Sample preparation system for speech identity identification device and identification capability evaluation method

Abstract

The present invention relates to the technical field of judicial identification instrument evaluation, and specifically to a sample preparation system and an identification capability evaluation method for a voice identity identification device, comprising: a parameter generation module, the parameter generation module generates a target parameter according to an evaluation requirement; a voice generation module, the voice generation module is connected to the parameter generation module, and the sample generation module generates voice data according to the target parameter; and a sample output module, the sample output module is connected to the voice generation module, and the sample output module receives the voice data and generates an experimental sample before outputting it. The beneficial effect of the present invention is that by adding a parameter generation module to the sample preparation system to parse the evaluation requirement, and based on the target parameter obtained by the parsing, sequentially constructing voice data, processing the voice data and then outputting it, the experimental sample can better meet the evaluation requirements of the voice identity identification device, and can better evaluate the identification capability of the voice identity identification device.

Description

Technical Field

[0001] The present invention relates to the technical field of judicial identification instrument evaluation, and in particular to a sample preparation system of a speech identity identification device and an identification capability evaluation method. Background Art

[0002] Voice identity forensic identification refers to a specialized technology that identifies and judges the identity of voices recorded in audio-visual materials through comparison and analysis. With the development of computer technology, voice identity identification methods have developed to the point where voiceprint analysis and feature extraction of recorded voice samples are performed through computer technology for comparative identification. However, due to the limitations of actual hardware conditions, the actual identification results may have certain errors due to the error factors of the identification device itself. Therefore, before conducting voice identity identification, it is necessary to verify the capability of the identification device. Voice identity capability verification is a method used to evaluate the capability of laboratories that implement voice identity identification. It is usually carried out in the form of "participating institutions register - organizers issue operating guidelines and samples to be tested - participating institutions feedback test results - organizers evaluate".

[0003] In the prior art, samples used to evaluate speech identity identification devices are usually prepared based on the following process: proficiency verification program experts determine the key points of the inspection, technical experts prepare samples according to the plan and verify them, add unique identifiers and copy and distribute them, thereby obtaining samples that can be used to evaluate the identification capabilities of speech identity identification devices.

[0004] However, during the actual implementation process, the inventors found that the prior art has the following problems:

[0005] (1) Random sample preparation process: There is no unified standard and estimate for the content and inspection points of proficiency testing samples, lack of standardized production process, and unpredictable evaluation results;

[0006] (2) Poor reusability of sample preparation content: Each sample preparation can only be used once, and parallel samples prepared during the sample preparation process cannot be utilized;

[0007] (3) Large workload: Each time a sample is prepared, from design to delivery, a lot of creative work is required from professional and experienced preparation technicians. The process is cumbersome, and trial and error and rework also take up a lot of working time. Summary of the invention

[0008] In view of the above problems existing in the prior art, a sample preparation system for a speech identity identification device is provided; on the other hand, a method for evaluating the identification capability of a speech identity identification device is also provided.

[0009] The specific technical solutions are as follows:

[0010] A sample preparation system for a speech identity identification device, comprising:

[0011] A parameter generation module, wherein the parameter generation module generates a target parameter according to an evaluation requirement;

[0012] A speech generation module, the speech generation module is connected to the parameter generation module, and the sample generation module generates speech data according to the target parameters;

[0013] A sample output module, the sample output module is connected to the speech generation module, and the sample output module receives the speech data and generates experimental samples for output.

[0014] Preferably, the parameter generation module includes:

[0015] A difficulty control submodule, wherein the difficulty control submodule generates a difficulty parameter according to the evaluation requirement and adds the difficulty parameter to the target parameter;

[0016] A sample type setting submodule generates a sample type parameter according to the evaluation requirement and adds the sample type parameter to the target parameter.

[0017] Preferably, the difficulty parameters include: the number of inspection points, the types of inspection points and the inspection point selection method;

[0018] The sample type parameters include recording source, recording mode and recording parameters.

[0019] Preferably, the speech generation module comprises:

[0020] A sample collection submodule, which obtains recording samples and stores them in a sample database;

[0021] A screening submodule, wherein the screening submodule retrieves available speech from the sample database according to the difficulty parameter and the sample type parameter;

[0022] A checking submodule, wherein the checking submodule is connected to the screening submodule, and the checking submodule checks the available voice according to the evaluation requirement to generate the voice data.

[0023] Preferably, the inspection submodule includes:

[0024] a matching degree checking submodule, wherein the matching degree checking submodule parses the evaluation requirement to obtain an expected sample of inspection points of the experimental sample required by the evaluation requirement, and evaluates the number and type of inspection points in the speech data based on the expected sample of inspection points to generate a first inspection result;

[0025] An inspection point association checking submodule, wherein the inspection point association checking submodule parses the evaluation requirement to generate an inspection point distribution of the experimental sample required by the evaluation requirement, and evaluates the distribution of the inspection points in the speech data based on the inspection point distribution to generate a second inspection result;

[0026] A logic analysis submodule, wherein the logic analysis submodule parses the evaluation requirement to generate brief case information of the experimental sample required by the evaluation requirement, and performs a logic correlation analysis on the voice content of the voice data based on the brief case information to generate a third inspection result;

[0027] A quality evaluation submodule, wherein the quality evaluation submodule is respectively connected to the matching degree inspection submodule, the inspection point association inspection submodule and the logic analysis submodule, and the quality evaluation submodule eliminates the available voices that do not meet the evaluation requirements according to the first inspection result, the second inspection result and the third inspection result.

[0028] Preferably, the quality assessment submodule performs a weighted summation on the first inspection result, the second inspection result and the third inspection result, and compares the calculated value with a preset inspection threshold to eliminate the voice data that does not meet the assessment requirements;

[0029] The quality evaluation submodule further compares the third inspection result with a preset logic inspection threshold to eliminate the speech data that does not conform to the logic relationship.

[0030] Preferably, the sample output module comprises:

[0031] An anti-collusion submodule, the anti-collusion submodule is connected to the voice generation module, the anti-collusion submodule receives the voice data, and adds a unique identifier to each group of the voice data to generate a unique voice;

[0032] A replication submodule, wherein the replication submodule is connected to the anti-collusion submodule, and the replication submodule replicates the unique voice according to the evaluation requirements and outputs it as the experimental sample.

[0033] Preferably, the sample output module further includes:

[0034] A check submodule, wherein the check submodule is connected to the replication submodule, and the check submodule respectively obtains the unique speech and the experimental sample and calculates a hash value for verifying the consistency of the unique speech and the experimental sample.

[0035] A method for evaluating the identification ability of a speech identity identification device, using the experimental samples generated by the sample preparation system, comprises:

[0036] Step S1: generating an evaluation requirement corresponding to the speech identity identification device for the speech identity identification device;

[0037] Step S2: generating the experimental sample based on the evaluation requirement;

[0038] Step S3: using the experimental samples to evaluate the identification capability of the speech identity identification device.

[0039] The above technical solution has the following advantages or beneficial effects: by adding a parameter generation module to the sample preparation system to parse the evaluation requirements, and constructing voice data in sequence based on the target parameters obtained by the analysis, and processing the voice data and then outputting it, the experimental samples can better meet the evaluation requirements of the voice identity identification device, and can better evaluate the identification ability of the voice identity identification device; at the same time, by setting up a sample output module to process and output the voice data, better sample reusability is achieved, and in this process, the influence of human factors is eliminated, thereby improving the quality of sample preparation. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] The embodiments of the present invention will be described more fully with reference to the attached drawings, which are provided for illustration and description only and are not intended to limit the scope of the present invention.

[0041] Figure 1 It is a principle block diagram of a sample preparation system in an embodiment of the present invention;

[0042] Figure 2 This is a schematic diagram of the inspection submodule in the embodiment of this aspect;

[0043] Figure 3 Schematic diagram of the identification capability evaluation method in the embodiment of this aspect. DETAILED DESCRIPTION

[0044] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0045] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0046] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments, but they are not intended to limit the present invention.

[0047] The present invention comprises:

[0048] A sample preparation system for a speech identity identification device, such as Figure 1 As shown, including:

[0049] Parameter generation module 1, parameter generation module 1 generates target parameters according to an evaluation requirement;

[0050] The speech generation module 2 is connected to the parameter generation module 1, and the sample generation module 2 generates speech data according to the target parameters;

[0051] The sample output module 3 is connected to the speech generation module 2. The sample output module 3 receives speech data and generates experimental samples for output.

[0052] Specifically, in the prior art, the evaluation process of the identification ability of the speech identity identification device usually adopts the method of manually preparing samples, and there is no specific operation process, which leads to the problem that the sample preparation process is very arbitrary and the quality is unstable due to the influence of artificial factors. In this embodiment, by constructing the above-mentioned sample preparation system and setting the parameter generation module 1 in the sample preparation system, the parameter generation module 1 can generate the corresponding target parameters according to the evaluation requirements; then, the speech generation module 2 is used to collect and screen the speech samples based on the target parameters, so as to obtain the speech data that meets the evaluation requirements. In this process, by setting the target parameters, a better selection of speech data of different types and contents is achieved, so that a more comprehensive evaluation effect can be achieved for the speech identity identification device. At the same time, by adjusting the target parameters, it is also easier to control the difficulty, parameters, etc. of the generated speech data, so that the experimental samples are more in line with the evaluation requirements. Finally, the speech data is copied and verified by the sample output module 3 and output as the experimental sample, so as to achieve better reusability of the speech data and ensure better consistency of the experimental samples of the same batch.

[0053] During implementation, the above-mentioned sample preparation system is usually set up in a computer system as a software embodiment, which is used to parse the evaluation requirements to realize the collection and processing process of voice samples, and then generate experimental samples that can be used to evaluate the identification ability of the voice identity identification device. The experimental sample refers to the sample required in the process of evaluating the identification ability of the voice identity identification device. It is embodied as a digital audio file with specific audio features, and in order to achieve the uniqueness of the experimental sample, it is also added with a unique identification string for identifying different samples. The evaluation requirements are the key points of the inspection that need to be inspected based on the voice identity identification device, which are embodied in different auditory features, spectral features, feature stability, recording methods, etc. The target parameter is an "expected parameter" for the experimental sample determined based on the evaluation requirements, which is used to judge the acquired voice sample in the process of extracting voice data, so as to obtain the actual required voice data.

[0054] In a preferred embodiment, the parameter generation module 1 includes:

[0055] A difficulty control submodule 11, which generates a difficulty parameter according to the evaluation requirements and adds the difficulty parameter to the target parameter;

[0056] The channel parameter setting submodule 12 generates channel parameters according to the evaluation requirements and adds the channel parameters to the target parameters.

[0057] Specifically, in view of the problem that the samples used for evaluating the identification ability of the speech identity identification device in the prior art are mainly manually selected samples, which are highly subjective and cannot cover the evaluation requirements well, in this embodiment, by respectively setting the difficulty control submodule 11 and the channel parameter setting submodule 12, the evaluation requirements are analyzed to obtain the difficulty parameters and the channel parameters, and the subsequent speech data generation process is guided based on the parameters of these two dimensions, so as to achieve a better experimental sample preparation effect and a better control effect on the desired generated speech data.

[0058] In a preferred embodiment, the difficulty parameters include: the number of inspection points, the types of inspection points and the inspection point selection method;

[0059] The channel parameters include at least one of a recording source, a recording mode and a recording parameter.

[0060] In the implementation process, the difficulty parameter refers to the expected difficulty of the assessment of the identification ability of the speech identity identification device determined based on the key points of the assessment, which is reflected by the number of available assessment points of the speech identity identification device; the type of assessment point, that is, the available inspection angles during the inspection; and the selection method of the assessment point, which is reflected by the number of auditory features and spectral features corresponding to the assessment points. Specifically, the inspection angle refers to the specific conforming features and difference features that can be extracted from the entire audio file during the inspection process, including "dialect accent", "pause rhythm", "wrong pronunciation", "speech defect", "timbre", "formant frequency", etc. The actual available inspection angle depends on the case content corresponding to the evaluation requirements and the indicators of the speech identity identification device itself. The selection method of the assessment point refers to the coverage of the assessment point on different speech features, such as whether the assessment points are evenly distributed in the entire speech content, whether the assessment points completely cover all speech features, etc. By changing the above parameters, the speech identity identification device can use different experimental samples according to different degrees of difficulty during the assessment of the identification ability. Channel parameters refer to the source type information of different voice samples, which are used to characterize different audio files, including recording sources, specifically telephone recording, video recording, instant messaging software recording, and direct recording of human voice using recording equipment; recording method refers to the method of directly obtaining the audio file, such as extracting software cache, secondary recording using recording equipment, reading through software interface, etc.; recording parameters refer to the basic audio information used to characterize the audio file, including channel parameters, sampling rate, number of bits, etc.

[0061] In a preferred embodiment, the speech generation module 2 includes:

[0062] The sample collection submodule 21 obtains the recording sample and stores it in the sample database;

[0063] A screening submodule 22, the screening submodule 22 retrieves available speech from a sample database according to target parameters;

[0064] The checking submodule 23 is connected to the screening submodule 22 , and the checking submodule 23 checks the available voice according to the evaluation requirements to generate voice data.

[0065] Specifically, in view of the problem that the samples used for evaluating the identification ability of the speech identity identification device in the prior art are mainly manually selected samples, which are highly subjective and cannot well cover the evaluation requirements, in this embodiment, a plurality of different recording samples are pre-acquired and stored in the sample database by setting the sample collection submodule 21. Subsequently, the screening submodule 22 performs field screening from the sample database according to the above-mentioned target parameters, so as to find the available speech that meets the requirements, and the inspection submodule 23 inspects the extracted set of available speech, so as to output the available speech that meets the evaluation requirements as speech data, so as to achieve a better speech data generation effect.

[0066] In a preferred embodiment, the inspection submodule 23 includes:

[0067] The matching degree checking submodule 231 analyzes the evaluation requirements to obtain the expected sample of inspection points of the experimental samples required by the evaluation requirements, and evaluates the number and type of inspection points in the speech data based on the expected sample of inspection points to generate a first inspection result;

[0068] During the implementation process, the expected sample of inspection points is the expected score determined based on the evaluation requirements to characterize the coverage of the inspection points; the matching degree check submodule 231 obtains several marked inspection points from the voice data, and matches the inspection point type of each inspection point respectively, obtains the inspection point weight corresponding to each inspection point, performs weighted calculation based on the inspection point weight and the number of inspection points, compares the calculation result with the expected score, and obtains the first inspection result, so as to judge whether the coverage of the inspection points meets expectations.

[0069] During the implementation process, in order to achieve a better evaluation effect on the speech identity identification device, this embodiment analyzes the evaluation requirements to obtain the inspection points of the sample that the speech identity identification device needs to identify in the actual identification process as the expected sample of inspection points, and evaluates the number and type of inspection points in the speech data to examine the degree of matching between the inspection points of the sample and the sample.

[0070] An inspection point association checking submodule 232, the inspection point association checking submodule 232 analyzes the evaluation requirements to generate an inspection point distribution of the experimental sample required for the evaluation requirements, and evaluates the distribution of the inspection points in the speech data based on the inspection point distribution to generate a second inspection result;

[0071] During the implementation process, the distribution of inspection points includes: the distribution of inspection points on different speech features in the entire speech data. In the pre-set difficulty parameter setting process, the inspection angle and selection method of the inspection points have been pre-extracted. Based on these two points, the speech features in the audio data can be extracted to obtain the distribution of speech features in the audio data, and the distribution of speech features can be compared with the inspection points to determine whether the inspection points effectively cover all speech features required in the identification process, and whether the number of inspection points selected for the same speech feature is sufficient.

[0072] The logic analysis submodule 233 analyzes the evaluation requirements to generate brief case information of the experimental samples required for the evaluation requirements, and performs a logic correlation analysis on the voice content of the voice data based on the brief case information to generate a third inspection result;

[0073] The quality evaluation submodule 234 is connected to the matching degree inspection submodule 231, the inspection point association inspection submodule 232 and the logic analysis submodule 233 respectively. The quality evaluation submodule 234 removes the available speech that does not meet the evaluation requirements according to the first inspection result, the second inspection result and the third inspection result.

[0074] Specifically, in view of the problem that the samples used for evaluating the identification ability of the speech identity identification device in the prior art are mainly manually selected samples, which are highly subjective and cannot cover the evaluation requirements well, in this embodiment, by setting the inspection submodule 23 and performing further inspection based on the number of inspection points, the distribution of inspection points and the logical relationship in the evaluation requirements, it is determined whether the available speech meets the evaluation requirements, so as to eliminate the available speech that does not meet the evaluation requirements and generate speech data.

[0075] During the implementation process, when the assessment requirements are generated, brief case information in the work guide will be extracted. By pre-training an NLP recognition model to process the brief case information and the voice content obtained through recognition, the first semantic vector of the brief case information and the second semantic vector of the voice content are obtained respectively. By comparing the first semantic vector and the second semantic vector, the quantitative calculation of the logical relationship between the two can be realized.

[0076] In a preferred embodiment, the quality evaluation submodule 234 performs a weighted summation on the first inspection result, the second inspection result, and the third inspection result, and compares the calculated value with a preset inspection threshold to eliminate available speech that does not meet the evaluation requirements;

[0077] The quality evaluation submodule 234 further compares the third inspection result with a preset logic inspection threshold to eliminate usable speech that does not conform to the logical relationship.

[0078] Specifically, in order to achieve a better inspection effect on the available voice, in this embodiment, the first inspection result, the second inspection result and the third inspection result are quantified, and then different weights are set respectively and weighted summation is performed to achieve the scoring of the available voice, and the inspection threshold is set to eliminate the available voice with too low score to achieve a better screening effect. At the same time, a logical inspection threshold can be further set to determine whether there is a strong logical correlation between the available voice and the brief case information. In the implementation process, it can be reflected in calculating the angle between the first semantic vector and the second semantic vector, and setting a logical inspection threshold for the angle. When the angle is greater than the logical inspection threshold, it indicates that the logical relationship between the two is weak and is not suitable for output as voice data.

[0079] In the implementation process, the first inspection result, the second inspection result and the third inspection result can also be set with different scoring ranges, and each scoring range corresponds to a different score value. For example, the first inspection result can be divided into: more, more, less, less, and less, and quantitative processing is achieved by setting different scores for each level.

[0080] In a preferred embodiment, the sample output module 3 includes:

[0081] The anti-collusion submodule 31 is connected to the voice generation module 2. The anti-collusion submodule 31 receives voice data and adds a unique identifier to each group of voice data to generate a unique voice.

[0082] The replication submodule 32 is connected to the anti-collusion submodule 31 . The replication submodule 32 replicates the unique voice according to the evaluation requirements and outputs it as an experimental sample.

[0083] Specifically, in view of the problem of poor sample reusability during the preparation of experimental samples used to evaluate speech identity identification devices in the prior art, in this embodiment, an anti-collusion submodule 31 and a replication submodule 32 are sequentially added to the sample output module 3, and the anti-collusion submodule 31 adds a unique identifier to each group of voice data received, thereby achieving the uniqueness of different experimental samples, and the replication submodule 32 replicates the unique voice with the unique identifier and outputs it as an experimental sample, thereby allowing the same group of experimental samples to be reused in multiple evaluation processes.

[0084] In a preferred embodiment, the sample output module 3 further includes:

[0085] The check submodule 33 is connected to the replication submodule 32. The check submodule 33 obtains the unique speech and the experimental sample respectively and calculates the hash value, which is used to verify the consistency of the unique speech and the experimental sample.

[0086] Specifically, in order to address the problem that errors may occur in the copying process of the experimental samples, thereby reducing the accuracy of the samples, in this embodiment, a verification submodule 33 is set after the copying submodule 32 to perform a hash value verification on the unique speech generated after the copying, so that the experimental samples and the unique speech have better consistency.

[0087] In one embodiment, a SHA256 hash value is used to verify the consistency of the unique speech and the experimental sample.

[0088] A method for evaluating the identification ability of a speech identity identification device, using the experimental samples generated by the sample preparation system, such as Figure 3 As shown, including:

[0089] Step S1: generating an evaluation requirement corresponding to the speech identity identification device for the speech identity identification device;

[0090] Step S2: Generate experimental samples based on evaluation requirements;

[0091] Step S3: using experimental samples to evaluate the identification capability of the speech identity identification device.

[0092] Specifically, in view of the problem of poor accuracy of samples used in the evaluation process of the identification capability of the speech identity identification device in the prior art, in this embodiment, the operating guide required for the evaluation process is determined for the speech identity identification device to be evaluated, and evaluation requirements are generated based on the operating guide. Subsequently, the above-mentioned sample preparation system is used to prepare experimental samples based on the evaluation requirements, thereby achieving better evaluation results.

[0093] The beneficial effects of the present invention are as follows: by adding a parameter generation module to the sample preparation system to analyze the evaluation requirements, and constructing speech data in sequence based on the target parameters obtained by the analysis, and processing the speech data and then outputting it, the experimental samples can better meet the evaluation requirements of the speech identity identification device, and can better evaluate the identification ability of the speech identity identification device; at the same time, by setting a sample output module to process and output the speech data, better sample reusability is achieved, and in this process, the influence of human factors is eliminated, thereby improving the quality of sample preparation.

[0094] The above are only preferred embodiments of the present invention, and are not intended to limit the implementation methods and protection scope of the present invention. Those skilled in the art should be aware that all solutions obtained by equivalent substitutions and obvious changes made using the description and illustrations of the present invention should be included in the protection scope of the present invention.

Claims

1. A sample preparation system for a speech identity identification device, characterized in that: include: A parameter generation module, wherein the parameter generation module generates a target parameter according to an evaluation requirement; A speech generation module, the speech generation module is connected to the parameter generation module, and the speech generation module generates speech data according to the target parameters: A sample output module, the sample output module is connected to the speech generation module, the sample output module receives the speech data and generates an experimental sample and then outputs it; The speech generation module comprises: A sample collection submodule, which obtains recording samples and stores them in a sample database; A screening submodule, wherein the screening submodule retrieves available speech from the sample database according to the target parameter; A checking submodule, the checking submodule is connected to the screening submodule, and the checking submodule checks the available voice according to the evaluation requirement to generate the voice data; The inspection submodule comprises: a matching degree checking submodule, wherein the matching degree checking submodule parses the evaluation requirement to obtain an expected sample of inspection points of the experimental sample required by the evaluation requirement, and evaluates the number and type of inspection points in the speech data based on the expected sample of inspection points to generate a first checking result; An inspection point association checking submodule, wherein the inspection point association checking submodule parses the evaluation requirement to generate an inspection point distribution of the experimental sample required by the evaluation requirement, and evaluates the distribution of the inspection points in the speech data based on the inspection point distribution to generate a second inspection result; a logic analysis submodule, wherein the logic analysis submodule parses the evaluation requirement to generate brief case information of the experimental sample required by the evaluation requirement, and performs a logic correlation analysis on the voice content of the voice data based on the brief case information to generate a third inspection result; A quality evaluation submodule, wherein the quality evaluation submodule is respectively connected to the matching degree inspection submodule, the inspection point association inspection submodule and the logic analysis submodule, and the quality evaluation submodule eliminates the available voices that do not meet the evaluation requirements according to the first inspection result, the second inspection result and the third inspection result.

2. The sample preparation system according to claim 1, characterized in that: The parameter generation module comprises: A difficulty control submodule, wherein the difficulty control submodule generates a difficulty parameter according to the evaluation requirement and adds the difficulty parameter to the target parameter; A channel parameter setting submodule is configured to generate channel parameters according to the evaluation requirements and add the channel parameters to the target parameters.

3. The sample preparation system according to claim 2, characterized in that: The difficulty parameters include: the number of inspection points, the type of inspection points and the inspection point selection method corresponding to the number of inspection points and the type of inspection points; The channel parameters include at least one of a recording source, a recording mode and a recording parameter.

4. The sample preparation system according to claim 1, characterized in that: The quality evaluation submodule performs a weighted summation on the first inspection result, the second inspection result and the third inspection result, and compares the calculated value with a preset inspection threshold to eliminate the available speech that does not meet the evaluation requirement; The quality evaluation submodule further compares the third inspection result with a preset logic inspection threshold to eliminate the available speech that does not conform to the logical relationship.

5. The sample preparation system according to claim 1, characterized in that: The sample output module comprises: An anti-collusion submodule, the anti-collusion submodule is connected to the voice generation module, the anti-collusion submodule receives the voice data, and adds a unique identifier to each group of the voice data to generate a unique voice; A replication submodule, wherein the replication submodule is connected to the anti-collusion submodule, and the replication submodule replicates the unique voice according to the evaluation requirements and outputs it as the experimental sample.

6. The sample preparation system according to claim 5, characterized in that: The sample output module also includes: A check submodule, wherein the check submodule is connected to the replication submodule, and the check submodule respectively obtains the unique speech and the experimental sample and calculates a hash value for verifying the consistency of the unique speech and the experimental sample.

7. A method for evaluating the identification ability of a speech identity identification device, characterized in that: An experimental sample generated by the sample preparation system according to any one of claims 1 to 6 comprises: Step S1: generating an evaluation requirement corresponding to the speech identity identification device for the speech identity identification device; Step S2: generating the experimental sample based on the evaluation requirement; Step S3: using the experimental samples to evaluate the identification capability of the speech identity identification device.

Images