Radar terminology speech recognition method and system based on dual similarity matching technology

Abstract

The invention discloses a radar terminology speech recognition method and system based on a dual similarity matching technology, and the method specifically comprises the steps: firstly removing punctuations of a Chinese long text obtained through speech recognition, and generating a candidate substring set through a sliding window; then pinyin is obtained from the candidate substrings and the field word bank entries, and a fuzzy rule is applied to obtain normalized pinyin; the sequence similarity and the normalized editing similarity are calculated, and error correction mapping is generated through double-threshold joint judgment; and finally, performing fragment replacement controlled by a position occupation set on the original text according to the error correction mapping to avoid overlapping replacement and cascading error replacement, and outputting an error correction text. According to the method, the accuracy of radar terminology speech recognition is improved, a lexicon is supported to provide fixed pronunciation for polyphones, the stability of terminology matching is improved, the readability and consistency of output are ensured, the algorithm is easy to implement, the requirement for hardware is low, the method does not depend on massive tagged corpora, and engineering implementation and deployment are easy.

Description

Technical Field

[0001] This invention relates to the field of speech recognition and text correction technology, and in particular to a speech recognition method and system for radar terminology based on dual similarity matching technology. Background Technology

[0002] Radar, as an important target detection device, is widely used in military, meteorological, geological, and transportation fields. During radar operation, relying on voice commands and log input is a novel method of radar operation and logging. In the field of speech recognition, the language recognition model based on Automatic Speech Recognition (ASR) technology is the most widely used. However, commonly used general-purpose speech recognition models have not been trained on corpora for many specialized terms in the radar field. Furthermore, due to limitations such as ambient noise, accent differences, and equipment pickup conditions, ASR suffers from numerous recognition errors when handling radar voice input.

[0003] In radar voice operation commands, technical terms and slang are easily confused. For example, "CFAR" is often pronounced as "cifar," and "07" is often pronounced as "dongguai." The similar pronunciations of "track / hangji" and "point track / electrode" can also lead to confusion. In addition, if polyphonic characters are not assigned a pronunciation, it often leads to pronunciation normalization deviation, which further affects the accuracy of matching.

[0004] While additional training of speech models tailored to the specific characteristics of the radar domain can significantly improve recognition errors, it requires massive amounts of labeled corpora, resulting in high training costs. Existing general post-processing methods largely rely on language models or contextual statistics for error correction, making insufficient use of domain-specific lexicons and often employing simple character set replacement strategies. In scenarios where radar terminology is concentrated, this can easily lead to high error correction rates, failing to meet the high-precision speech recognition requirements of the radar domain. Summary of the Invention

[0005] The purpose of this invention is to provide a radar terminology speech recognition method and system based on dual similarity matching technology, which has high error correction accuracy, stable terminology matching, simple algorithm, low hardware requirements, does not rely on massive labeled corpora, and is easy to implement and deploy in engineering.

[0006] The technical solution to achieve the purpose of this invention is: a speech recognition method for radar technical terms based on dual similarity matching technology, comprising the following steps:

[0007] Step 1, Text Segmentation: After removing punctuation from the long Chinese text obtained from speech recognition, a sliding window is used to generate a set of candidate substrings;

[0008] Step 2, Pinyin Acquisition and Fuzzy Rule Normalization: Obtain the pinyin from the candidate substrings and domain vocabulary entries, and apply fuzzy rules to obtain normalized pinyin;

[0009] Step 3, Two-stage matching and dual-threshold determination: The first stage calculates sequence similarity and the second stage normalized edit similarity, and the error correction map is generated by joint determination of dual thresholds;

[0010] Step 4, Safe Replacement: Based on the error correction mapping, replace segments of the original text according to the positional occupancy set control to avoid overlapping replacement and cascading erroneous replacement, and output the error correction text.

[0011] Furthermore, the text segmentation described in step 1 involves removing punctuation from the long Chinese text obtained from speech recognition and then generating a candidate substring set using a sliding window, as detailed below:

[0012] Punctuation removal and whitespace normalization are performed on the input long text T. Then, a candidate set S is generated using a sliding window of n∈{2,3,4,5,6}, covering short near-homophones and common terminology fragments.

[0013] S = { T[i:i+n] | 0 ≤i < |T|, n in {2,3,4,5,6}, i+n ≤|T|}.

[0015] Furthermore, in step 1, Arabic numerals and unit symbols are protected by marking them before text segmentation to prevent them from being mis-segmented.

[0016] Furthermore, step 2 involves obtaining the pinyin and normalizing it using fuzzy rules: obtaining the pinyin from the candidate substrings and domain vocabulary entries, and applying fuzzy rules to obtain normalized pinyin, as detailed below:

[0017] For any candidate substring s∈S or a word w∈W in the domain lexicon W, after obtaining the pinyin, apply the fuzzy rule F to obtain the normalized pinyin P(x), where x=s or w. When the lexicon provides a polyphonic character with a specified pronunciation P0(x), the criterion of P(x) = F(pinyin(x)), or P(x) = F(P0(x)) if P0(x) provided is preferred.

[0018] Furthermore, the fuzzy rule F described in step 2 covers common near-sound phenomena:

[0019] F: { z->zh, c->ch, s->sh, en->eng, in->ing}.

[0021] Furthermore, the two-stage matching and dual-threshold determination described in step 3: the first stage calculates sequence similarity and the second stage normalized edit similarity, and the error correction mapping is generated through joint determination of dual thresholds, as detailed below:

[0022] Step 3.1: Calculate the similarity r between sequences P(s) and P(w), where r ∈ [0,1]:

[0023] r = R(P(s), P(w)), r in [0,1]

[0024] Step 3.2: Use Levenshtein-based normalized edit similarity l, where D represents the Levenshtein edit distance, calculated as follows:

[0025] l = 1 - D(P(s), P(w)) / max( |P(s)|, |P(w)| )

[0026] Step 3.3: Generate the error correction mapping M[s] = w through joint determination of dual thresholds:

[0027] r >τ1 and l >τ2 => M[s] = w

[0028] Among them, τ1 is used to control the coarse screen width, and τ2 is used to control the verification intensity. The parameters are adjusted according to the noise intensity, accent complexity and vocabulary size, and 0<τ1<τ2<1.

[0029] Furthermore, in step 3, a length consistency filter is introduced before the first stage. When the dictionary entry is in Chinese and |s|≠|w|, the comparison is skipped; when the entry is in English, the restriction is relaxed.

[0030] not is_all_english(w) and |s| < > |w| => skip

[0031] is_all_english(w) is a function that determines whether the string w consists entirely of English letters.

[0032] The expression `not is_all_english(w)` means that the letter 'w' is not entirely composed of English letters.

[0033] |s|| and |w| represent the lengths of strings s and w, respectively;

[0034] < > indicates inequality, that is, |s| is not equal to |w|;

[0035] "and" indicates that both conditions must be met;

[0036] => skip means skipping the current processing if the condition is true;

[0037] The entire expression means: if w is not all English letters and the length of s is not equal to the length of w, then skip the current processing.

[0038] Furthermore, the safe replacement described in step 4 involves replacing segments of the original text according to the error correction map, controlling the positional occupancy set to avoid overlapping and cascading erroneous replacements, and outputting the error-corrected text, as detailed below:

[0039] After traversing the candidate set S, a mapping M is obtained. The final replacement uses the position-occupying set Ω to avoid overlapping replacements and cascading erroneous replacements.

[0040] Omega subset {0,1,...,|T|-1}, J=[a,b] replaceable iff J ∩ Omega =empty

[0041] The replacement process follows the original index order. Candidates are first located by the first and last characters. After removing common punctuation marks, if they are exactly equal to the key 's', they are replaced with 'w', and the replacement interval is merged into 'Ω'. If there is an overlap in intervals, they are skipped to ensure the structural stability and readability of the output text.

[0042] A speech recognition system for radar terminology based on dual similarity matching technology is disclosed. This system implements the aforementioned speech recognition method for radar terminology based on dual similarity matching technology. The system includes a text segmentation module, a pinyin acquisition and fuzzy rule normalization module, a two-stage matching and dual threshold determination module, and a secure replacement module.

[0043] Text segmentation module: After removing punctuation from long Chinese text obtained from speech recognition, a sliding window is used to generate a set of candidate substrings;

[0044] Pinyin Acquisition and Fuzzy Rule Normalization Module: Acquires pinyin from candidate substrings and domain lexicon entries, and applies fuzzy rules to obtain normalized pinyin;

[0045] Two-stage matching and dual-threshold determination module: The first stage calculates sequence similarity and the second stage normalizes edit similarity, and the dual thresholds are used to jointly determine and generate error correction mapping;

[0046] The safe replacement module replaces segments of the original text according to the error correction mapping, based on the positional occupancy set control, avoiding overlapping replacements and cascading erroneous replacements, and outputs the error correction text.

[0047] An electronic device includes a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to perform the radar terminology speech recognition method based on dual similarity matching technology.

[0048] Compared with the prior art, the present invention has the following significant advantages: (1) It is oriented towards the characteristics of Chinese and introduces fuzzy rules of pinyin to cover near-sound phenomena such as retroflex consonants and nasal consonants, thereby improving the accuracy of error correction; (2) It reduces the error replacement rate by using a two-stage verification of "sequence similarity + distance deviation" and a joint judgment of dual thresholds; (3) It supports the dictionary to provide fixed pronunciations for polyphonic characters, thereby improving the stability of professional terminology matching; (4) It introduces a safe replacement strategy of position occupancy set to suppress overlapping replacement and cascading pollution, thereby ensuring the readability and consistency of the output; (5) The algorithm is simple to implement, has low hardware requirements, does not rely on massive labeled corpora, and is easy to implement and deploy in engineering. Attached Figure Description

[0049] Figure 1 This is a flowchart of the radar terminology speech recognition method based on dual similarity matching technology of the present invention. Detailed Implementation

[0050] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0051] This invention provides a method for speech recognition and error correction of radar terminology based on dual similarity matching technology. The method takes a long text output by ASR (Automatic Speech Recognition) as input, segments it into 2-6 character pieces to form candidates, first performs pinyin normalization and fuzzy rule processing, then performs coarse screening based on sequence similarity, and then performs a second verification using Levenshtein distance deviation method. Error correction mapping is generated through dual threshold joint judgment, and finally, a safe replacement based on positional occupancy set constraints is used to obtain the corrected text, achieving accurate error correction for radar terminology and slang. This method is simple to implement, has low requirements for processing equipment, achieves good text correction results after speech recognition in the radar field, does not rely on massive labeled corpora, and is easy to implement and deploy in engineering.

[0052] like Figure 1 As shown, this invention provides a speech recognition method for radar terminology based on dual similarity matching technology. Combining pronunciation normalization and fuzzy rules, it employs a two-stage joint determination using sequence similarity and distance deviation thresholds to achieve accurate error correction of radar terminology and slang. The method includes the following steps:

[0053] Step 1, Text Segmentation: After removing punctuation from the long Chinese text obtained from speech recognition, a sliding window is used to generate a set of candidate substrings, as follows:

[0054] Punctuation removal and whitespace normalization are performed on the input long text T. Then, a candidate set S is generated using a sliding window of n∈{2,3,4,5,6}, covering short near-homophones and common terminology fragments.

[0055] S = { T[i:i+n] | 0 ≤i < |T|, n in {2,3,4,5,6}, i+n ≤|T|}

[0056] As a specific example, in step 1, Arabic numerals and unit symbols are protected before text segmentation to prevent them from being mis-segmented.

[0057] Step 2, Pinyin Acquisition and Fuzzy Rule Normalization: Pinyin is acquired from candidate substrings and domain terminology entries, and fuzzy rules are applied to obtain normalized Pinyin, as detailed below:

[0058] For any candidate substring s∈S or a word w∈W in the domain lexicon W, after obtaining the pinyin, apply the fuzzy rule F to obtain the normalized pinyin P(x), where x=s or w. When the lexicon provides a polyphonic character with a specified pronunciation P0(x), the criterion of P(x) = F(pinyin(x)), or P(x) = F(P0(x)) if P0(x) provided is preferred.

[0059] The fuzzy rule F described in step 2 covers common near-sound phenomena:

[0060] F: { z->zh, c->ch, s->sh, en->eng, in->ing}

[0061] Step 3, Two-stage matching and dual-threshold determination: The first stage calculates sequence similarity and the second stage normalizes edit similarity. Error correction mappings are generated through joint determination using dual thresholds, as detailed below:

[0062] Step 3.1: Calculate the similarity r between sequences P(s) and P(w), where r ∈ [0,1]:

[0063] r = R(P(s), P(w)), r in [0,1]

[0064] Step 3.2: Use Levenshtein-based normalized edit similarity l, where D represents the Levenshtein edit distance, calculated as follows:

[0065] l = 1 - D(P(s), P(w)) / max( |P(s)|, |P(w)| )

[0066] Step 3.3: Generate the error correction mapping M[s] = w through joint determination of dual thresholds:

[0067] r >τ1 and l >τ2 => M[s] = w

[0068] Among them, τ1 is used to control the coarse screen width, and τ2 is used to control the verification intensity. The parameters are adjusted according to the noise intensity, accent complexity and vocabulary size, and 0<τ1<τ2<1.

[0069] In step 3, a length consistency filter is introduced before the first stage. When the dictionary entry is in Chinese and |s|≠|w|, the comparison is skipped; when the entry is in English, the restriction is relaxed.

[0070] not is_all_english(w) and |s| < > |w| => skip

[0071] is_all_english(w) is a function that determines whether the string w consists entirely of English letters.

[0072] The expression `not is_all_english(w)` means that the letter 'w' is not entirely composed of English letters.

[0073] |s|| and |w| represent the lengths of strings s and w, respectively;

[0074] < > indicates inequality, that is, |s| is not equal to |w|;

[0075] "and" indicates that both conditions must be met;

[0076] => skip means skipping the current processing if the condition is true;

[0077] The entire expression means: if w is not all English letters and the length of s is not equal to the length of w, then skip the current processing.

[0078] Step 4, Safe Replacement: Based on the error correction map, perform segment replacement on the original text according to the positional occupancy set control to avoid overlapping replacement and cascading erroneous replacement, and output the error correction text, as follows:

[0079] After traversing the candidate set S, a mapping M is obtained. The final replacement uses the position-occupying set Ω to avoid overlapping replacements and cascading erroneous replacements.

[0080] Omega subset {0,1,...,|T|-1}, J=[a,b] replaceable iff J ∩ Omega =empty

[0081] The replacement process follows the original index order. Candidates are first located by the first and last characters. After removing common punctuation marks, if they are exactly equal to the key 's', they are replaced with 'w', and the replacement interval is merged into 'Ω'. If there is an overlap in intervals, they are skipped to ensure the structural stability and readability of the output text.

[0082] The above steps provide a relatively simple way to implement this method. It is easy to implement, has low requirements for processing equipment, and can achieve good text correction results after speech recognition in the radar field. It does not rely on massive labeled corpora, is easy to implement and deploy in engineering, and can be widely used in various radar systems.

[0083] This invention also provides a radar terminology speech recognition system based on dual similarity matching technology. This system implements the aforementioned radar terminology speech recognition method based on dual similarity matching technology. The system includes a text segmentation module, a pinyin acquisition and fuzzy rule normalization module, a two-stage matching and dual threshold determination module, and a secure replacement module.

[0084] Text segmentation module: After removing punctuation from long Chinese text obtained from speech recognition, a sliding window is used to generate a set of candidate substrings;

[0085] Pinyin Acquisition and Fuzzy Rule Normalization Module: Acquires pinyin from candidate substrings and domain lexicon entries, and applies fuzzy rules to obtain normalized pinyin;

[0086] Two-stage matching and dual-threshold determination module: The first stage calculates sequence similarity and the second stage normalizes edit similarity, and the dual thresholds are used to jointly determine and generate error correction mapping;

[0087] The safe replacement module replaces segments of the original text according to the error correction mapping, based on the positional occupancy set control, avoiding overlapping replacements and cascading erroneous replacements, and outputs the error correction text.

[0088] The present invention also provides an electronic device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the radar terminology speech recognition method based on dual similarity matching technology.

[0089] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the electronic device.

[0090] The electronic device may be a desktop computer, laptop, handheld computer, or smart tablet, etc. The electronic device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above components are merely examples of electronic devices and do not constitute a limitation on the electronic device. It may include more or fewer components than described above, or combine certain components, or different components. For example, the electronic device may also include input / output devices, network access devices, buses, etc.

[0091] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the electronic device, connecting all parts of the electronic device via various interfaces and lines.

[0092] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0093] Wherein, if the modules / units integrated in the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0094] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and 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 modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0095] In summary, this invention constructs a high-precision speech recognition error correction method that takes "pronunciation normalization" as the first principle, combines Chinese fuzzy pronunciation rules with a two-stage verification mechanism of "sequence similarity + edit distance deviation", robustly generates "error → correct" mapping under dual threshold control, and suppresses overlapping replacement and secondary pollution through a safe replacement strategy of position occupancy set, thus meeting the engineering needs of high-precision error correction for radar terminology.

[0096] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A speech recognition method for radar terminology based on dual similarity matching technology, characterized in that, Includes the following steps: Step 1, Text Segmentation: After removing punctuation from the long Chinese text obtained from speech recognition, a sliding window is used to generate a set of candidate substrings; Step 2, Pinyin Acquisition and Fuzzy Rule Normalization: Obtain the pinyin from the candidate substrings and domain vocabulary entries, and apply fuzzy rules to obtain normalized pinyin; Step 3, Two-stage matching and dual-threshold determination: The first stage calculates sequence similarity and the second stage normalized edit similarity, and the error correction map is generated by joint determination of dual thresholds; Step 4, Safe Replacement: Based on the error correction mapping, replace segments of the original text according to the positional occupancy set control to avoid overlapping replacement and cascading erroneous replacement, and output the error correction text.

2. The radar terminology speech recognition method based on dual similarity matching technology according to claim 1, characterized in that, The text segmentation in step 1 involves removing punctuation from the long Chinese text obtained from speech recognition and then generating a candidate substring set using a sliding window, as detailed below: Punctuation removal and whitespace normalization are performed on the input long text T. Then, a candidate set S is generated using a sliding window of n∈{2,3,4,5,6}, covering short near-homophones and common terminology fragments. S = { T[i:i+n] | 0 ≤i < |T|, n in {2,3,4,5,6}, i+n ≤|T|}.

3. The radar terminology speech recognition method based on dual similarity matching technology according to claim 2, characterized in that, In step 1, Arabic numerals and unit symbols are protected before text segmentation to prevent them from being mis-segmented.

4. The radar terminology speech recognition method based on dual similarity matching technology according to claim 3, characterized in that, Step 2, Pinyin Acquisition and Fuzzy Rule Normalization: Pinyin is acquired from candidate substrings and domain terminology entries, and fuzzy rules are applied to obtain normalized Pinyin, as detailed below: For any candidate substring s∈S or a word w∈W in the domain lexicon W, after obtaining the pinyin, apply the fuzzy rule F to obtain the normalized pinyin P(x), where x=s or w. When the lexicon provides a polyphonic character with a specified pronunciation P0(x), the criterion of P(x) = F(pinyin(x)), or P(x) = F(P0(x)) if P0(x) provided is preferred.

5. The radar terminology speech recognition method based on dual similarity matching technology according to claim 4, characterized in that, The fuzzy rule F described in step 2 covers common near-sound phenomena: F: { z->zh, c->ch, s->sh, en->eng, in->ing}.

6. The radar terminology speech recognition method based on dual similarity matching technology according to claim 1, characterized in that, Step 3 involves two-stage matching and dual-threshold determination: the first stage calculates sequence similarity and the second stage normalized edit similarity, and the error correction mapping is generated through joint determination of dual thresholds, as detailed below: Step 3.1: Calculate the similarity r between sequences P(s) and P(w), where r ∈ [0,1]: r = R(P(s), P(w)), r in [0,1] Step 3.2: Use Levenshtein-based normalized edit similarity l, where D represents the Levenshtein edit distance, calculated as follows: l = 1 - D(P(s), P(w)) / max( |P(s)|, |P(w)| ) Step 3.3: Generate the error correction mapping M[s] = w through joint determination of dual thresholds: r >τ1 and l >τ2 => M[s] = w Among them, τ1 is used to control the coarse screen width, and τ2 is used to control the verification intensity. The parameters are adjusted according to the noise intensity, accent complexity and vocabulary size, and 0<τ1<τ2<1.

7. The radar terminology speech recognition method based on dual similarity matching technology according to claim 6, characterized in that, In step 3, a length consistency filter is introduced before the first stage. When the dictionary entry is in Chinese and |s|≠|w|, the comparison is skipped; when the entry is in English, the restriction is relaxed. not is_all_english(w) and |s| < > |w| => skip; is_all_english(w) is a function that determines whether the string w consists entirely of English letters. The expression `not is_all_english(w)` means that the letter 'w' is not entirely composed of English letters. |s|| and |w| represent the lengths of strings s and w, respectively; < > indicates inequality, that is, |s| is not equal to |w|; "and" indicates that both conditions must be met; => skip means skipping the current processing if the condition is true; The entire expression means: if w is not all English letters and the length of s is not equal to the length of w, then skip the current processing.

8. The radar terminology speech recognition method based on dual similarity matching technology according to claim 7, characterized in that, Step 4 describes safe replacement: based on the error correction map, the original text is replaced with segments controlled by positional occupancy sets to avoid overlapping replacements and cascading erroneous replacements, and the error-corrected text is output as follows: After traversing the candidate set S, a mapping M is obtained. The final replacement uses the position-occupying set Ω to avoid overlapping replacements and cascading erroneous replacements. Omega subset {0,1,...,|T|-1}, J=[a,b] replaceable iff J ∩ Omega = empty The replacement process follows the original index order. Candidates are first located by the first and last characters. After removing common punctuation marks, if they are exactly equal to the key 's', they are replaced with 'w', and the replacement interval is merged into 'Ω'. If there is an overlap in intervals, they are skipped to ensure the structural stability and readability of the output text.

9. A speech recognition system for radar terminology based on dual similarity matching technology, characterized in that, This system is used to implement the radar terminology speech recognition method based on dual similarity matching technology as described in any one of claims 1 to 8. The system includes a text segmentation module, a pinyin acquisition and fuzzy rule normalization module, a two-stage matching and dual threshold determination module, and a secure replacement module. Text segmentation module: After removing punctuation from long Chinese text obtained from speech recognition, a sliding window is used to generate a set of candidate substrings; Pinyin Acquisition and Fuzzy Rule Normalization Module: Acquires pinyin from candidate substrings and domain lexicon entries, and applies fuzzy rules to obtain normalized pinyin; Two-stage matching and dual-threshold determination module: The first stage calculates sequence similarity and the second stage normalizes edit similarity, and the dual thresholds are used to jointly determine and generate error correction mapping; The safe replacement module replaces segments of the original text according to the error correction mapping, based on the positional occupancy set control, avoiding overlapping replacements and cascading erroneous replacements, and outputs the error correction text.

10. An electronic device, characterized in that, include: The system includes a memory and a processor, which are interconnected. The memory stores computer instructions, and the processor executes the computer instructions to perform the radar terminology speech recognition method based on dual similarity matching technology as described in any one of claims 1 to 8.

Images