A voice test method, device, equipment, storage medium and vehicle

Abstract

A voice test method, device, equipment, storage medium and vehicle, relating to the technical field of artificial intelligence, comprising: obtaining a test task, the test task comprising first indication information indicating a test environment and second indication information indicating an audio feature used for executing the test task. Target background audio is obtained based on the first indication information, target test audio is obtained based on the second indication information, and mixing processing is performed based on the target background audio and the target test audio to obtain target audio for testing a voice recognition system using the target audio. That is, the target audio corresponding to the test task is obtained by mixing in the present application, without the need for on-site collection, thereby reducing test cost and improving efficiency.

Description

Technical Field

[0001] This application relates to the field of voice testing technology, and in particular to a voice testing method, apparatus, device, storage medium, and vehicle. Background Technology

[0002] Voice assistant testing refers to the process of comprehensively evaluating the functionality, performance, and user experience of a voice assistant system. Testing tasks include wake-up testing, recognition testing, command execution testing, and feedback testing, ensuring that the voice assistant can accurately execute user commands in different scenarios.

[0003] Currently, for each test task, corresponding audio data needs to be collected for that task in order to conduct the test. When there are a large number of test tasks, audio data needs to be collected for each task, resulting in a large data collection volume, high cost, and low efficiency. Summary of the Invention

[0004] In view of the above problems, this application provides a voice testing method, apparatus, device, storage medium and vehicle to improve testing efficiency and reduce testing costs.

[0005] In a first aspect, this application provides a speech testing method, comprising: acquiring a test task to be executed, the test task including first instruction information and second instruction information, the first instruction information indicating a test environment, and the second instruction information indicating the characteristics of the test audio used to execute the test task; acquiring a target background audio according to the first instruction information, and acquiring a target test audio according to the second instruction information; mixing the target background audio and the target test audio to obtain a target audio; and testing a speech recognition system based on the target audio.

[0006] Based on the technical solution provided in this application, a test task to be executed is obtained. This test task includes first instruction information and second instruction information. The first instruction information indicates the environmental characteristics when executing the test task, for example, the first instruction information indicates that the test task should be executed in an environment where the vehicle is stationary, the air conditioning is off, and the multimedia system is off. The second instruction information indicates the characteristics of the audio used to execute the test task, for example, the second instruction information indicates a speech test on an adult male under normal emotional conditions. Target background audio is obtained based on the first instruction information, and target test audio is obtained based on the second instruction information. The target background audio and target test audio are mixed to obtain target audio, which is then used to test the speech recognition system. In other words, this application obtains the target audio corresponding to the test task by mixing background audio and test audio, eliminating the need for on-site data collection, reducing testing costs, and improving efficiency.

[0007] In one possible implementation, the target test audio is obtained based on the target speech generation model corresponding to the second instruction information. Different second instruction information corresponds to different speech generation models, and the test audio output by the different speech generation models has different characteristics. In this application, different second instruction information corresponds to different speech generation models, which can obtain different types of test audio to meet diverse testing needs. Specifically, the text information in the test task is input into the target speech generation model, and the target speech generation model uses the text information to generate the target test audio.

[0008] In one possible implementation, obtaining the target test audio according to the second indication information includes:

[0009] Determine the target speech generation model corresponding to the second indication information;

[0010] The text information in the test task is input into the target speech generation model to obtain the target test audio output by the target speech generation model.

[0011] In one possible implementation, the step of inputting the text information from the test task into the target speech generation model to obtain the target test audio output by the target speech generation model includes:

[0012] The text information and control parameters in the test task are input into the target speech generation model to obtain the target test audio output by the target speech generation model. The control parameters include speech rate.

[0013] In one possible implementation, the step of inputting the text information from the test task into the target speech generation model to obtain the target test audio output by the target speech generation model includes:

[0014] Obtain the training audio used to train the target speech generation model and the corresponding text information of the training audio;

[0015] The text information from the test task, the audio, and the text information corresponding to the training audio are input into the target speech generation model to obtain the target test audio output by the target speech generation model. The features of the target test audio are matched with the features of the training audio.

[0016] In one possible implementation, the training process of the target speech generation model includes:

[0017] Acquire multiple sets of training audio, wherein at least one feature of one set of training audio differs from the features of the other sets of training audio.

[0018] For any set of training audio, the initial speech generation model is fine-tuned using the training audio to obtain the target speech generation model corresponding to the training audio, wherein the initial speech generation model is a pre-trained model.

[0019] In one possible implementation, the step of mixing the target background audio and the target test audio to obtain the target audio includes:

[0020] Obtain the mixing algorithm and mixing time used during the mixing process, wherein the mixing time indicates the time period during which the target Beijing audio and the target test audio are mixed;

[0021] The target background audio and the target test audio are mixed based on the mixing algorithm and the mixing time to obtain the target audio.

[0022] In one possible implementation, the voice recognition system is deployed on a vehicle, and the background audio is audio collected by the vehicle in different environments. The different environments refer to at least one of the following: the vehicle's operating state, the air conditioning operating state inside the vehicle, the state of the windows, and the multimedia system operating state.

[0023] Secondly, this application also provides a voice testing device, which includes: an acquisition unit, a processing unit, and a testing unit;

[0024] An acquisition unit is used to acquire a test task to be executed, the test task including first indication information and second indication information, the first indication information indicating the test environment, and the second indication information indicating the characteristics of the audio used to execute the test task;

[0025] The acquisition unit is further configured to acquire target background audio according to the first indication information, and acquire target test audio according to the second indication information;

[0026] The processing unit is used to mix the target background audio and the target test audio to obtain the target audio;

[0027] The test unit is used to test the speech recognition system based on the target audio.

[0028] Thirdly, this application also provides a voice testing device, including: a processor, a memory, and a system bus. The processor and the memory are connected via the system bus; the memory is used to store one or more programs, the one or more programs including instructions, which, when executed by the processor, cause the processor to perform the method described in the first aspect and any implementation thereof.

[0029] Fourthly, this application also provides a computer-readable storage medium, characterized in that the computer-readable storage medium stores instructions that, when executed on a voice testing device, cause the voice testing device to perform the method described in the first aspect and any implementation thereof.

[0030] Fifthly, this application also provides a vehicle, which includes the voice testing device provided in the third aspect above. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 This application provides a schematic diagram of a process for obtaining a speech generation model.

[0033] Figure 2 A flowchart of a voice testing method provided in an embodiment of this application;

[0034] Figure 3 A framework diagram of a voice testing technology provided in this application embodiment;

[0035] Figure 4 A structural diagram of a voice testing device provided in an embodiment of this application;

[0036] Figure 5 A schematic diagram of a voice testing device provided in an embodiment of this application;

[0037] Figure 6 This is a schematic diagram of a vehicle provided in an embodiment of this application. Detailed Implementation

[0038] To enable those skilled in the art to better understand the solution of this application, the application scenario of the technical solution of this application will be described first below.

[0039] With the continuous development of artificial intelligence technology, in order to improve user experience, more and more device manufacturers are adding voice assistant (also known as voice recognition system) functionality to their devices. For example, voice assistant software is being deployed on smartphones, in-vehicle systems, and other devices. Through intelligent dialogue and convenient operation, voice assistants greatly save users' time and effort. Voice recognition technology is the foundation for realizing voice assistant functionality.

[0040] Generally, the higher the recognition accuracy of a voice assistant, the stronger its ability to correctly execute commands, the more natural the dialogue, the more stable its operation in daily situations, and the higher its acceptance rate. To ensure this, voice assistants must be properly tested and optimized in real acoustic environments. Common testing tasks include wake-up tests, false wake-up tests, one-shot tests, and automatic speech recognition (ASR) tests.

[0041] Among them, wake-up: test the wake-up mechanism of the voice assistant to ensure that users can successfully wake up the voice assistant through voice buttons or wake words;

[0042] Accidental wake-up: refers to the activation of a voice assistant without receiving a specific wake word;

[0043] ASR: Tests the accuracy and stability of the voice assistant when executing various commands, including controlling vehicle functions (such as air conditioning, windows, navigation, etc.), multimedia control (such as playing music), and free dialogue;

[0044] Unlike wake-up words, OneShot combines wake-up and recognition into one. For example, if you say "XXX, how's the weather today?" in one sentence, the voice assistant will not give a wake-up response, but will directly execute ASR and subsequent processes.

[0045] When testing a voice assistant, it is necessary to collect audio data in a real-world environment using the device where the voice assistant is deployed, and then use this audio data to test the voice assistant. For example, when testing an in-vehicle voice assistant, audio is collected through the vehicle's microphone under different operating conditions for different test tasks, and then simulated tests are conducted in a laboratory.

[0046] However, to comprehensively test a given task, it is necessary to collect audio data from users of different ages in real-vehicle environments under varying noise levels (quiet, low speed, high speed, air conditioning fan speed, media volume) for that specific test task. When there are many types of test tasks, the data collection volume is large, resulting in high costs and low efficiency. To address these issues, speech synthesis has been proposed to obtain the necessary audio for testing. However, current speech synthesis results suffer from limited timbre and diversity, failing to meet the diverse testing requirements. Speech synthesis refers to converting text into speech signals and generating audible speech using a speech synthesizer.

[0047] Based on this, this application provides a voice testing method. When a test task needs to be performed, a matching target background audio is obtained according to the first instruction information in the test task, and a matching target test audio is obtained according to the second instruction information in the test task. The target background audio and the target test audio are mixed to obtain the target audio. It is evident that by mixing the background audio and test audio instead of actual audio recording, costs are reduced and efficiency is improved.

[0048] The acquisition of target test audio based on the second instruction information can include two methods. One method involves pre-establishing a correspondence between different second instruction information and test audio. After acquiring the second instruction information for a specific test task, the target test audio is determined based on this second instruction information and the aforementioned correspondence. The other method involves pre-training multiple different speech generation models, each outputting test audio with different characteristics. After acquiring the second instruction information for a specific test task, the target test audio is obtained using the target speech generation model matched by this second instruction information. This method eliminates the need to collect test audio and enhances the diversity of speech synthesis to meet different testing needs.

[0049] To facilitate understanding of the technical solution of this application, the training process of the speech generation model will be explained below.

[0050] See Figure 1 The figure is a flowchart of a method for obtaining a speech generation model provided in an embodiment of this application, as shown below. Figure 1 As shown, the method includes:

[0051] S101: Obtain multiple sets of training audio.

[0052] In this training audio dataset, at least one feature of one set of training audio differs from the features of the other sets of training audio. Specifically, the features of the training audio can be user features, including user attributes and user emotions. User attributes include at least age and gender; user emotions include normal, happy, sad, excited, etc. That is, the difference between any two sets of training audio can be at least one of different ages, genders, or emotions. The training audio can be audio data actually collected through recording equipment or audio data obtained from existing open-source audio libraries.

[0053] Typically, when testing voice assistants, the audio used is the user's audio under normal emotional conditions. However, considering that users' emotions can fluctuate, in order for the voice assistant to respond to the user's voice commands under different emotional conditions, the training audio obtained includes audio of the same user under different emotions, such as audio of the user when angry, audio of the user when sad, etc.

[0054] For example, as shown in Table 1, the training audio types divide age into three age groups. For each age group, audio is collected under different genders and different emotions. For instance, training audio is collected for men aged 14-50 who are angry; for men aged 14-50 who are happy; for men under 14 who are normal; for women under 14 who are normal; for men over 50 who are normal; and for women over 50 who are normal.

[0055] Table 1 Training Audio

[0056]

[0057] It should be noted that Table 1 is only an example to understand different training audios and is not intended to limit this application. In practical applications, it is also possible to segment users under the age of 14 into multiple different emotions and obtain training audios under different emotions.

[0058] S102: For any set of training audio, fine-tune the initial speech generation model using the training audio to obtain the target speech generation model that matches the features corresponding to the training audio.

[0059] The initial speech generation model is a pre-trained model. For any set of training audio, the initial speech generation model can be fine-tuned using the training audio to obtain a target speech generation model that matches the features corresponding to the training audio.

[0060] For example, in Table 1, there are 14 different sets of training audio. For these 14 sets of training audio, the initial speech generation model is fine-tuned using the first set of training audio to obtain speech generation model 1; the initial speech generation model is fine-tuned using the second set of training audio to obtain speech generation model 2; the initial speech generation model is fine-tuned using the third set of training audio to obtain speech generation model 3, and so on, to obtain 14 speech generation models.

[0061] As can be seen, through the above training, multiple different speech generation models can be obtained, which can generate test audio with different features, thereby achieving diversity of test audio and meeting different testing needs.

[0062] Based on the speech generation model obtained above, the following will describe how to implement speech testing based on this speech generation model with specific embodiments.

[0063] See Figure 2 The figure is a flowchart of a voice testing method provided in an embodiment of this application, as shown below. Figure 2 As shown, the method includes:

[0064] S201: Obtain the test task to be executed, which includes first instruction information and second instruction information.

[0065] The first instruction information indicates the test environment, such as a quiet or noisy test environment; the second instruction information indicates the characteristics of the test audio used to perform the test task, such as the second instruction information indicating a voice test for men aged 14-50 with normal emotions, and a voice test for women aged 14-50 with normal emotions.

[0066] S202: Obtain the target background audio according to the first instruction information, and obtain the target test audio according to the second instruction information.

[0067] In this embodiment, background audio under different environments can be pre-collected, and a correspondence between different environments and background audio can be established. When executing the test task, the target background audio is selected according to the first instruction information and the aforementioned correspondence. For example, if the speech recognition system is deployed in a vehicle, the background audio refers to the audio collected by the vehicle in different environments. These different environments refer to at least one of the following: the vehicle's operating state, the air conditioning's operating state, the window state, and the multimedia system's operating state being different. Alternatively, background audio under different environments can be obtained from an open-source audio library, and a correspondence between different environments and background audio can be established. The open-source audio library may include a large number of background audio tracks under different environments, and the required background audio can be obtained from the open-source audio library according to the actual test requirements.

[0068] The vehicle's operating status can include stationary, low-speed driving, and high-speed driving; the air conditioning's operating status can include air conditioning off, air conditioning on, and air conditioning level; and the multimedia system's operating status can include multimedia off, multimedia on, and volume level when multimedia is on.

[0069] For example, Table 2 shows the background audio of the vehicle under different environments, including quiet, noisy, and acoustic echo cancellation (AEC). For any given environment type, background audio is acquired under specific vehicle, air conditioning, roof / window, and media conditions.

[0070] Table 2 Background Audio Acquisition

[0071]

[0072] It should be noted that Table 2 is only an example for understanding background audio and does not constitute a limitation on background audio. In practical applications, background audio corresponding to more components in different states can be collected.

[0073] After obtaining the first indication information, a target background audio that matches the first indication information can be determined from a variety of pre-collected background audios. For example, if the first indication information is {30~80 Level 4 Off None}, then the target background audio is the background audio corresponding to the second row in Table 2.

[0074] Typically, a single test task will be conducted in multiple different test environments. Therefore, a test task can include multiple first indication messages, such as the first first indication message being {30-80, level 4, off, none}, and the second first indication message being {80-120, level 4, off, none}. Alternatively, a single first indication message can indicate multiple different test environments, for example, the first indication message being {30-80, level 4, off, none; 80-120, level 4, off, none}. Regardless of the method used, the background audio for each test environment needs to be acquired.

[0075] The acquisition of the target test audio based on the second instruction information can include the following implementation methods:

[0076] One approach involves obtaining the target test audio based on the second instruction information and a corresponding relationship, whereby the corresponding relationship includes the relationship between the second instruction information and the target test audio. Specifically, test audio with different features can be pre-collected, and a correspondence between different features and different test audio can be established. Alternatively, multiple different test audio can be pre-generated using multiple different speech generation models, and a correspondence between different features and different test audio can be established. The test audio generated by different speech generation models will have different features. After obtaining the second instruction information in the test task, the target test audio matching the features indicated by the second instruction information and the aforementioned correspondence is obtained.

[0077] Another approach is to generate target test audio in real time using the target speech generation model corresponding to the second instruction information. Different second instruction information corresponds to different speech generation models, and the test audio output by different speech generation models has different characteristics.

[0078] pass Figure 1 As illustrated in the example, multiple different speech generation models are pre-trained. After obtaining the second instruction information, a target speech generation model matching the second instruction information is selected from these multiple models, and the test audio is obtained using this target speech generation model. Specifically, the target speech generation model corresponding to the second instruction information is determined; the text information from the test task is input into the target speech generation model to obtain the target test audio output by the target speech generation model. The test task also includes text information, which the target speech generation model converts into the target test audio.

[0079] For example, if the test task is a wake-up test, and the text information in the test task includes a wake-up word, and the second instruction information indicates the test range as {14-50, male, normal emotion}, then obtain the speech generation model corresponding to the fourth row in Table 1, input the text information into the speech generation model, and obtain the target test audio, which includes the wake-up word.

[0080] In practical implementation, the test device executing the test task can deploy multiple different speech generation models. After acquiring the test task, the test device determines the target speech generation model based on the second instruction information and uses the target speech generation model to generate the target test audio. Alternatively, multiple different speech generation models can be deployed on other devices. After acquiring the test task, the test device sends the second instruction information to the other devices, which then determine the target speech generation model based on the second instruction information, generate the target test audio using the target speech generation model, and send the target test audio to the test device, which then executes step S203.

[0081] Typically, for a given test task, it's necessary to test audio with different features to improve the performance of the speech recognition system. Therefore, a test task can include multiple second indications, for example, the first second indication might be {14-50, male, normal emotion}, the second second indication might be {14-50, female, normal emotion}, etc. Alternatively, a single second indication can indicate multiple different test audio, for example, {14-50, male, normal emotion; 14-50, female, normal emotion; 14-50, male, excited; 14-50, female, excited}. Regardless of the method used, for each indicated audio feature, a speech generation model corresponding to that feature is obtained, and each speech generation model is used to obtain the test audio.

[0082] Optionally, obtaining the target test audio includes: inputting the text information and control parameters from the test task into the target speech generation model to obtain the target test audio output by the target speech generation model. The control parameters include speech rate. That is, in this implementation, the test task includes not only text information but also speech rate. When generating the target test audio using the target speech generation model, the text information and speech rate are used as input to control the speech rate corresponding to the target test audio generated by the target speech generation model, thereby obtaining target test audio with a specific speech rate to meet different testing needs.

[0083] For example, if the speaking speed is measured in words per minute (WPM), and the speaking speed included in the test task is 120 WPM, then the speaking speed corresponding to the generated target test audio is 120 WPM; if the speaking speed included in the test task is 200 WPM, then the speaking speed corresponding to the generated target test audio is 200 WPM.

[0084] The control parameters can also include other parameters, such as the sampling rate. The sampling rate affects the quality and naturalness of the generated speech; generally, a higher sampling rate results in higher sound fidelity. In this implementation, the sampling rate is used as input, thus controlling the quality of the target test audio generated by the target speech generation model.

[0085] To ensure that the timbre of the generated target test audio is consistent with that of the training audio used when training the target speech generation model, the training audio used to train the speech generation model and its corresponding text information are obtained when generating the target test audio using the target speech generation model. This involves inputting not only the text information from the test task into the target speech generation model, but also the training audio and its corresponding text information, to promote feature matching between the features of the target test audio generated by the target speech generation model and the features of the training audio. Specifically, the training audio used to train the target speech generation model and its corresponding text information are obtained; the text information from the test task, the training audio, and its corresponding text information are input into the target speech generation model to obtain the target test audio output by the target speech generation model. The features of the target test audio match the features of the training audio.

[0086] For example, when training the target speech generation model, if the training audio is the audio corresponding to {14-50 male, normal mood}, then when generating the target test audio using the target speech generation model, in addition to the first text information from the test task, the training audio corresponding to {14-50 male, normal mood} and the second text information corresponding to that training audio must also be input. In this way, when generating the target test audio corresponding to the first text information, the target speech generation model uses the training audio as a reference target, controlling the features of the target test audio to approach the features corresponding to the reference target as closely as possible, thus ensuring that the features of the target test audio are as similar as possible to the features of the training audio.

[0087] S203: Mix the target background audio and the target test audio to obtain the target audio.

[0088] In this embodiment, to obtain the audio used for testing, the acquired target background audio and target test audio are mixed to obtain the target audio. This eliminates the need to actually record the audio required for testing, reducing testing costs and improving efficiency.

[0089] As mentioned above, multiple target background audio tracks and multiple target test audio tracks can be obtained. Each target background audio track needs to be mixed with all the target test audio tracks to obtain multiple target audio tracks. For example, if three target background audio tracks and two test audio tracks are obtained, after mixing, six target audio tracks will be obtained.

[0090] Optionally, audio mixing can be achieved by: obtaining the mixing algorithm and mixing time used during the mixing process; and mixing a target background audio and a target test audio based on the mixing algorithm and mixing time to obtain a target audio. Here, audio mixing refers to integrating sounds from multiple sources into a stereo track or mono track.

[0091] The mixing time refers to the time period during which the target background audio and the target test audio are mixed. Specifically, the mixing time may include at least one mixing start time and a mixing duration. During mixing, the computer starts mixing the target background audio and the target test audio from the mixing start time, and ends mixing after the preset mixing duration. The mixing start time typically refers to the point in time when mixing begins after recording is completed; the mixing duration refers to the duration of one mixing cycle. The mixing start time and mixing duration can be determined according to the actual application scenario, and this embodiment does not impose any limitations. Mixing algorithms include, but are not limited to, direct summation, average weight adjustment, summation with clamping, and adaptive mixing weighting (screening factor method), etc. These mixing algorithms are mature technologies, and this embodiment will not describe them in detail.

[0092] In practical implementation, the mixing time can include multiple mix start times and at least one mix duration. Mixing is performed at each start time, and ends when the mix duration reaches the preset duration. Mixing begins again at the next start time. Alternatively, the mixing time can include multiple start times and multiple durations, with one start time corresponding to one duration. For example, the mixing time could include two start times, t1 and t2, and two durations, step1 and step2. When playback time reaches t1, the target background audio and target test audio are mixed; mixing stops when the duration is step1. When playback time reaches t2, the target background audio and target test audio are mixed again; mixing stops when the duration is step2.

[0093] It should be noted that, to ensure the quality of the mixed audio, alignment processing is required before mixing the target background audio and the target test audio. This includes, for example, adjusting the alignment format, sampling rate, frame length, bit depth, and number of channels.

[0094] S204: Test the speech recognition system based on the target audio.

[0095] For each target audio segment acquired, the speech recognition system is tested using that target audio segment to obtain the recognition performance of the speech recognition system, so as to optimize the speech recognition system based on the test results.

[0096] As can be seen, this embodiment obtains the target audio corresponding to the test task by mixing background audio and test audio, eliminating the need for on-site data collection, thus reducing testing costs and improving efficiency. Furthermore, different second instruction information in this application can correspond to different speech generation models, thereby obtaining different types of test audio to meet diverse testing needs.

[0097] For a better understanding of the overall implementation of the voice test, please refer to [link / reference]. Figure 3 The diagram shown illustrates the framework for implementing voice testing, which includes two parts: one part is for acquiring test audio, and the other part is for acquiring the audio required for the test.

[0098] In the first part, a large speech model is deployed and loaded, and a large amount of training audio is obtained. Based on the training requirements, training audio matching the training requirements is obtained. The large speech model is then fine-tuned using the training audio to obtain the fine-tuned large speech model. Based on the second instruction information in the test task, the fine-tuned large speech model is loaded, and test audio is obtained.

[0099] In the second part, the test audio and the background audio indicated by the first instruction information in the test task are mixed to obtain the target audio, and the voice assistant in the vehicle is tested using the target audio.

[0100] Based on the above method embodiments, this application provides a voice testing device and equipment, which will be described below with reference to the accompanying drawings.

[0101] See Figure 4 This figure is a structural diagram of a voice testing device provided in an embodiment of this application, as shown below. Figure 4 As shown, the device 400 includes: an acquisition unit 401, a processing unit 402, and a testing unit 403.

[0102] The acquisition unit 401 is used to acquire a test task to be executed. The test task includes first indication information and second indication information. The first indication information indicates the test environment, and the second indication information indicates the characteristics of the audio used to execute the test task.

[0103] The acquisition unit 401 is further configured to acquire target background audio according to the first indication information, and acquire target test audio according to the second indication information;

[0104] Processing unit 402 is used to mix the target background audio and the target test audio to obtain the target audio;

[0105] Test unit 403 is used to test the voice assistant based on the target audio.

[0106] In one possible implementation, the acquisition unit 401 is specifically used to determine the target speech generation model corresponding to the second instruction information. Different second instruction information corresponds to different speech generation models, and the test audio output by different speech generation models has different characteristics. The text information in the test task is input into the target speech generation model to obtain the target test audio output by the target speech generation model.

[0107] In one possible implementation, the acquisition unit 401 is specifically used to input the text information and control parameters in the test task into the target speech generation model to obtain the target test audio output by the target speech generation model, wherein the control parameters include speech rate.

[0108] In one possible implementation, the acquisition unit 401 is specifically used to acquire the training audio used to train the target speech generation model and the text information corresponding to the training audio; input the text information in the test task, the training audio, and the text information corresponding to the training audio into the target speech generation model to obtain the target test audio output by the target speech generation model, wherein the features of the target test audio match the features of the training audio.

[0109] In one possible implementation, the training process of the target speech generation model includes:

[0110] Acquire multiple sets of training audio, wherein at least one feature of one set of training audio differs from the features of other sets of training audio.

[0111] For any set of training audio, the initial speech generation model is fine-tuned using the training audio to obtain a target speech generation model that matches the features corresponding to the training audio. The initial speech generation model is a pre-trained model.

[0112] In one possible implementation, the processing unit 402 is specifically used to obtain the mixing algorithm and mixing time used when performing the mixing process, wherein the mixing time indicates the time period during which the target background audio and the target test audio are mixed; and to perform mixing processing on the target background audio and the target test audio based on the mixing algorithm and the mixing time to obtain the target audio.

[0113] In one possible implementation, the voice recognition system is deployed on a vehicle, and the background audio is audio collected by the vehicle in different environments. The different environments refer to at least one of the following: the vehicle's operating state, the air conditioning's operating state, the state of the windows, and the multimedia system's operating state.

[0114] It should be noted that the specific implementation of each of the above units can be found in the relevant descriptions in the above method embodiments.

[0115] Furthermore, embodiments of this application also provide another voice testing device, such as... Figure 5 As shown, the voice testing device 500 includes a processor 501, a memory 502, and a system bus 503. The processor 501 and the memory 502 are connected via the system bus 503. The memory 502 is used to store one or more programs, which include instructions. When the instructions are executed by the processor, they cause the processor to perform the aforementioned voice testing method.

[0116] Furthermore, embodiments of this application also provide a computer-readable storage medium storing instructions that, when executed on a voice testing device, cause the voice testing device to perform the aforementioned voice testing method.

[0117] Furthermore, this application embodiment also provides a computer program product, which, when run on a voice testing device, causes the voice testing device to execute the above-described voice testing method.

[0118] Furthermore, embodiments of this application also provide a vehicle, such as... Figure 6 As shown, the vehicle 600 includes the voice testing device 500 described in the above embodiments. This voice testing device 500 can be an in-vehicle infotainment system or other in-vehicle mobile terminal; this application does not specifically limit its functionality. The vehicle can be a battery electric vehicle (BEV), a hybrid electric vehicle (HEV), a plug-in hybrid electric vehicle (PHEV), an extended-range electric vehicle (EREV), a fuel cell vehicle (FCV), etc.

[0119] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that all or part of the steps in the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, a server, or a network communication device such as a media gateway, etc.) to execute the methods described in various embodiments or some parts of the embodiments of this application.

[0120] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0121] It should also be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0122] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A voice testing method, characterized in that, The method includes: Obtain a test task to be executed, the test task including first indication information and second indication information, the first indication information indicating the test environment, and the second indication information indicating the characteristics of the test audio used to execute the test task; The target background audio is obtained according to the first instruction information, and the target test audio is obtained according to the second instruction information; The target background audio and the target test audio are mixed to obtain the target audio. The speech recognition system was tested based on the target audio.

2. The method according to claim 1, characterized in that, The step of obtaining the target test audio according to the second indication information includes: Determine the target speech generation model corresponding to the second instruction information. Different second instruction information corresponds to different speech generation models, and the test audio output by the different speech generation models has different features. The text information in the test task is input into the target speech generation model to obtain the target test audio output by the target speech generation model.

3. The method according to claim 2, characterized in that, The step of inputting the text information from the test task into the target speech generation model to obtain the target test audio output by the target speech generation model includes: The text information and control parameters in the test task are input into the target speech generation model to obtain the target test audio output by the target speech generation model. The control parameters include speech rate.

4. The method according to claim 2 or 3, characterized in that, The step of inputting the text information from the test task into the target speech generation model to obtain the target test audio output by the target speech generation model includes: Obtain the training audio used to train the target speech generation model and the corresponding text information of the training audio; The text information in the test task, the training audio, and the text information corresponding to the training audio are input into the target speech generation model to obtain the target test audio output by the target speech generation model. The features of the target test audio are matched with the features of the training audio.

5. The method according to claim 2, characterized in that, The training process of the target speech generation model includes: Acquire multiple sets of training audio, wherein at least one feature of one set of training audio differs from the features of the other sets of training audio. For any set of training audio, the initial speech generation model is fine-tuned using the training audio to obtain a target speech generation model that matches the features corresponding to the training audio. The initial speech generation model is a pre-trained model.

6. The method according to claim 1, characterized in that, The step of mixing the target background audio and the target test audio to obtain the target audio includes: Obtain the mixing algorithm and mixing time used during the mixing process, wherein the mixing time indicates the time period during which the target background audio and the target test audio are mixed; The target background audio and the target test audio are mixed based on the mixing algorithm and mixing time to obtain the target audio.

7. The method according to claim 1, characterized in that, The voice recognition system is deployed in the vehicle, and the background audio is audio collected by the vehicle in different environments. The different environments refer to at least one of the following: the vehicle's operating status, the air conditioning operating status, the window status, and the multimedia system operating status.

8. A voice testing device, characterized in that, The device includes: an acquisition unit, a processing unit, and a testing unit; An acquisition unit is used to acquire a test task to be executed, the test task including first indication information and second indication information, the first indication information indicating the test environment, and the second indication information indicating the characteristics of the audio used to execute the test task; The acquisition unit is further configured to acquire target background audio according to the first indication information, and acquire target test audio according to the second indication information; The processing unit is used to mix the target background audio and the target test audio to obtain the target audio. The test unit is used to test the speech recognition system based on the target audio.

9. A voice testing device, characterized in that, include: Processor, memory, system bus; The processor and the memory are connected via the system bus; The memory is used to store one or more programs, the one or more programs including instructions that, when executed by the processor, cause the processor to perform the method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed on a task processing device, cause the task processing device to perform the method of any one of claims 1-7.

11. A vehicle, characterized in that, The vehicle includes the voice testing equipment as described in claim 9.

Images