are stacked in the time dimension to obtain a target spectrogram, which is used as the initial speech data.

[0095] In step S203 of some embodiments, the initial speech data is filtered using a preset Mel-Cepstral Spectrum filter bank. First, a logarithmic operation is performed on the initial speech data to obtain the target logarithmic spectrum. Then, an inverse Fourier transform is performed on the target logarithmic spectrum to obtain the target Mel-Cepstral Spectrum. Further, feature extraction is performed on the target Mel-Cepstral Spectrum to obtain the target Mel-Cepstral coefficients, which are then used as the target speech data.

[0096] Through the above steps S201 to S203, the original speech data can be easily converted into spectral features, and the spectral features can be filtered to obtain the target Mel-frequency cepstral coefficients (i.e., the target speech data). This allows the speech content of the target speech data to be identified through the spectral features, thereby improving the accuracy of speech recognition.

[0097] Please see Figure 3 In some embodiments, step S102 may include, but is not limited to, steps S301 to S303:

[0098] Step S301: Input the target speech data into a preset acoustic model, which includes a temporal convolutional layer and a fully connected layer.

[0099] Step S302: Extract features from the target speech data based on the temporal convolutional layer to obtain preliminary acoustic features;

[0100] Step S303: Based on the fully connected layer, perform feature filtering on the preliminary acoustic features to obtain the target acoustic features.

[0101] In step S301 of some embodiments, the acoustic model may be constructed based on a deep neural network or a hidden Markov model. For example, a deep neural network may be used to construct the acoustic model, which includes temporal convolutional layers and fully connected layers.

[0102] In step S302 of some embodiments, feature extraction is performed on the target speech data based on the temporal convolutional layer to obtain temporal feature information in the target speech data and obtain preliminary acoustic features.

[0103] In step S303 of some embodiments, the initial acoustic features are classified based on the prediction function of the fully connected layer (e.g., the softmax function, etc.) to obtain labeled acoustic features. Each labeled acoustic feature includes an initial acoustic feature and an acoustic label for the initial acoustic feature. The acoustic label includes timbre type, pitch type, etc. Based on the acoustic label, target acoustic features that meet the current requirements are selected from multiple labeled acoustic features. The target acoustic features may contain various audio information such as pitch, duration, and frequency of the target speech data.

[0104] For example, the speech features of target speech data include semantic features, emotional features, regional features, and speech rate features. Speech features could be: "Inquiring about credit card issues"; historical business features could be: "Credit card delinquency, more than two delinquencies in history, low salary"; user statistical features could be: "Preference for high-limit credit cards"; and interactive content could be: "Hello, you have been delinquent multiple times, which has greatly affected your credit. We hope you can repay as soon as possible." Simultaneously, speech features could be: "Normal mood," "A City dialect," and "Normal speech rate." User static features could be: "Male," "Residence in A City," "Age 38," "Introverted personality," and "Loves anime"; user statistical features could be: "Preference for deposit business"; interactive emotional features could be: "Normal mood," interactive regional features could be: "A City dialect," interactive stylistic features could be: "Game style," and interactive speech rate features could be: "Normal speech rate."

[0105] Through the above steps S301 to S303, various acoustic features of the target speech data, such as pitch, duration, and frequency, can be easily identified, enabling the acquired target acoustic features to be used in the subsequent speech recognition process, thereby improving the accuracy of speech recognition.

[0106] In step S103 of some embodiments, the preset decoding model can be a static decoder based on WFST, where WFST refers to a weighted finite-state transducer. Using this decoding model to decode the target speech data converts the decoded content of the target speech data into a directed graph, resulting in a target word graph (Lattice). The target word graph (Lattice) stores the recognized word sequences, which can be used to represent the speech content of the recognized target speech data. The target word graph (Lattice) includes multiple word nodes and speech feature sequences, each speech feature sequence including at least two speech words, with each speech word connected to two adjacent word nodes. This method decodes the target speech data into a directed graph, visually displaying the decoded content of the target speech data, which is beneficial for subsequent speech recognition.

[0107] Please see Figure 4 In some embodiments, the word node includes a start node, an end node, and multiple intermediate nodes, and step S104 may include, but is not limited to, steps S401 to S404:

[0108] Step S401: Calculate the first target weight between the starting node and the intermediate node according to the preset algorithm;

[0109] Step S402: Calculate the second target weight between the end node and the intermediate node according to the preset algorithm;

[0110] Step S403: Calculate the third objective weight between each intermediate node and other intermediate nodes according to the preset algorithm;

[0111] Step S404: Based on the first target weight, the second target weight, and the third target weight, perform speech word traversal on each intermediate node to obtain multiple candidate sentence paths.

[0112] In step S401 of some embodiments, the preset algorithm can be multiple algorithms included in the Hidden Markov Model, such as the forward algorithm, the Viterbi algorithm, and the forward-backward algorithm. The Hidden Markov Model is used to calculate the sequence probability of the starting node and intermediate nodes to obtain the weight value between the starting node and each intermediate node. The weight value with the smallest value is taken as the first target weight α, and the intermediate node corresponding to the first target weight α is taken as the optimal successor node starting from the starting node.

[0113] In step S402 of some embodiments, the sequence probability is calculated for the starting node and intermediate nodes using a hidden Markov model to obtain the weight value between the intermediate node and each intermediate node. The weight value with the smallest value is taken as the second target weight β, and the intermediate node corresponding to the second target weight β is taken as the optimal predecessor node of the ending node.

[0114] In step S403 of some embodiments, a hidden Markov model is used to calculate the sequence probability of each intermediate node and other intermediate nodes to obtain the weight value between each intermediate node and other intermediate nodes. The weight value is used as the third objective weight between each intermediate node and other intermediate nodes. The magnitude of the third objective weight can reflect the optimal predecessor node and the optimal successor node of each intermediate node.

[0115] In step S404 of some embodiments, when performing speech word traversal on each intermediate node according to the first target weight, the second target weight, and the third target weight, each edge originating from each intermediate node is traversed, that is, the speech words between the intermediate node and all its optimal successor nodes are traversed, thereby generating multiple candidate sentence paths. It should be noted that a candidate sentence path includes multiple word nodes and speech words on a valid path, wherein the valid path is the path between the start node and the end node.

[0116] Through the above steps S401 to S404, it is relatively convenient to construct the connection relationship between word nodes based on the weight between word nodes, thereby forming multiple candidate sentence paths, making the generated candidate sentence paths more reasonable.

[0117] In step S105 of some embodiments, a Neural Networks Learn Language Model (NNLM model) can be used to initially score the speech words in each candidate sentence path, obtaining preliminary word scores. Specifically, the speech words in the input candidate sentence path are one-hot encoded based on the neural network language model to obtain speech coding features. Then, each speech coding feature is multiplied by a preset reference matrix to obtain a target speech word matrix. Finally, the prediction function of the neural network language model is used to initially score each target speech word matrix, obtaining a preliminary word score for each speech word. The prediction function can be a softmax function, a tanh function, etc., without limitation. This method can conveniently use candidate sentence paths as a batch unit to score multiple speech words belonging to the same candidate sentence path using the neural network language model, which can significantly improve scoring efficiency and thus improve the efficiency of speech recognition.

[0118] Please see Figure 5In some embodiments, step S106 may include, but is not limited to, steps S501 to S502:

[0119] Step S501: Re-evaluate the speech words in each candidate sentence path to obtain the intermediate word score;

[0120] Step S502: The preliminary word scores are scored and corrected according to the preset formula, target acoustic features, and intermediate word scores to obtain the target word scores.

[0121] In step S501 of some embodiments, a common language model (e.g., an N-gram model, i.e., an N-gram model, etc.) can be used to re-score the speech words in each candidate sentence path to obtain intermediate word scores. For example, an n-gram model can be used to perform one-hot encoding on the speech words of the input candidate sentence path to obtain speech encoding features. The prediction function of the n-gram model can be used to score each speech encoding feature to calculate the intermediate word score for each speech word. The prediction function can be a softmax function, a tanh function, etc., without limitation.

[0122] In step S502 of some embodiments, the correlation between each speech word and the target acoustic feature can be calculated first. For example, cosine similarity algorithm, coordinated filtering algorithm, Euclidean distance, etc., can be used to calculate the correlation between the speech word and the target acoustic feature to obtain the acoustic correlation M. Then, according to a preset formula, the acoustic correlation M is first subtracted from the intermediate word score Q, and the result of the subtraction is added to the preliminary word score P to obtain the target word score N. That is, the preset formula is:

[0123] The target word score N = acoustic relevance M - intermediate word score Q + preliminary word score P.

[0124] Through the above steps S501 to S502, it is relatively convenient to use ordinary language models to score multiple speech words belonging to the same candidate sentence path again, based on the candidate sentence path, to obtain intermediate word scores. Then, based on the aforementioned target acoustic features and intermediate word scores, the initial word scores are corrected to obtain more accurate target word scores. This method can improve the scoring accuracy and thus improve the efficiency of speech recognition.

[0125] Please see Figure 6 In some embodiments, step S107 includes, but is not limited to, steps S601 to S603:

[0126] Step S601: Sum the target word scores for each candidate sentence path to obtain the candidate sentence score for the candidate sentence path;

[0127] Step S602: Filter the candidate sentence paths according to the candidate sentence scores to obtain the target sentence path;

[0128] Step S603: Concatenate the speech words in the target sentence path according to the preset sentence template to obtain the target sentence data.

[0129] In step S601 of some embodiments, the sum function or other statistical functions can be used to sum the target word scores of each candidate sentence path to obtain the candidate sentence score of the candidate sentence path. For example, if a candidate sentence path contains 5 speech words, and each speech word corresponds to a target word score, then the sum function is used to sum the target word scores of these 5 speech words to obtain the sum of the 5 target word scores, and the sum of the 5 target word scores is used as the candidate sentence score of the candidate sentence path.

[0130] In step S602 of some embodiments, the size of the candidate sentence score can intuitively reflect whether the candidate sentence path is the shortest path. The shorter the candidate sentence path, that is, the smaller the candidate sentence score, the closer the sentence content represented by the candidate sentence path is to the true semantic content of the current target speech. Therefore, the candidate sentence path can be screened according to the candidate sentence score, and the candidate sentence path with the smaller candidate sentence score can be selected as the target sentence path.

[0131] In step S603 of some embodiments, the preset sentence template includes preset sentence format content, such as the font size, font type, character spacing, paragraph spacing, etc. All speech words in the candidate sentence path are concatenated according to the node order of the word nodes in the target sentence path to obtain the target sentence path.

[0132] For example, candidate sentence path A includes the following phonetic words and their target word scores: orange (0.5), is (0.2), fruit (0.1); candidate sentence path B includes the following phonetic words and their target word scores: sentence (0.2), is (0.2), fruit (0.1); then the candidate sentence score for candidate sentence path A is 0.5 + 0.2 + 0.1 = 0.8; the candidate sentence score for candidate sentence path B is 0.2 + 0.2 + 0.1 = 0.5; therefore, based on the candidate sentence scores of the candidate sentence paths, candidate sentence path A is selected as the target sentence path, and the phonetic words in the target sentence path are concatenated to obtain the target sentence data "orange is fruit".

[0133] Through the above steps S601 to S603, the candidate sentence score of each candidate sentence path can be obtained relatively easily, and the target sentence path can be selected from multiple candidate sentence paths based on the candidate sentence score. The candidate sentence path can be selected in a score quantification manner, and the target sentence data corresponding to the target speech data can be determined based on the target sentence path. The target sentence data is used to represent the speech content of the target speech data, which can improve the accuracy of speech recognition.

[0134] Please see Figure 7 In some embodiments, step S602 may include, but is not limited to, steps S701 to S702:

[0135] Step S701: Compare the candidate sentence scores of all candidate sentence paths;

[0136] Step S702: Select the candidate sentence path with the smallest candidate sentence score as the target sentence path.

[0137] In step S701 of some embodiments, since the shorter the candidate sentence path, that is, the smaller the candidate sentence score, the closer the sentence content represented by the candidate sentence path is to the true semantic content of the current target speech, the candidate sentence path can be screened according to the candidate sentence score. The candidate sentence scores of all candidate sentence paths are compared, and the target sentence path is selected from multiple candidate sentence paths according to the relationship between the candidate sentence scores of multiple candidate sentence paths.

[0138] In step S702 of some embodiments, in order to improve the accuracy of speech recognition, when filtering the target sentence path from multiple candidate sentence paths according to the relationship between the candidate sentence scores of multiple candidate sentence paths, the candidate sentence path with the smallest candidate sentence score is selected. The sentence content of the candidate sentence path with the smallest candidate sentence score is closest to the true semantic content of the current target speech, and the candidate sentence path with the smallest candidate sentence score is taken as the target sentence path.

[0139] Through the above steps S701 to S702, the candidate sentence score of each candidate sentence path can be obtained relatively easily. Based on the candidate sentence scores of all candidate sentence paths, the target sentence path can be selected from multiple candidate sentence paths. The selection of candidate sentence paths can be achieved by quantifying the scores, thereby improving the accuracy and rationality of path selection.

[0140] The speech recognition method of this application acquires target speech data; extracts acoustic features from the target speech data to obtain target acoustic features, which can easily identify the target acoustic features of the target speech data. Further, the target speech data is decoded based on a preset decoding model to obtain a target word graph; wherein the target word graph includes word nodes and a speech feature sequence, the speech feature sequence including at least two speech words, each speech word connecting two adjacent word nodes. This method can decode the target speech data into a graph form, intuitively displaying the decoded content of the target speech data, which is beneficial for subsequent speech recognition. Further, based on the word nodes and speech feature sequence of the target word graph, multiple candidate sentence paths are constructed, which can combine multiple speech words based on the connection relationships between word nodes to form multiple candidate sentence paths. The system performs preliminary scoring on the speech words in each candidate sentence path to obtain preliminary word scores. Then, it corrects the preliminary word scores based on the target acoustic features to obtain target word scores. This method can score multiple speech words belonging to the same candidate sentence path in batches, which can significantly improve scoring efficiency and thus speech recognition efficiency. At the same time, the system corrects the preliminary word scores based on the aforementioned target acoustic features to obtain more accurate target word scores, which can significantly improve scoring accuracy. Finally, candidate sentence paths are filtered based on target word scores to obtain target sentence paths. The speech words within these target sentence paths are then concatenated to obtain target sentence data. This method allows for the quantification of candidate sentence paths by score, and the determination of the corresponding target sentence data based on the target speech data. This significantly improves the accuracy of speech recognition, enabling intelligent customer service robots to more accurately identify the needs expressed in the speech data of service recipients during dialogues. This leads to more targeted responses and service feedback, effectively improving the quality and effectiveness of dialogues in financial transactions. Ultimately, intelligent voice dialogue services enhance customer service quality and satisfaction, thereby increasing business conversion rates.

[0141] Please see Figure 8 This application also provides a speech recognition device that can implement the above-described speech recognition method. The device includes:

[0142] Data acquisition module 801 is used to acquire target speech data;

[0143] Feature extraction module 802 is used to extract acoustic features from target speech data to obtain target acoustic features;

[0144] The decoding module 803 is used to decode the target speech data based on a preset decoding model to obtain a target word map; wherein, the target word map includes word nodes and speech feature sequences, the speech feature sequences include at least two speech words, and each speech word connects two adjacent word nodes;

[0145] The path construction module 804 is used to construct multiple candidate sentence paths based on the word nodes and speech feature sequences of the target word graph;

[0146] The preliminary scoring module 805 is used to give a preliminary score to the phonetic words in each candidate sentence path and obtain a preliminary word score.

[0147] The scoring correction module 806 is used to correct the initial word scores based on the target acoustic features to obtain the target word scores.

[0148] The filtering module 807 is used to filter candidate sentence paths based on the target word score to obtain the target sentence path, and to concatenate the speech words in the target sentence path to obtain the target sentence data.

[0149] The specific implementation of this speech recognition device is basically the same as the specific implementation of the speech recognition method described above, and will not be repeated here.

[0150] This application also provides an electronic device, which includes: a memory, a processor, a program stored in the memory and executable on the processor, and a data bus for communication between the processor and the memory. When the program is executed by the processor, it implements the aforementioned speech recognition method. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0151] Please see Figure 9 , Figure 9 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes:

[0152] The processor 901 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application.

[0153] The memory 902 can be implemented as a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 902 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 902 and is called and executed by the processor 901 using the speech recognition method of the embodiments of this application.

[0154] The input / output interface 903 is used to implement information input and output;

[0155] The communication interface 904 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0156] Bus 905 transmits information between various components of the device (e.g., processor 901, memory 902, input / output interface 903, and communication interface 904);

[0157] The processor 901, memory 902, input / output interface 903, and communication interface 904 are connected to each other within the device via bus 905.

[0158] This application also provides a computer-readable storage medium storing one or more programs that can be executed by one or more processors to implement the above-described speech recognition method.

[0159] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0160] The speech recognition method, speech recognition device, electronic device, and computer-readable storage medium provided in this application acquire target speech data; extract acoustic features from the target speech data to obtain target acoustic features, which can conveniently identify the target acoustic features of the target speech data. Further, the target speech data is decoded based on a preset decoding model to obtain a target word graph; wherein the target word graph includes word nodes and a speech feature sequence, the speech feature sequence including at least two speech words, each speech word connecting two adjacent word nodes. This method can decode the target speech data into a graph form, intuitively displaying the decoded content of the target speech data, which is beneficial for subsequent speech recognition. Further, based on the word nodes and speech feature sequence of the target word graph, multiple candidate sentence paths are constructed, which can combine multiple speech words based on the connection relationship between word nodes to form multiple candidate sentence paths. The system performs preliminary scoring on the speech words in each candidate sentence path to obtain preliminary word scores. Then, it corrects the preliminary word scores based on the target acoustic features to obtain target word scores. This method can score multiple speech words belonging to the same candidate sentence path in batches, which can significantly improve scoring efficiency and thus speech recognition efficiency. At the same time, the system corrects the preliminary word scores based on the aforementioned target acoustic features to obtain more accurate target word scores, which can significantly improve scoring accuracy. Finally, candidate sentence paths are filtered based on target word scores to obtain target sentence paths. The speech words within these target sentence paths are then concatenated to obtain target sentence data. This method allows for the quantification of candidate sentence paths by score, and the determination of the corresponding target sentence data based on the target speech data. This significantly improves the accuracy of speech recognition, enabling intelligent customer service robots to more accurately identify the needs expressed in the speech data of service recipients during dialogues. This leads to more targeted responses and service feedback, effectively improving the quality and effectiveness of dialogues in financial transactions. Ultimately, intelligent voice dialogue services enhance customer service quality and satisfaction, thereby increasing business conversion rates.

[0161] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0162] It will be understood by those skilled in the art that Figure 1-7 The technical solutions shown do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0163] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0164] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0165] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0166] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0167] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0168] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0169] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0170] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0171] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A speech recognition method, characterized in that, The method includes: Acquire target speech data; Acoustic features are extracted from the target speech data to obtain the target acoustic features; The target speech data is decoded based on a preset decoding model to obtain a target word map; wherein the target word map includes word nodes and speech feature sequences, the speech feature sequences include at least two speech words, and each speech word is connected to two adjacent word nodes; Based on the word nodes of the target word map and the speech feature sequence, multiple candidate sentence paths are constructed; A preliminary word score is obtained by scoring the phonetic words in each candidate sentence path. The initial word scores are corrected based on the target acoustic features to obtain the target word scores; The candidate sentence paths are filtered based on the target word scores to obtain the target sentence paths, and the speech words in the target sentence paths are concatenated to obtain the target sentence data. The word node includes a start node, an end node, and multiple intermediate nodes. The construction of multiple candidate sentence paths based on the word nodes of the target word graph and the speech feature sequence includes: The first target weight between the starting node and the intermediate node is calculated according to a preset algorithm; The second target weight between the end node and the intermediate node is calculated according to the preset algorithm; The third target weight between each intermediate node and other intermediate nodes is calculated according to the preset algorithm. Based on the first target weight, the second target weight, and the third target weight, a speech word traversal is performed on each intermediate node to obtain multiple candidate sentence paths.

2. The speech recognition method according to claim 1, characterized in that, The step of extracting acoustic features from the target speech data to obtain target acoustic features includes: The target speech data is input into a preset acoustic model, which includes a temporal convolutional layer and a fully connected layer. Based on the temporal convolutional layer, feature extraction is performed on the target speech data to obtain preliminary acoustic features; The target acoustic features are obtained by performing feature filtering on the preliminary acoustic features based on the fully connected layer.

3. The speech recognition method according to claim 1, characterized in that, The step of refining the initial word score based on the target acoustic features to obtain the target word score includes: The speech words in each candidate sentence path are scored again to obtain the intermediate word score; The initial word score is scored and corrected based on the preset formula, the target acoustic features, and the intermediate word scores to obtain the target word score.

4. The speech recognition method according to claim 1, characterized in that, The process of filtering the candidate sentence paths based on the target word scores to obtain the target sentence paths, and then concatenating the speech words in the target sentence paths to obtain the target sentence data, includes: The target word scores of each candidate sentence path are summed to obtain the candidate sentence score of the candidate sentence path; The candidate sentence paths are filtered based on the candidate sentence scores to obtain the target sentence path; The target sentence data is obtained by concatenating the phonetic words in the target sentence path according to the preset sentence template.

5. The speech recognition method according to claim 4, characterized in that, The step of filtering the candidate sentence paths based on the candidate sentence scores to obtain the target sentence path includes: Compare the candidate sentence scores for all the candidate sentence paths; The candidate sentence path with the lowest candidate sentence score is taken as the target sentence path.

6. The speech recognition method according to any one of claims 1 to 5, characterized in that, The acquisition of target speech data includes: Acquire raw speech data; The original speech data is processed by frame segmentation to obtain initial speech data; The initial speech data is subjected to spectral transformation to obtain the target speech data, wherein the target speech data is a Mel cepstral plot.

7. A voice recognition device, characterized in that, The device includes: The data acquisition module is used to acquire target speech data; The feature extraction module is used to extract acoustic features from the target speech data to obtain target acoustic features; A decoding module is used to decode the target speech data based on a preset decoding model to obtain a target word map; wherein, the target word map includes word nodes and speech feature sequences, the speech feature sequences include at least two speech words, and each speech word connects two adjacent word nodes; The path construction module is used to construct multiple candidate sentence paths based on the word nodes of the target word graph and the speech feature sequence; The preliminary scoring module is used to give a preliminary score to the speech words in each of the candidate sentence paths and obtain a preliminary word score; The scoring correction module is used to correct the preliminary word scores based on the target acoustic features to obtain the target word scores. The filtering module is used to filter the candidate sentence paths according to the target word scores to obtain the target sentence paths, and to concatenate the speech words in the target sentence paths to obtain the target sentence data; The word node includes a start node, an end node, and multiple intermediate nodes. The construction of multiple candidate sentence paths based on the word nodes of the target word graph and the speech feature sequence includes: The first target weight between the starting node and the intermediate node is calculated according to a preset algorithm; The second target weight between the end node and the intermediate node is calculated according to the preset algorithm; The third target weight between each intermediate node and other intermediate nodes is calculated according to the preset algorithm. Based on the first target weight, the second target weight, and the third target weight, a speech word traversal is performed on each intermediate node to obtain multiple candidate sentence paths.

8. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the speech recognition method as described in any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the speech recognition method according to any one of claims 1 to 6.

Images