Multi-modal speech language large model training method and device, equipment and medium

Abstract

The invention provides a multi-modal speech language large model training method and device, equipment and a medium, and relates to the technical field of artificial intelligence, in particular to the technical field of large models, speech data processing and data generation. According to the implementation scheme, first inquiry voice data is input into a multi-mode voice language large model, and first reply voice data generated by the multi-mode voice language large model is obtained; determining an inquiry text corresponding to the first inquiry voice data and a reply text corresponding to the first reply voice data; determining a first score based on the inquiry text and the reply text; determining a second score based on voice features of the first inquiry voice data and voice features of the first reply voice data, the voice features including at least one of voice sharpness, speed features, tone features, intonation features and emotion features; and adjusting parameters of the multi-modal speech language large model based on the first score and the second score.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, in particular to the field of speech data processing and data generation technology, and specifically to a training method for a large multimodal speech language model, a speech data generation method and device, an electronic device, a computer-readable storage medium, and a computer program product. Background Art

[0002] Artificial intelligence (AI) is the study of how computers can simulate certain human thought processes and intelligent behaviors (such as learning, reasoning, thinking, and planning). It encompasses both hardware and software technologies. AI hardware technologies generally include sensors, specialized AI chips, cloud computing, distributed storage, and big data processing. AI software technologies primarily encompass computer vision, speech recognition, natural language processing, machine learning / deep learning, big data processing, and knowledge graphs.

[0003] With the development of computer technology, generative models based on artificial intelligence can be applied to various forms of natural language processing tasks, including the processing of natural language text and natural language speech, especially the ability to generate response content based on the user's query content to achieve interaction with the user.

[0004] The approaches described in this section are not necessarily approaches that have been previously conceived or employed. Unless otherwise indicated, it should not be assumed that any approach described in this section is prior art simply by virtue of its inclusion in this section. Similarly, unless otherwise indicated, the issues raised in this section should not be considered as having been recognized in any prior art. Summary of the Invention

[0005] The present disclosure provides a method, apparatus, electronic device, computer-readable storage medium, and computer program product for training a large multimodal speech language model.

[0006] According to one aspect of the present disclosure, a training method for a multimodal speech language large model is provided, comprising: obtaining first reply speech data generated by the multimodal speech language large model by inputting first inquiry speech data into the multimodal speech language large model; determining an inquiry text corresponding to the first inquiry speech data and a reply text corresponding to the first reply speech data; determining a first score based on the inquiry text and the reply text; determining a second score based on speech features of the first inquiry speech data and speech features of the first reply speech data, wherein the speech features include at least one of speech clarity, speaking speed features, timbre features, intonation features and emotional features; and adjusting parameters of the multimodal speech language large model based on the first score and the second score.

[0007] According to one aspect of the present disclosure, a voice data generation method is provided, comprising: obtaining inquiry voice data from a user; obtaining reply voice data generated by the multimodal voice language large model by inputting the inquiry voice data into a multimodal voice language large model trained using the above-mentioned multimodal voice language large model training method; and returning the reply voice data to the user.

[0008] According to one aspect of the present disclosure, a training device for a multimodal speech language large model is provided, comprising: a first acquisition unit, configured to acquire first reply speech data generated by the multimodal speech language large model by inputting first inquiry speech data into the multimodal speech language large model; a first determination unit, configured to determine an inquiry text corresponding to the first inquiry speech data and a reply text corresponding to the first reply speech data; a second determination unit, configured to determine a first score based on the inquiry text and the reply text; a third determination unit, configured to determine a second score based on speech features of the first inquiry speech data and speech features of the first reply speech data, wherein the speech features include at least one of speech clarity, speaking speed features, timbre features, intonation features and emotional features; and an adjustment unit, configured to adjust parameters of the multimodal speech language large model based on the first score and the second score.

[0009] According to one aspect of the present disclosure, a speech data generation device is provided, comprising: a multimodal speech language large model trained by the above-mentioned multimodal speech language large model training device; a second acquisition unit, configured to acquire inquiry speech data from a user; a third acquisition unit, configured to acquire reply speech data generated by the multimodal speech language large model by inputting the inquiry speech data into the multimodal speech language large model; and a return unit, configured to return the reply speech data to the user.

[0010] According to one aspect of the present disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute at least one of the above-mentioned multimodal speech language large model training method and speech data generation method.

[0011] According to one aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable the computer to execute at least one of the above-mentioned multimodal speech language large model training method and speech data generation method.

[0012] According to one aspect of the present disclosure, a computer program product is provided, comprising a computer program, wherein the computer program, when executed by a processor, can implement at least one of the above-mentioned multimodal speech language large model training method and speech data generation method.

[0013] According to one or more embodiments of the present disclosure, the quality of voice data generation can be improved, and more accurate human-computer voice question-and-answer interaction can be achieved.

[0014] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present disclosure, nor are they intended to limit the scope of the present disclosure. Other features of the present disclosure will become readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The accompanying drawings illustrate exemplary embodiments and constitute a part of the specification. Together with the description of the specification, they serve to explain exemplary implementation of the embodiments. The illustrated embodiments are for illustrative purposes only and do not limit the scope of the claims. Throughout the drawings, the same reference numerals designate similar, but not necessarily identical, elements.

[0016] Figure 1 A schematic diagram illustrating an exemplary system in which the various methods described herein may be implemented according to exemplary embodiments of the present disclosure;

[0017] Figure 2 A flowchart of a method for training a large multimodal speech language model according to an exemplary embodiment of the present disclosure is shown;

[0018] Figure 3 A schematic structural diagram of a multimodal speech language large model according to an exemplary embodiment of the present disclosure is shown;

[0019] Figure 4 A schematic structural diagram of a question-answering evaluation model according to an exemplary embodiment of the present disclosure is shown;

[0020] Figure 5 A schematic diagram illustrating a training process of a large multimodal speech language model according to an exemplary embodiment of the present disclosure is shown;

[0021] Figure 6 A flowchart of a method for generating speech data according to an exemplary embodiment of the present disclosure is shown;

[0022] Figure 7 A schematic diagram illustrating a process of generating speech data according to an exemplary embodiment of the present disclosure is shown;

[0023] Figure 8 A structural block diagram of a training device for a multimodal speech language large model according to an exemplary embodiment of the present disclosure is shown;

[0024] Figure 9 shows a structural block diagram of a voice data generating apparatus according to an exemplary embodiment of the present disclosure;

[0025] Figure 10 A structural block diagram of an exemplary electronic device that can be used to implement the embodiments of the present disclosure is shown. DETAILED DESCRIPTION

[0026] The following description of exemplary embodiments of the present disclosure is made in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be appreciated by those skilled in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0027] In this disclosure, unless otherwise specified, the use of terms such as "first" and "second" to describe various elements is not intended to limit the positional relationship, temporal relationship, or importance relationship of these elements. Such terms are only used to distinguish one element from another. In some examples, the first element and the second element may refer to the same instance of the element, while in some cases, based on the context of the description, they may also refer to different instances.

[0028] The terms used in the descriptions of the various examples described in this disclosure are for the purpose of describing specific examples only and are not intended to be limiting. Unless the context clearly indicates otherwise, if the number of elements is not specifically limited, the element may be one or more. In addition, the term "and / or" used in this disclosure encompasses any one and all possible combinations of the listed items.

[0029] The embodiments of the present disclosure will be described in detail below with reference to the accompanying drawings.

[0030] Figure 1 FIG2 is a schematic diagram of an exemplary system 100 in which the various methods and apparatuses described herein may be implemented according to an embodiment of the present disclosure. Figure 1 , the system 100 includes one or more client devices 101, 102, 103, 104, 105, and 106, a server 120, and one or more communication networks 110 coupling the one or more client devices to the server 120. The client devices 101, 102, 103, 104, 105, and 106 can be configured to execute one or more applications.

[0031] In an embodiment of the present disclosure, the server 120 may run one or more services or software applications that enable the execution of a method for training a large multimodal speech language model.

[0032] In some embodiments, server 120 may also provide other services or software applications, which may include non-virtualized environments and virtualized environments. In some embodiments, these services may be provided as web-based services or cloud services, such as provided to users of client devices 101, 102, 103, 104, 105, and / or 106 under a software as a service (SaaS) model.

[0033] exist Figure 1 In the configuration shown, the server 120 may include one or more components that implement the functions performed by the server 120. These components may include software components, hardware components, or a combination thereof that can be executed by one or more processors. Users operating client devices 101, 102, 103, 104, 105, and / or 106 may, in turn, utilize one or more client applications to interact with the server 120 to utilize the services provided by these components. It should be understood that a variety of different system configurations are possible, which may differ from the system 100. Therefore, Figure 1 is one example of a system for implementing the various methods described herein and is not intended to be limiting.

[0034] The user may use client devices 101, 102, 103, 104, 105 and / or 106 to send inquiry voice data. The client device may provide an interface that enables the user of the client device to interact with the client device. The client device may also output information to the user via the interface. Figure 1 Only six client devices are depicted, but one skilled in the art will appreciate that the present disclosure can support any number of client devices.

[0035] Client devices 101, 102, 103, 104, 105, and / or 106 may include various types of computer devices, such as portable handheld devices, general-purpose computers (such as personal computers and laptops), workstation computers, wearable devices, smart screen devices, self-service terminal devices, service robots, gaming systems, thin clients, various messaging devices, sensors or other sensing devices, etc. These computer devices may run various types and versions of software applications and operating systems, such as MICROSOFT Windows, APPLE iOS, UNIX-like operating systems, Linux, or Linux-like operating systems (such as GOOGLE Chrome OS); or include various mobile operating systems, such as MICROSOFT Windows Mobile OS, iOS, Windows Phone, and Android. Portable handheld devices may include cellular phones, smartphones, tablet computers, personal digital assistants (PDAs), etc. Wearable devices may include head-mounted displays (such as smart glasses) and other devices. Gaming systems may include various handheld gaming devices, internet-enabled gaming devices, etc. The client device is capable of executing various different applications, such as various Internet-related applications, communication applications (eg, email applications), Short Message Service (SMS) applications, and may use various communication protocols.

[0036] The network 110 may be any type of network known to those skilled in the art that can support data communications using any of a variety of available protocols, including but not limited to TCP / IP, SNA, IPX, etc. By way of example only, the one or more networks 110 may be a local area network (LAN), an Ethernet-based network, a token ring, a wide area network (WAN), the Internet, a virtual network, a virtual private network (VPN), an intranet, an extranet, a blockchain network, a public switched telephone network (PSTN), an infrared network, a wireless network (e.g., Bluetooth, WIFI), and / or any combination of these and / or other networks.

[0037] Server 120 may include one or more general-purpose computers, specialized server computers (e.g., PC (personal computer) servers, UNIX servers, mid-range servers), blade servers, mainframe computers, server clusters, or any other suitable arrangement and / or combination. Server 120 may include one or more virtual machines running virtual operating systems, or other computing architectures involving virtualization (e.g., one or more flexible pools of logical storage devices that may be virtualized to maintain a server's virtual storage device). In various embodiments, server 120 may run one or more services or software applications that provide the functionality described below.

[0038] The computing units in the server 120 may run one or more operating systems including any of the operating systems described above as well as any commercially available server operating systems. The server 120 may also run any of a variety of additional server applications and / or middle-tier applications, including HTTP servers, FTP servers, CGI servers, JAVA servers, database servers, and the like.

[0039] In some implementations, server 120 may include one or more applications to analyze and consolidate data feeds and / or event updates received from users of client devices 101, 102, 103, 104, 105, and 106. Server 120 may also include one or more applications to display the data feeds and / or real-time events via one or more display devices of client devices 101, 102, 103, 104, 105, and 106.

[0040] In some embodiments, server 120 may be a distributed system server or a server integrated with blockchain. Server 120 may also be a cloud server, or an intelligent cloud computing server or intelligent cloud host equipped with artificial intelligence technology. A cloud server is a host product within the cloud computing service system that addresses the management difficulties and poor scalability of traditional physical hosts and virtual private servers (VPS) services.

[0041] The system 100 may also include one or more databases 130. In some embodiments, these databases may be used to store data and other information. For example, one or more of the databases 130 may be used to store information such as audio files and video files. The databases 130 may reside in a variety of locations. For example, the database used by the server 120 may be local to the server 120, or may be remote from the server 120 and communicate with the server 120 via a network-based or dedicated connection. The databases 130 may be of different types. In some embodiments, the databases used by the server 120 may be, for example, relational databases. One or more of these databases may store, update, and retrieve data to and from the databases in response to commands.

[0042] In some embodiments, one or more of the databases 130 may also be used by applications to store application data. The databases used by the applications may be different types of databases, such as a key-value store, an object store, or a conventional store backed by a file system.

[0043] Figure 1The system 100 may be configured and operated in various ways to enable application of the various methods and apparatuses described in accordance with the present disclosure.

[0044] In related technologies, when using an AI-based data generation model to conduct human-computer dialogue with a user, the data generation model is typically a large language model capable of processing and generating natural language text. In this case, when a voice dialogue between a user and a computer is required, a cascade system consisting of a speech recognition module, a large language model, and a speech synthesis module can be used to generate reply voice data for broadcasting to the user. This involves using speech recognition technology to convert the user's input voice query data into query text, having the large language model generate reply text based on the query text, and then using speech synthesis technology to generate reply voice data corresponding to the reply text. While this approach can leverage the thinking capabilities of existing large language models, system cascade errors can affect the overall voice question-answering performance, resulting in insufficient accuracy in voice responses. Furthermore, this approach fails to utilize the AI network to generate voice responses with diverse speech speed, timbre, and emotion, failing to fully meet user needs. The quality of the reply voice data still needs improvement.

[0045] Based on this, the present disclosure provides a training method for a large multimodal speech and language model, which sets reward scores from two perspectives: text content and speech features for the reply speech data directly generated by the large multimodal speech and language model. This allows the accuracy of the reply content and the quality of the reply speech (for example, higher clarity, matching of speaking emotions with inquiry data, appropriate speaking speed, etc.) to be optimized based on the text score and speech score, respectively, and a more accurate voice reply can be achieved using the optimized model.

[0046] Figure 2 FIG. 2 shows a flow chart of a method 200 for training a large multimodal speech language model according to an exemplary embodiment of the present disclosure. Figure 2 As shown, the method 200 includes:

[0047] Step S201: Inputting first inquiry voice data into the multimodal voice language model to obtain first reply voice data generated by the multimodal voice language model;

[0048] Step S202: Determine the inquiry text corresponding to the first inquiry voice data and the reply text corresponding to the first reply voice data;

[0049] Step S203: determining a first score based on the inquiry text and the reply text;

[0050] Step S204: determining a second score based on the voice features of the first inquiry voice data and the voice features of the first reply voice data, wherein the voice features include at least one of voice clarity, speech rate, timbre, intonation, and emotion; and

[0051] Step S205: Adjust the parameters of the multimodal speech language model based on the first score and the second score.

[0052] By applying the above method 200, it is possible to set reward scores from both the text content and voice feature perspectives for the reply voice data directly generated by the model during the training process of the multimodal speech language large model, thereby being able to more accurately optimize the accuracy of the reply content and the quality of the reply voice based on the text content score and the voice feature score. For example, the reply voice can be made clearer and at a more appropriate speed, or the emotional expression of the reply voice can be made more consistent with the emotional expression of the inquiry voice, and the optimized model can be used to achieve more accurate and high-quality voice replies.

[0053] In some examples, the multimodal speech language large model can be a speech language large model pre-trained using large-scale corpus, and the multimodal speech language large model can intelligently understand and generate content for multiple modalities (such as speech and text information). In some examples, the multimodal speech language large model adopts a cross-modal encoder-decoder architecture and an attention mechanism to achieve semantic alignment and joint modeling of multimodal information. By applying the above method 200 to optimize the training of the pre-trained large model, the diversity of the speech features of the speech data can be further optimized on the basis of fully utilizing the semantic understanding and speech generation capabilities of the pre-trained large model, so as to more efficiently obtain a multimodal speech language large model that can express rich and accurate speech features, thereby achieving more accurate human-computer voice question and answer.

[0054] In some examples, determining the inquiry text corresponding to the first inquiry voice data and the reply text corresponding to the first reply voice data in step S202 can be achieved using automatic speech recognition (ASR) technology, that is, performing text recognition on the first inquiry voice data and the first reply voice data, and determining the inquiry text and reply text based on the text recognition results, so as to more accurately evaluate the accuracy of the reply content based on the text content and improve the quality of the model reply.

[0055] In some examples, in step S203, the inquiry text and the reply text are input into the scoring model to obtain a first score output by the scoring model. The first score can be used to indicate whether the facts described in the reply text are accurate, and can also indicate whether the content of the reply text conforms to the question intention of the inquiry text. In some examples, the scoring model can be obtained by supervised training using question-answer text pairs annotated with reference scores, and then the output of the scoring model can be used to evaluate the content quality of the reply text. In some examples, the scoring model can be constructed based on a large language model (LLM) trained using large-scale corpus. For example, it can be fine-tuned based on the pre-trained large language model using the above-mentioned annotated data, so that the evaluation model can more comprehensively and accurately evaluate the quality of the reply content.

[0056] According to some embodiments, determining the second score based on the voice features of the first inquiry voice data and the voice features of the first reply voice data in step S204 includes: in response to determining that the first inquiry voice data includes descriptive information for the voice features of the first reply voice data, determining the second score based on the descriptive information, the voice features of the first inquiry voice data, and the voice features of the first reply voice data. Thus, explicit instructions contained in the inquiry data (e.g., explicit requirements for reply timbre, intonation, and speaking speed) can be obtained during the scoring process, so that the optimized model can output reply data that better meets the requirements of the question.

[0057] According to some embodiments, determining the second score based on the voice features of the first inquiry voice data and the voice features of the first reply voice data in step S204 includes: determining the identity features of the speaker based on the voice features of the first inquiry voice data; and determining the second score based on the identity features, the reply text, and the voice features of the first reply voice data. Thus, the voice features of the inquiry data can be captured during the scoring process to avoid inconsistencies between the reply data and the inquiry data. For example, when the inquiry data is from a female voice, no terms specific to men should appear in the reply content, thereby improving the accuracy of the reply.

[0058] According to some embodiments, determining the second score based on the voice features of the first inquiry voice data and the voice features of the first reply voice data in step S204 includes: inputting the first inquiry voice data and the first reply voice data into a question-and-answer evaluation model to determine the second score output by the question-and-answer evaluation model, wherein the question-and-answer evaluation model is trained using the first sample inquiry voice data, the first sample reply voice data, and a reference score. By training the question-and-answer evaluation model with labeled data, and then using the question-and-answer evaluation model to output a reward score for the voice features of the voice data during the training of a large multimodal speech and language model, scoring efficiency and accuracy can be improved.

[0059] In some examples, the question-answering evaluation model can be trained using multiple question-answering voice data pairs annotated with reference scores. In order to improve the accuracy of the question-answering evaluation model in evaluating the voice features of the reply voice data, the training data may include question-answering voice data pairs with the same text content but different voice features. For example, the training data may include: voice data 1 corresponding to content A and timbre B, voice data 2 corresponding to content A and timbre C. For another example, the training data may also include: voice data 3 corresponding to content A and emotion D, voice data 4 corresponding to content A and emotion E. By configuring question-answering voice data pairs with the same text content but different voice features in the training data, the impact of differences in text content on the voice data features can be eliminated, enabling the evaluation model to more accurately learn different voice features (such as voice clarity, speaking speed features, timbre features, intonation features, and emotional features) to more accurately evaluate the quality of voice data with different voice features.

[0060] In some examples, steps S203 and S204 may also determine the first score and the second score based on other methods, such as predefined scoring rules. In this example, the predefined scoring rules may include multiple dimensions, such as content conciseness, speech clarity, and whether the speech data expresses positive emotions. By predefining scoring rules and using them to guide the optimization training of the multimodal speech and language model, the output results of the multimodal speech and language model can be made more consistent with the needs of actual application scenarios, achieving higher-quality human-machine voice question and answer.

[0061] According to some embodiments, adjusting the parameters of the large multimodal speech and language model based on the first and second scores in step S205 includes: determining reward information based on the first and second scores; and adjusting the parameters of the large multimodal speech and language model based on a reinforcement learning strategy corresponding to the reward information. In this way, reinforcement learning training can be performed based on the score information to obtain a large multimodal speech and language model with improved performance.

[0062] In some examples, step S205 may utilize the RLHF (Reinforcement Learning with Human Feedback) method to perform model tuning, i.e., model parameters may be adjusted based on various types of reinforcement learning strategies, such as a proximal strategy optimization strategy, a group strategy relative optimization strategy, a direct preference optimization strategy, etc. By applying the scoring reward information and the strategy optimization algorithm to adjust the parameters of the multimodal speech and language large model, the output results of the multimodal speech and language large model can be made as close as possible to the expected results of the scoring reward information, i.e., higher quality response voice data can be obtained, thereby improving the accuracy of human-computer voice interaction.

[0063] According to some embodiments, method 200 further includes: inputting second inquiry voice data into the multimodal speech and language model to obtain second reply voice data generated by the multimodal speech and language model; obtaining reference reply voice data corresponding to the second inquiry voice data; and adjusting parameters of the multimodal speech and language model based on the second reply voice data and the reference reply voice data. In this way, the quality of the reply voice data output by the model can be further improved by combining reward-based scoring with supervised training using labeled data.

[0064] In some examples, the first inquiry voice data and the second inquiry voice data can be the same, that is, after the inquiry voice data is input into the multimodal speech language model, the quality score of the reply voice data generated by the model is performed, and the loss value can also be calculated based on the reply voice data generated by the model and the reference reply voice data corresponding to the inquiry voice data, or the quality score result can be affected by the difference between the model output result and the reference result, so as to perform model tuning training in combination with the reference annotation information and quality score information of the sample data. In some examples, the first inquiry voice data and the second inquiry voice data can be different, that is, different training methods are applied to the multimodal speech language model. In one example, the training process of the multimodal speech language model includes the following two stages: in the first stage, supervised fine-tuning (SFT) can be performed on the basis of the pre-trained speech language model, that is, the sample inquiry voice data annotated with the reference reply voice data is input into the pre-trained model, and the loss value is calculated based on the model output result and the annotation content, and then the model parameters are adjusted according to the loss value, which corresponds to the technical means described above for training using the second inquiry voice data and the reference reply voice data corresponding to the second inquiry voice data. Training at this stage might, for example, employ autoregressive learning with a maximum likelihood objective to improve training efficiency and effectiveness. In the second stage, the unlabeled first-question voice data might be fed into a large multimodal speech and language model. The model output is then scored for content quality and audio quality. Reinforcement learning training is then conducted based on these two dimensions of reward scoring to more precisely optimize the accuracy of response content and the model's ability to represent a wide variety of speech features, ultimately improving model performance.

[0065] According to some embodiments, the multimodal speech language large model generates the second reply speech data in the following manner: generating predicted thought information based on the second inquiry speech data, wherein the predicted thought information includes descriptive information for the speech features of the second reply speech data; and generating the second reply speech data based on the predicted thought information, and wherein method 200 further includes: obtaining reference thought information corresponding to the second inquiry speech data; and adjusting the parameters of the multimodal speech language large model based on the predicted thought information and the reference thought information. In this way, the thinking process based on the thought chain can be combined in the reply data generation process, that is, the large model decomposes the generation task, outputs based on the logical chain of first thinking and then generating, and uses the thought information to indicate the speech features that need to be output, so that the generated reply speech data is more accurate.

[0066] In some examples, the reasoning mechanism of the multimodal speech language large model is built based on the Chain-of-Thought (CoT) technology, that is, the model simulates the human cognitive mode of "step-by-step thinking and step-by-step deduction". After receiving the input inquiry voice data, it first outputs thinking information, and then further generates reply voice data. By making the model think about the voice features of the reply voice data to be generated first, the quality of the voice data generated by the model can be improved. In this example, the sample data in the model training process is annotated by reference thinking information, that is, the model's thinking process can be accurately optimized based on the reference thinking information and the predicted thinking information output by the model, further improving the accuracy of model reasoning and improving the quality of data generation.

[0067] According to some embodiments, the reference thought information is text information, and the generating of predicted thought information based on the second inquiry voice data includes: encoding the second inquiry voice data into an inquiry semantic vector in a semantic vector space; and generating a thought semantic vector in the semantic vector space based on the inquiry semantic vector, and wherein the generating of the second reply voice data based on the predicted thought information includes: generating a reply semantic vector in the semantic vector space based on the thought semantic vector; and decoding the reply semantic vector into the second reply voice data, and wherein adjusting the parameters of the multimodal speech language model based on the predicted thought information and the reference thought information includes: decoding the thought semantic vector into a predicted thought text; and adjusting the parameters of the multimodal speech language model based on the predicted thought text and the reference thought information. Thus, text information can be used to more easily and accurately annotate the model's thought information, thereby improving the accuracy of the model's thinking. In this case, the inquiry-thinking-reply chain of voice question-answering using the model includes voice modal data and text modal data. By converting the voice modal data and text modal data into the same semantic vector space, multimodal unified intelligent content understanding is achieved, avoiding the cascade loss caused by modal conversion, optimizing model performance, and improving the accuracy of data generation.

[0068] In some examples, the operation of encoding speech modal data and text modal data into semantic vectors in the speech vector space is implemented based on Tokenizer technology. Specifically, the word segmenter can split the original speech modal data and text modal data into discrete text tokens and audio tokens, and then map them into digital sequences that can be processed by the computer, that is, encoded into semantic vectors. In this example, the mapping vocabulary of text tokens and the mapping vocabulary of audio tokens are spliced to form the overall vocabulary space of the multimodal speech language model, that is, the multimodal information is mapped to the same semantic vector space, so that the model can understand and process multimodal information and improve the quality of data generation.

[0069] According to some embodiments, the reference reply voice data is marked with voice breakpoints, and the second reply voice data is composed of a first voice segment and a second voice segment. Adjusting the parameters of the multimodal voice language model based on the second reply voice data and the reference reply voice data includes: splitting the reference reply voice data into a first reference segment and a second reference segment based on the voice breakpoints; adjusting the parameters of the multimodal voice language model based on the first voice segment and the first reference segment; and adjusting the parameters of the multimodal voice language model based on the second voice segment and the second reference segment. In some examples, the multimodal voice language model is configured to perform phased output during the data generation process. When part of the reply voice has been generated, the generated part of the reply voice is first output to the user, and the remaining part of the reply voice continues to be generated while the output is broadcast. In this case, the multimodal voice language model can broadcast voice replies to the user more quickly during the human-computer voice question and answer process, reducing user waiting time. By optimizing the model parameters using training data marked with speech breakpoints, the model can learn when to split the reply speech data, and then determine when to output partial reply speech first during the data generation process. By outputting the reply speech data in the form of a sequence of speech fragments, there is no need to wait for all the reply data to be generated before voice broadcasting, thereby improving the model's response fluency during human-computer voice interaction and optimizing the user experience.

[0070] Figure 3 FIG. 1 shows a schematic diagram of the structure of a multimodal speech language model according to an exemplary embodiment of the present disclosure. Figure 3As shown, the multimodal speech language large model includes a speech encoder 301, a speech decoder 302, and a thinking and generation network 303. In this example, the speech encoder 301 is used to encode the inquiry speech data into an inquiry semantic vector, so that the thinking and generation network 303 can think and reply based on the inquiry semantic vector, and then output the thinking semantic vector and the reply semantic vector in sequence. In one example, the thinking and generation network 303 first determines the thinking semantic vector based on the inquiry semantic vector, indicates the content and speech features to be generated based on the thinking content, and then further thinks and generates based on the inquiry semantic vector and the thinking semantic vector to obtain the final reply semantic vector. In one example, the thinking semantic vector can be decoded into a thinking text and output to show explicit thinking information to the user, and the user can provide feedback to the model according to needs after viewing the thinking information. The speech decoder 302 is used to decode the reply semantic vector into reply speech data and output it. In this example, the query semantic vector, thinking semantic vector, and reply semantic vector are in the same semantic vector space. By mapping multimodal information to the semantic vector space, the thinking and generation network 303 can uniformly and intelligently understand and process the multimodal information to improve the accuracy of data generation.

[0071] Figure 4 FIG. 1 shows a schematic diagram of the structure of a question-answer evaluation model according to an exemplary embodiment of the present disclosure. Figure 4 As shown, the question-answering evaluation model includes a speech encoder 401 and an evaluation network 402. The speech encoder 401 is used to encode the inquiry speech data and the reply speech data to be evaluated into an inquiry semantic vector and a reply semantic vector, so that the evaluation network 402 can obtain a second score based on the inquiry semantic vector and the reply semantic vector. In some examples, the speech encoder 401 in the question-answering evaluation model and the speech encoder 301 in the thinking and generation network 303 described above can be the same unit, that is, the thinking and generation network 303 described above and the evaluation network 402 in the question-answering evaluation model can perform intelligent understanding and information processing based on the same semantic vector space.

[0072] Figure 5 FIG. 1 is a schematic diagram showing a training process of a multimodal speech language large model according to an exemplary embodiment of the present disclosure. Figure 5As shown, after the inquiry voice data is input into the multimodal voice language model 501 and the reply voice data is obtained, the voice recognition system 502 is used to determine the inquiry text corresponding to the inquiry voice data and the reply text corresponding to the reply voice data, and then the first evaluation model 503 is used to output a first score based on the inquiry text and the reply text, which can indicate whether the reply content to the inquiry content is accurate and comprehensive. The second evaluation model 504 is used to output a second score for voice features based on the inquiry voice data and the reply voice data to indicate whether the reply voice data is clear or whether its speaking speed, timbre, and emotion meet the requirements of the inquiry data. By training and optimizing the multimodal voice language model 503 based on the first score and the second score, the content accuracy and voice performance richness of the reply output by the multimodal voice language model 503 can be accurately optimized, and the trained model can be used to improve the accuracy of human-machine voice question and answer.

[0073] In one example, the multimodal speech and language model 503 is fine-tuned using labeled sample data prior to the aforementioned training phase. The labeled sample data can be annotated with both reference speech response data and reference thought information. This annotation information can then be used to specifically optimize the model's thought process and data generation process, improving training efficiency and optimizing model performance.

[0074] In one example, the multimodal speech language model 503 outputs reply speech data in the form of a sequence of speech segments. In this case, after the multimodal speech language model 503 has finished outputting all the reply speech segments, it can be spliced into complete reply speech data, which can then be converted into reply text and scored. When the speech breakpoints of the reference speech reply data are marked in the annotated sample data, the timing of splitting the reply speech segments output by the model can be indicated based on the speech breakpoint information. By tuning based on the annotated information, the model can learn the accurate timing of splitting the speech reply, thereby improving the fluency of the speech reply.

[0075] According to one aspect of the present disclosure, a method for generating speech data is provided. Figure 6 FIG. 6 shows a flow chart of a method 600 for generating speech data according to an exemplary embodiment of the present disclosure. Figure 6 As shown, the method 600 includes:

[0076] Step S601: Acquire inquiry voice data from the user;

[0077] Step S602: Inputting the inquiry voice data into the multimodal voice language model trained using method 200 to obtain reply voice data generated by the multimodal voice language model; and

[0078] Step S603: Return the reply voice data to the user.

[0079] By applying the above-mentioned multimodal speech language model to conduct human-computer voice question and answer, it is possible to return to the user reply voice data with richer and more accurate voice features such as timbre, emotion, and speaking speed, so that the reply content is more in line with user needs and improves the user experience.

[0080] Figure 7 FIG. 1 shows a schematic diagram of a voice data generation process according to an exemplary embodiment of the present disclosure. Figure 7 As shown, after receiving inquiry voice 01, the multimodal speech language model 700 can first think and output stage-by-stage thought information A, and then output a response voice a based on thought information A. While response voice a is playing, the model can continue to think and output the next stage of thought information B, and then output a response voice b based on thought information B. When the model determines that it has output all responses to inquiry voice 01, a symbol indicating that the answer is complete can be added to the end of response voice b, indicating the end of the question-and-answer round for inquiry voice 01. After the previous round ends, the user can continue to input inquiry voice 02. The multimodal speech language model 700 can then think and generate data based on the new inquiry voice 02, outputting thought information C and response voice c. By having the model output response voice data as a sequence of voice segments during the data generation process, sequential voice output can be achieved during the voice interaction process, eliminating the need to wait for all response data to be generated before voice playback. This reduces user waiting time, improves the fluency of voice responses, and ultimately enhances the user experience.

[0081] According to one aspect of the present disclosure, a device for training a large multimodal speech language model is provided. Figure 8 FIG. 8 shows a structural block diagram of a training apparatus 800 for a multimodal speech language large model according to an exemplary embodiment of the present disclosure. Figure 8 As shown, the apparatus 800 includes:

[0082] The first acquisition unit 801 is configured to input the first inquiry voice data into the multimodal voice language model to acquire the first reply voice data generated by the multimodal voice language model;

[0083] A first determining unit 802 is configured to determine an inquiry text corresponding to the first inquiry voice data and a reply text corresponding to the first reply voice data;

[0084] A second determining unit 803 is configured to determine a first score based on the inquiry text and the reply text;

[0085] a third determining unit 804 configured to determine a second score based on voice features of the first inquiry voice data and voice features of the first reply voice data, wherein the voice features include at least one of voice clarity, speech rate, timbre, intonation, and emotion; and

[0086] The adjustment unit 805 is configured to adjust the parameters of the multimodal speech language model based on the first score and the second score.

[0087] According to some embodiments, the third determination unit 804 is configured to: in response to determining that the first inquiry voice data includes descriptive information for the voice features of the first reply voice data, determine the second score based on the descriptive information, the voice features of the first inquiry voice data, and the voice features of the first reply voice data.

[0088] According to some embodiments, the third determination unit 804 includes: a first determination subunit, configured to determine the identity feature of the speaker based on the voice feature of the first inquiry voice data; and a second determination subunit, configured to determine the second score based on the identity feature, the reply text and the voice feature of the first reply voice data.

[0089] According to some embodiments, the third determination unit 804 is configured to determine the second score output by the question and answer evaluation model by inputting the first inquiry voice data and the first reply voice data into a question and answer evaluation model, wherein the question and answer evaluation model is trained using the first sample inquiry voice data, the first sample reply voice data and the reference score.

[0090] According to some embodiments, the adjustment unit 805 includes: a third determination subunit, configured to determine reward information based on the first score and the second score; and a first adjustment subunit, configured to adjust the parameters of the multimodal speech language model based on the reinforcement learning strategy corresponding to the reward information.

[0091] According to some embodiments, the first acquisition unit 801 is also configured to obtain second reply voice data generated by the multimodal speech language model by inputting second inquiry voice data into the multimodal speech language model, and the device 800 also includes: a fourth acquisition unit, configured to obtain reference reply voice data corresponding to the second inquiry voice data, and wherein the adjustment unit 805 is further configured to adjust the parameters of the multimodal speech language model based on the second reply voice data and the reference reply voice data.

[0092] According to some embodiments, the multimodal speech language large model is configured to: generate predicted thinking information based on the second inquiry voice data, wherein the predicted thinking information includes descriptive information of the voice features of the second reply voice data; and generate the second reply voice data based on the predicted thinking information, and wherein the device 800 also includes: a fifth acquisition unit, configured to obtain reference thinking information corresponding to the second inquiry voice data, and the adjustment unit 805 is configured to adjust the parameters of the multimodal speech language large model based on the predicted thinking information and the reference thinking information.

[0093] According to some embodiments, the reference thinking information is text information, and the multimodal speech language large model is configured to: encode the second inquiry speech data into an inquiry semantic vector in a semantic vector space; generate a thinking semantic vector in the semantic vector space based on the inquiry semantic vector; generate a reply semantic vector in the semantic vector space based on the thinking semantic vector; and decode the reply semantic vector into the second reply speech data, and wherein the adjustment unit 805 includes: a decoding subunit, configured to decode the thinking semantic vector into a predicted thinking text; and a second adjustment subunit, configured to adjust the parameters of the multimodal speech language large model based on the predicted thinking text and the reference thinking information.

[0094] According to some embodiments, the reference reply voice data is marked with voice breakpoints, the second reply voice data is composed of a first voice segment and a second voice segment, and the adjustment unit 805 includes: a splitting subunit, configured to split the reference reply voice data into a first reference segment and a second reference segment based on the voice breakpoints; and a third adjustment subunit, configured to adjust the parameters of the multimodal speech language model based on the first voice segment and the first reference segment; and adjust the parameters of the multimodal speech language model based on the second voice segment and the second reference segment.

[0095] According to one aspect of the present disclosure, a speech data generating apparatus is provided. Figure 9 FIG. 1 shows a structural block diagram of a voice data generating apparatus 900 according to an exemplary embodiment of the present disclosure. Figure 9 As shown, the apparatus 900 includes:

[0096] A multimodal speech and language large model 901 obtained by training using the multimodal speech and language large model training device 800 as described above;

[0097] The second acquiring unit 902 is configured to acquire inquiry voice data from the user;

[0098] The third acquiring unit 903 is configured to acquire the reply voice data generated by the multimodal voice language model by inputting the inquiry voice data into the multimodal voice language model; and

[0099] The returning unit 904 is configured to return the reply voice data to the user.

[0100] In the technical solutions disclosed herein, the collection, storage, use, processing, transmission, provision and disclosure of user personal information involved comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0101] According to one aspect of the present disclosure, an electronic device is also provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute at least one of the above-mentioned multimodal speech language large model training method and speech data generation method.

[0102] According to one aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is also provided, wherein the computer instructions are used to enable the computer to execute at least one of the above-mentioned multimodal speech language large model training method and speech data generation method.

[0103] According to one aspect of the present disclosure, a computer program product is also provided, comprising a computer program, wherein when the computer program is executed by a processor, at least one of the above-mentioned multimodal speech language large model training method and speech data generation method is implemented.

[0104] refer to Figure 10 , a block diagram of an electronic device 1000 that can serve as a server or client of the present disclosure will now be described, which is an example of a hardware device that can be applied to various aspects of the present disclosure. The electronic device is intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0105] like Figure 10As shown, the device 1000 includes a computing unit 1001, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 1002 or a computer program loaded from a storage unit 1008 into a random access memory (RAM) 1003. Various programs and data required for the operation of the device 1000 can also be stored in the RAM 1003. The computing unit 1001, the ROM 1002, and the RAM 1003 are connected to each other via a bus 1004. An input / output (I / O) interface 1005 is also connected to the bus 1004.

[0106] Multiple components in the device 1000 are connected to the I / O interface 1005, including: an input unit 1006, an output unit 1007, a storage unit 1008, and a communication unit 1009. The input unit 1006 can be any type of device that can input information to the device 1000. The input unit 1006 can receive input digital or character information and generate key signal input related to user settings and / or function control of the electronic device, and can include but is not limited to a mouse, a keyboard, a touch screen, a trackpad, a trackball, a joystick, a microphone, and / or a remote control. The output unit 1007 can be any type of device that can present information, and can include but is not limited to a display, a speaker, a video / audio output terminal, a vibrator, and / or a printer. The storage unit 1008 can include but is not limited to a magnetic disk and an optical disk. The communication unit 1009 allows the device 1000 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks, and may include but is not limited to a modem, a network card, an infrared communication device, a wireless communication transceiver and / or a chipset, such as a Bluetooth™ device, an 802.11 device, a WiFi device, a WiMax device, a cellular communication device and / or the like.

[0107] The computing unit 1001 can be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 1001 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units that run machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 1001 performs the various methods and processes described above, such as the training method or speech data generation method of the multimodal speech language large model. For example, in some embodiments, the training method or speech data generation method of the multimodal speech language large model can be implemented as a computer software program, which is tangibly included in a machine-readable medium, such as a storage unit 1008. In some embodiments, part or all of the computer program can be loaded and / or installed on the device 1000 via the ROM 1002 and / or the communication unit 1009. When the computer program is loaded into the RAM 1003 and executed by the computing unit 1001, one or more steps of the training method or speech data generation method of the multimodal speech language large model described above can be performed. Alternatively, in other embodiments, the computing unit 1001 may be configured to execute a method for training a large multimodal speech language model or a method for generating speech data in any other appropriate manner (e.g., by means of firmware).

[0108] Various embodiments of the systems and techniques described above can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), system-on-chip systems (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system that includes at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0109] The program code for implementing the method of the present disclosure can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device so that when the program code is executed by the processor or controller, the functions / operations specified in the flow chart and / or block diagram are implemented. The program code can be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0110] In the context of the present disclosure, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in conjunction with an instruction execution system, device or equipment. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium can include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0111] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0112] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), the Internet, and a blockchain network.

[0113] A computer system may include a client and a server. The client and server are generally remote from each other and typically interact through a communication network. The client-server relationship arises through computer programs running on the respective computers and having a client-server relationship with each other. The server may be a cloud server, a server in a distributed system, or a server integrated with a blockchain.

[0114] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this disclosure can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved. This is not limited herein.

[0115] Although the embodiments or examples of the present disclosure have been described with reference to the accompanying drawings, it should be understood that the above-mentioned methods, systems and devices are merely exemplary embodiments or examples, and the scope of the present invention is not limited by these embodiments or examples, but is only limited by the claims after authorization and their equivalents. Various elements in the embodiments or examples may be omitted or replaced by their equivalents. In addition, the steps may be performed in an order different from that described in this disclosure. Further, the various elements in the embodiments or examples may be combined in various ways. It is important that as technology evolves, many of the elements described herein may be replaced by equivalent elements that appear after this disclosure.

Claims

1. A method for training a large multimodal speech and language model, comprising: Inputting first inquiry voice data into the multimodal voice language model to obtain first reply voice data generated by the multimodal voice language model; Determining a query text corresponding to the first query voice data and a reply text corresponding to the first reply voice data; Determining a first score based on the inquiry text and the reply text; determining a second score based on voice features of the first inquiry voice data and voice features of the first reply voice data, wherein the voice features include at least one of voice clarity, speech rate, timbre, intonation, and emotion; and Based on the first score and the second score, parameters of the multimodal speech language large model are adjusted.

2. The method according to claim 1, wherein Determining the second score based on the voice features of the first inquiry voice data and the voice features of the first reply voice data includes: In response to determining that the first inquiry voice data includes description information for voice features of the first reply voice data, the second score is determined based on the description information, the voice features of the first inquiry voice data, and the voice features of the first reply voice data.

3. The method according to claim 1 or 2, wherein Determining the second score based on the voice features of the first inquiry voice data and the voice features of the first reply voice data includes: Determining the identity characteristics of the speaker based on the voice characteristics of the first inquiry voice data; and The second score is determined based on the identity feature, the reply text, and a voice feature of the first reply voice data.

4. The method according to any one of claims 1 to 3, wherein Determining the second score based on the voice features of the first inquiry voice data and the voice features of the first reply voice data includes: The second score output by the question-answering evaluation model is determined by inputting the first inquiry voice data and the first reply voice data into a question-answering evaluation model, wherein the question-answering evaluation model is trained using the first sample inquiry voice data, the first sample reply voice data and the reference score.

5. The method according to any one of claims 1 to 4, wherein The adjusting the parameters of the multimodal speech language model based on the first score and the second score includes: determining reward information based on the first score and the second score; and Based on the reinforcement learning strategy corresponding to the reward information, the parameters of the multimodal speech language model are adjusted.

6. The method according to any one of claims 1 to 5, further comprising: Inputting the second inquiry voice data into the multimodal voice language model to obtain second reply voice data generated by the multimodal voice language model; Obtaining reference reply voice data corresponding to the second inquiry voice data; as well as Based on the second reply voice data and the reference reply voice data, the parameters of the multimodal speech language model are adjusted.

7. The method according to claim 6, wherein: The multimodal speech language model generates the second reply speech data in the following manner: generating predicted thought information based on the second inquiry voice data, wherein the predicted thought information includes descriptive information of voice features of the second reply voice data; and generating the second reply voice data based on the predicted thinking information, And wherein, the method further comprises: Obtaining reference thinking information corresponding to the second inquiry voice data; and Based on the predicted thought information and the reference thought information, parameters of the multimodal speech language large model are adjusted.

8. The method of claim 7, wherein: The reference thought information is text information, and generating predicted thought information based on the second inquiry voice data includes: Encoding the second inquiry speech data into an inquiry semantic vector in a semantic vector space; and Generating a thought semantic vector in the semantic vector space based on the query semantic vector, wherein generating the second reply voice data based on the predicted thought information includes: generating a reply semantic vector in the semantic vector space based on the thought semantic vector; and Decoding the reply semantic vector into the second reply voice data, And wherein, adjusting the parameters of the multimodal speech language model based on the predicted thinking information and the reference thinking information includes: Decoding the thought semantic vector into a predicted thought text; and Based on the predicted thought text and the reference thought information, the parameters of the multimodal speech language model are adjusted.

9. The method according to any one of claims 6 to 8, wherein The reference reply voice data is marked with voice breakpoints, and the second reply voice data is composed of a first voice segment and a second voice segment. And wherein, adjusting the parameters of the multimodal speech language model based on the second reply voice data and the reference reply voice data includes: Splitting the reference reply voice data into a first reference segment and a second reference segment based on the voice breakpoint; Adjusting parameters of the multimodal speech language model based on the first speech segment and the first reference segment; and Adjust parameters of the multimodal speech language model based on the second speech segment and the second reference segment.

10. A method for generating speech data, comprising: Obtaining inquiry voice data from users; Inputting the inquiry voice data into a multimodal speech language model trained using the method according to any one of claims 1 to 9 to obtain reply voice data generated by the multimodal speech language model; and The reply voice data is returned to the user.

11. A training device for a large multimodal speech language model, comprising: A first acquisition unit is configured to acquire first reply voice data generated by the multimodal voice language model by inputting the first inquiry voice data into the multimodal voice language model; a first determining unit configured to determine an inquiry text corresponding to the first inquiry voice data and a reply text corresponding to the first reply voice data; a second determining unit, configured to determine a first score based on the inquiry text and the reply text; a third determining unit configured to determine a second score based on voice features of the first inquiry voice data and voice features of the first reply voice data, wherein the voice features include at least one of voice clarity, speech rate, timbre, intonation, and emotion; as well as An adjustment unit is configured to adjust parameters of the multimodal speech language large model based on the first score and the second score.

12. A voice data generating device, comprising: A large multimodal speech and language model trained using the apparatus according to claim 11; A second acquiring unit is configured to acquire inquiry voice data from a user; A third acquisition unit is configured to obtain response voice data generated by the multimodal voice language model by inputting the inquiry voice data into the multimodal voice language model; as well as The returning unit is configured to return the reply voice data to the user.

13. An electronic device comprising: at least one processor; as well as a memory communicatively coupled to the at least one processor; in The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 10.

14. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to make a computer execute the method according to any one of claims 1-10.

15. A computer program product comprising a computer program, wherein The computer program implements the method according to any one of claims 1 to 10 when executed by a processor.

Images