Voice generation method and device, electronic equipment, medium and outbound system

Abstract

The invention provides a voice generation method and device, electronic equipment, a medium and an outbound system. The voice generation method comprises the following steps: acquiring a to-be-synthesized text and preset voice information; fusing the to-be-synthesized text and the voice information based on an encoder by using the trained voice generation network to obtain an encoding information fusion result; and integrating the coding information fusion result based on the large model by using the trained voice generation network so as to generate voice and output the voice. According to the method, the to-be-synthesized text and the voice information are integrated by using the encoder, and the information of different dimensions can be fused in the step so as to integrate the information of different dimensions into the same feature distribution space, so that the accuracy is improved, and the humanoid degree of subsequent voice synthesis is improved; meanwhile, a large model is adopted for voice synthesis, and the synthesized voice is higher in humanoid degree and more vivid; and in combination with a real-time generation technology, the voice generation efficiency is high, and the speed is high.

Description

Technical Field

[0001] The present invention relates to the field of speech generation technology, and in particular to a speech generation method and device, electronic equipment, medium, and an outbound call system. Background Art

[0002] Text-to-Speech (TTS) is a technology that converts text information into spoken language. It involves multiple fields such as computer science, linguistics, acoustics, and digital signal processing. The goal of speech synthesis technology is to enable computers to "speak" naturally like humans. Common speech synthesis technologies usually use deep learning models, which can learn large amounts of text and voice data to generate more natural and realistic speech. These technologies are widely used, including virtual assistants, audiobooks, automatic news broadcasts, language learning software, and assistive devices for people with disabilities. With the advancement of technology, the naturalness and flexibility of speech synthesis continue to improve, making synthesized speech increasingly difficult to distinguish from natural human speech.

[0003] Current speech generation technology requires encoding various information to be generated, which is then fed into an audio synthesizer to output the synthesized audio. Existing speech generation technology requires encoding each piece of information to be generated in the model, which is a complex and lengthy process. If a certain part of the process does not work well, the overall model effect will be damaged. Summary of the invention

[0004] In view of the above defects in the prior art, the present invention provides a speech generation method and device, electronic equipment, medium, and outbound call system to solve the technical problems in the prior art that the speech generation process is complicated and lengthy and the model effect is poor.

[0005] To achieve the above-mentioned purpose and other related purposes, the present invention provides a speech generation method, comprising: obtaining a text to be synthesized and preset speech information; using a trained speech generation network to fuse the text to be synthesized and the speech information based on an encoder to obtain a fusion result of the encoding information; using the trained speech generation network to integrate the fusion result of the encoding information based on a large model to generate speech and output it; wherein a bidirectional streaming technology is used to simultaneously obtain the text to be synthesized and output the generated speech.

[0006] In one embodiment of the present invention, obtaining the text to be synthesized includes: obtaining the current characters one by one from the module that generates text data to obtain the text to be synthesized; using the trained speech generation network to integrate the encoding information fusion results based on the large model to generate speech and output, including: using the trained speech generation network to integrate the encoding information fusion results based on the large model to generate speech for each of the current characters and output.

[0007] In one embodiment of the present invention, obtaining the text to be synthesized includes: obtaining current characters one by one from a module that generates text data; and obtaining the text to be synthesized based on characters that have been obtained in a sentence where the current characters are located.

[0008] In one embodiment of the present invention, the trained speech generation network is used to integrate the encoding information fusion result based on a large model to generate speech and output it, including: using the trained speech generation network to integrate the encoding information fusion result based on a large model to generate a first speech; judging whether the current text is the first text of the sentence where the current text is located: if yes, outputting the first speech; if not, removing the speech output by the speech generation network last time from the first speech to obtain a second speech and output it.

[0009] In one embodiment of the present invention, the preset voice information includes volume, speaking speed, emotion and timbre.

[0010] In one embodiment of the present invention, the speech generation network is trained according to the following steps: generating a number of information pairs, wherein the information pairs consist of text to be synthesized and speech information; recording the real speech of the information pairs by manual recording; obtaining sample data based on the information pairs and their corresponding real speech; constructing a speech generation network, wherein the speech generation network includes an information fusion module and a speech synthesis module; and using the sample data to train the speech generation network to obtain the trained speech generation network.

[0011] To achieve the above-mentioned purpose and other related purposes, the present invention also provides a speech generation device, including: a data acquisition unit, used to acquire the text to be synthesized and the preset speech information; a fusion unit, used to use the trained speech generation network to fuse the text to be synthesized and the speech information based on the encoder to obtain the encoding information fusion result; and a generation unit, used to use the trained speech generation network to integrate the encoding information fusion result based on a large model to generate speech and output it.

[0012] To achieve the above-mentioned purpose and other related purposes, the present invention also provides an electronic device, including a processor, a memory and a communication bus; the communication bus is used to connect the processor and the memory; the processor is used to execute a computer program stored in the memory to implement a method provided in any one of the above embodiments.

[0013] To achieve the above-mentioned object and other related objects, the present invention further provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program is used to enable a computer to execute the method provided in any one of the above-mentioned embodiments.

[0014] To achieve the above-mentioned purpose and other related purposes, the present invention also provides an outbound call system, comprising: a speech recognition module, used to recognize the voice of the called user to convert it into first text data; a text analysis module, used to analyze the first text data to generate second text data of response; and a speech generation device, used to obtain the text to be synthesized from the second text data generated by the text analysis module, and generate a speech reply to the called user according to the speech generation method as described in any of the above items.

[0015] Beneficial effects of the present invention: The present invention proposes a speech generation method and device, electronic device, medium, and outbound call system. The method uses an encoder to integrate the text to be synthesized and the speech information. This step can fuse information of different dimensions to integrate the information of different dimensions into the same feature distribution space, thereby improving the accuracy and thus improving the degree of anthropomorphism of subsequent speech synthesis. At the same time, a large model is used for speech synthesis, and the synthesized speech has a higher degree of anthropomorphism and is more realistic. Combined with real-time generation technology, the speech generation efficiency is high and the speed is fast. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0017] Figure 1 A flowchart of a speech generation method provided by an embodiment of the present invention;

[0018] Figure 2 A framework diagram of a speech generation network provided by an embodiment of the present invention;

[0019] Figure 3 A schematic diagram of an information fusion module provided by an embodiment of the present invention;

[0020] Figure 4 A flowchart of a first speech generation method provided by an embodiment of the present invention;

[0021] Figure 5 A schematic diagram of a first type of real-time speech generation provided by an embodiment of the present invention;

[0022] Figure 6 A second flow chart for obtaining text provided by an embodiment of the present invention;

[0023] Figure 7 A second flow chart for generating speech provided by an embodiment of the present invention;

[0024] Figure 8 A schematic diagram of a second real-time speech generation method provided by an embodiment of the present invention;

[0025] Fig. 9 A schematic diagram of manually recording speech provided by an embodiment of the present invention;

[0026] Fig.10 A schematic diagram of a speech generating device provided by an embodiment of the present invention;

[0027] Fig.11 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention;

[0028] Fig.12 A schematic diagram of an outbound calling system provided in one embodiment of the present invention.

[0029] Explanation of the reference numerals: 100, speech recognition module; 200, text analysis module; 300, speech generation device; 301, data acquisition unit; 302, fusion unit; 303, generation unit; 401, processor; 402, memory. DETAILED DESCRIPTION

[0030] The following describes the embodiments of the present invention through specific specific embodiments, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. It should be noted that the following embodiments and the features in the embodiments can be combined with each other without conflict. In addition to the specific methods, equipment, and materials used in the embodiments, according to the technical personnel in this technical field's grasp of the prior art and the records of the present invention, any methods, equipment, and materials of the prior art that are similar or equivalent to the methods, equipment, and materials described in the embodiments of the present invention can also be used to implement the present invention.

[0031] It should be understood that the terms used in the embodiments of the present invention are for describing specific embodiments rather than for limiting the scope of protection of the present invention. Unless otherwise defined, all technical and scientific terms used in the present invention have the same meaning as those generally understood by those skilled in the art.

[0032] In the following description, numerous details are discussed to provide a more thorough explanation of the embodiments of the present invention. However, it is obvious to those skilled in the art that the embodiments of the present invention can be implemented without these specific details. In some of the embodiments, well-known structures and devices are shown in the form of block diagrams rather than in detail to avoid making the embodiments of the present invention difficult to understand.

[0033] The flowchart and block diagram in the accompanying drawings illustrate the system architecture, functions and operations that may be implemented according to the methods and computer program products of various embodiments disclosed in the present invention. In this regard, each box in the flowchart or block diagram can represent a module, a program segment, or a part of a code, and the module, the program segment, or a part of the code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or the flowchart, and the combination of the boxes in the block diagram and / or the flowchart can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0034] See also Figure 1 , Figure 1 A speech generation method provided by an embodiment of the present invention includes steps S100 to S300.

[0035] Step S100, obtaining text to be synthesized and preset voice information. The text to be synthesized is the text that needs to generate voice, which can be a character, a word, a sentence, or a paragraph, and can be generated by other modules, such as the text analysis module described later.

[0036] In a specific embodiment of the present invention, the preset voice information includes volume, speaking speed, emotion and timbre.

[0037] Volume refers to the loudness of the output sound, that is, the strength of the sound. In speech synthesis, the volume can be controlled by adjusting the amplitude of the synthesized speech.

[0038] Speech rate refers to the number of syllables or words spoken per unit time, usually measured in syllables or words per minute. In speech synthesis, speech rate can be controlled by adjusting the time interval between phonemes or words, thus affecting the listener's understanding of the information and grasp of the semantics.

[0039] Timbre refers to the characteristics of a sound that make it possible to distinguish different sounds even when they have the same volume, pitch, and intensity. Timbre is determined by the spectrum of the sound, including the fundamental frequency (pitch) and the harmonic structure of the sound. In speech synthesis, timbre can be controlled by adjusting the Mel-spectrogram or the parameters of the vocoder to simulate the characteristics of different people's voices.

[0040] Step S200: using the trained speech generation network, based on the encoder, fuse the text to be synthesized and the speech information to obtain the encoding information fusion result.

[0041] See also Figure 2 In a specific embodiment of the present invention, the speech generation network may include, for example, an information fusion module and a speech synthesis module. For example, the text to be synthesized is "Hello, I am a Huolala customer service representative." We hope that the speech information of the synthesized speech is as follows: low volume, slow speaking speed, happy emotion, and the timbre is Zhang San. The information fusion module fuses the text to be synthesized with the speech information, encodes the speech information of all dimensions together, and combines it with the text to be synthesized. The combined information is then sent as input to the speech synthesis module, which integrates all the information to generate a super-humanized voice that meets our speech information requirements.

[0042] See also Figure 3 In a specific embodiment of the present invention, the encoder in step S200 can be a Transformer encoder. Taking volume, speech rate, emotion, timbre and text to be synthesized as examples, they are input into the Transformer encoder together, and the voice information is spliced ​​with the text to be synthesized to obtain the input. The purpose of this step is to combine all the input information to prevent information loss caused by separate processing. In the Transformer encoder, various information is integrated, and the voice information such as volume, speech rate, emotion, timbre and the text to be synthesized are fused to obtain the coded information fusion result. This step fuses information of different dimensions and integrates information of different dimensions into the same feature distribution space to improve accuracy, thereby improving the degree of anthropomorphism of subsequent speech synthesis.

[0043] Step S300: Using the trained speech generation network, the coding information fusion results are integrated based on the large model to generate speech and output it. In the field of artificial intelligence and machine learning, "large models" usually refer to models with a large number of parameters, which usually require a large amount of data and computing resources to train. Large models usually contain millions or even billions of parameters, which learn patterns and relationships in the data during the training process. Due to the large number of parameters, large models can capture and learn complex patterns and nuances in the data. Compared with traditional small models, large models have stronger speech synthesis capabilities, and the synthesized speech is more humanized and more realistic.

[0044] When performing the above steps S100 - S300, the present invention uses a two-way streaming technology to simultaneously obtain the text to be synthesized and the generated output speech. In traditional speech synthesis technologies, the text to be synthesized is obtained first, then speech synthesis is performed on this text to be synthesized, and after synthesis, it is output. Only after the output is completed is the next text to be synthesized obtained. This will result in the inability to process the text to be synthesized during the speech synthesis process. Therefore, in this embodiment, a two-way streaming technology is adopted. The two-way streaming speech synthesis technology is a speech synthesis method that allows word segmentation and sentence breaking of the input text, local synthesis of speech features and audio by an acoustic model and a vocoder, and can output the synthesized audio in segments. This technology can reduce the waiting time for speech synthesis output, thereby enhancing the overall interaction experience. Two-way streaming speech synthesis is particularly suitable for speech interaction scenarios because it can respond faster to user input and generate speech output in real time.

[0045] Please refer to Figure 4 and Figure 5 , in a specific embodiment of the present invention, obtaining the text to be synthesized includes: S101. Sequentially obtain the current character from the module that generates text data to obtain the text to be synthesized. Step S300 includes: S301. Use the trained speech generation network to generate and output the speech corresponding to each current character based on the integrated coding information fusion result of the large model. In this embodiment, obtaining characters one by one constitutes the text to be synthesized. In this way, the speech generation network only needs to synthesize the speech corresponding to one character each time, and its processing speed is significantly faster.

[0046] Please refer to Figure 6 , in a specific embodiment of the present invention, obtaining the text to be synthesized includes steps S111 and S112.

[0047] Step S111. Sequentially obtain the current character from the module that generates text data; for example, if the module that generates text data intends to output "Hello, sir!", the first time, the current character obtained is the character "先 (xiān)", and the second time, the current character obtained is the character "生 (shēng)", and so on.

[0048] Step S112. Obtain the text to be synthesized based on the characters that have been obtained in the sentence where the current character is located. For the character "先 (xiān)", the characters that have been obtained in the sentence where it is located are only itself, so the corresponding text to be synthesized is "先 (xiān)"; for the character "生 (shēng)", the characters that have been obtained in the sentence where it is located are the two characters "先 (xiān)" and "生 (shēng)", and its corresponding text to be synthesized is "先生 (xiān shēng)", and so on.

[0049] The advantage of doing this is that the text to be synthesized will continuously change, including the previous characters, so that the subsequent text will be more natural and less rigid when generating speech.

[0050] Please refer to Figure 7and Figure 8 In a specific embodiment of the present invention, step S300 includes step S311 and step S312.

[0051] Step S311: Using the trained speech generation network, based on the integrated coding information fusion result of the large model, to generate the first speech. For example, the word "first" or "Mr." above, the speech generated according to them is called the first speech.

[0052] Step S312: Determine whether the current character is the first character of the sentence where the current character is located. If so, output the first speech. If not,剔除 the speech output by the speech generation network last time from the first speech, obtain the second speech and output it. After such processing, the speech of the word "first" will be normally output. After the speech of "Mr." is synthesized for the second time, the word "first" will be剔除, and only the speech of the word "sheng" will be output. In this way, the two output speeches will sound corresponding to the normal input "Mr.", but its speed is faster (so the corresponding speech will be output immediately after receiving the word "first"), and at the same time, it can also take into account the tone of the whole sentence when generating the speech of the subsequent characters, making it more natural.

[0053] Specifically, when剔除 the last output speech from the first speech, it can be剔除 according to time or according to the pronunciation gap of each character.

[0054] In a specific embodiment of the present invention, the speech generation network is trained according to the following steps: (1) Generate a number of information pairs, and the information pair consists of the text to be synthesized and speech information; (2) Adopt the method of manual recording to record the real speech of the information pair; Fig. 9 It shows how to record the corresponding real speech (the audio is shown in the figure) according to the information pair (the speech information and text are shown in the figure); (3) According to the information pair and its corresponding real speech, obtain the sample data; (4) Construct the speech generation network, and the speech generation network includes an information fusion module and a speech synthesis module; (5) Use the sample data to train the speech generation network to obtain the trained speech generation network.

[0055] The information pair in the above step (1) can be obtained, for example, by randomly matching the preset text to be synthesized and the preset speech information.

[0056] When performing manual recording in the above step (2), it is necessary to consider the recording environment, recording equipment and recording requirements to ensure that the recorded real speech is more standardized as much as possible.

[0057] It should be noted that the Chinese character "剔除" in the original text seems to be a wrong input. It might be "剔除" (remove). This translation is based on the understanding of the overall context. If there are any inaccuracies, please provide more correct information.The choice of recording environment mainly considers reducing noise and reverberation. You can use a professional recording studio for recording. Try to record in the same acoustic environment and complete the full recording at one time to ensure that the recording environment is the same. Avoid recording in a far-field environment. For example, you can be 1-2 fists away from the microphone to keep the position and distance stable to prevent inconsistent volume. Do not let the airflow of the speech face the microphone to prevent noise caused by microphone spray. When choosing a recording device, you should use a wired microphone with good sound reception. Recording supports pcm, m4a, mp3, wav, ogg, aac, ogg_opus formats. Ensure that the sampling rate is ≥24k. When exporting audio, do not add any equalization, compression, noise gate or other hardware or plug-in processing.

[0058] When recording, the pronunciation should be clear and accurate; the recording should be in a natural conversation state, and avoid deliberately lowering the voice, etc. The posture should be kept stable, and artificial noises such as the sound of the chair, the friction of clothes, and the tapping of the mouse and keyboard should be avoided. Before recording, the recording staff should be familiar with the copy and perform according to the determined style (volume, speed, emotion). The recording time can be set as needed, such as recording 30 minutes to 1 hour of audio.

[0059] It should be noted that the step division of the above methods is only for clear description. When implemented, they can be combined into one step or some steps can be split and decomposed into multiple steps. As long as they contain the same logical relationship, they are all within the scope of protection of this patent; adding insignificant modifications to the algorithm or process or introducing insignificant designs without changing the core design of the algorithm and process are all within the scope of protection of this patent.

[0060] See also Fig.10 , Fig.10 A speech generation device provided in one embodiment of the present invention includes a data acquisition unit 301, a fusion unit 302 and a generation unit 303. The data acquisition unit 301 is used to acquire a text to be synthesized and preset speech information; the fusion unit 302 is used to use a trained speech generation network to fuse the text to be synthesized and the speech information based on an encoder to obtain a fusion result of the encoding information; the generation unit 303 is used to use a trained speech generation network to integrate the fusion result of the encoding information based on a large model to generate speech and output it.

[0061] It should be noted that the speech generation device of the present embodiment is a device corresponding to the above-mentioned speech generation method, and the functional modules in the speech generation device may correspond to the corresponding steps in the speech generation method. The speech generation device of the present embodiment can be implemented in conjunction with the speech generation method, that is, in the absence of conflict, the relevant technical details mentioned in the speech generation method of the above-mentioned embodiment can also be applied to the speech generation device of the present embodiment.

[0062] See also Fig.11 , Fig.11 An electronic device provided for one embodiment of the present invention includes a processor 401, a memory 402 and a communication bus; the communication bus is used to connect the processor 401 and the memory 402; the processor 401 is used to execute the computer program stored in the memory 402 to implement the above-mentioned speech generation method.

[0063] The above-mentioned electronic device is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0064] The electronic device mentioned above can be any electronic product that can perform human-computer interaction with a user, such as a personal computer, a tablet computer, a smart phone, a personal digital assistant (PDA), a game console, an interactive network television (IPTV), a smart wearable device, etc.

[0065] The electronic device may also include a network device and / or a user device, wherein the network device includes, but is not limited to, a single network server, a server group consisting of multiple network servers, or a cloud consisting of a large number of hosts or network servers based on cloud computing.

[0066] The network where the above electronic devices are located includes but is not limited to the Internet, wide area network, metropolitan area network, local area network, virtual private network (VPN), etc.

[0067] The processor 401 may be, for example, a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components; the memory may include a random access memory (RAM), and may also include a non-volatile memory, such as at least one disk storage.

[0068] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored. The computer program is used to enable a computer to execute the above-mentioned speech generation method.

[0069] See also Fig.12 , Fig.12 An outbound call system provided in one embodiment of the present invention includes a speech recognition module 100, a text analysis module 200 and a speech generation device 300. The speech recognition module 100 is used to recognize the speech of the called user to convert it into first text data; the text analysis module 200 is used to analyze the first text data to generate second text data of the answer; the speech generation device 300 is used to obtain the text to be synthesized from the second text data generated by the text analysis module 200, and generate a speech reply to the called user according to the above speech generation method. The outbound call system can be used in any call system scenario, such as customer service, sales promotion, marketing, market research, etc.

[0070] In general, the present invention uses encoders to fuse various information codes to simplify the process and improve efficiency; uses a large model to fuse various information codes, which improves efficiency while greatly improving the degree of humanization of the generated speech; and also uses bidirectional streaming real-time generation technology to generate speech while generating text, thereby improving timeliness.

[0071] The above embodiments are merely illustrative of the principles and effects of the present invention, and are not intended to limit the present invention. Anyone familiar with the art may modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by a person of ordinary skill in the art without departing from the spirit and technical concept disclosed by the present invention shall still be covered by the claims of the present invention.

Claims

1. A speech generation method, characterized in that: include: Obtain the text to be synthesized and the preset voice information; Using the trained speech generation network, based on the encoder, the text to be synthesized and the speech information are fused to obtain a coded information fusion result; Using the trained speech generation network, integrating the coded information fusion results based on a large model to generate speech and output it; Among them, a bidirectional streaming technology is used to simultaneously obtain the text to be synthesized and output the generated speech.

2. The speech generation method according to claim 1, characterized in that: Get the text to be synthesized, including: Obtaining the current characters one by one from the module generating text data to obtain the text to be synthesized; Using the trained speech generation network, integrating the coded information fusion result based on the large model to generate speech and output it, including: The trained speech generation network is utilized to integrate the encoding information fusion results based on a large model to generate and output the speech of each current text.

3. The speech generation method according to claim 1, characterized in that: Get the text to be synthesized, including: Get the current text one by one from the module that generates text data; The text to be synthesized is obtained according to the characters acquired in the sentence where the current characters are located.

4. The speech generation method according to claim 3, characterized in that: Using the trained speech generation network, integrating the coded information fusion result based on the large model to generate speech and output it, including: Using the trained speech generation network, integrating the encoding information fusion result based on a large model to generate a first speech; Determine whether the current character is the first character of the sentence where the current character is located: If yes, output the first speech; If not, the speech outputted last by the speech generation network is removed from the first speech to obtain the second speech and output it.

5. The speech generation method according to claim 1, characterized in that: The preset voice information includes volume, speaking speed, emotion and timbre.

6. The speech generation method according to claim 1, characterized in that: The speech generation network is trained as follows: Generate a plurality of information pairs, wherein the information pairs consist of text to be synthesized and speech information; Using manual recording to record the real voice of the information pair; Obtaining sample data according to the information pair and the corresponding real speech; Constructing a speech generation network, wherein the speech generation network includes an information fusion module and a speech synthesis module; The speech generation network is trained using the sample data to obtain the trained speech generation network.

7. A speech generating device, characterized in that: include: A data acquisition unit, used to acquire the text to be synthesized and preset voice information; A fusion unit, used to use the trained speech generation network to fuse the text to be synthesized and the speech information based on the encoder to obtain a fusion result of the encoding information; as well as A generation unit is used to utilize the trained speech generation network to integrate the encoding information fusion result based on a large model to generate speech and output it.

8. An electronic device, characterized in that: The system comprises a processor, a memory and a communication bus; the communication bus is used to connect the processor and the memory; the processor is used to execute a computer program stored in the memory to implement the method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and the computer program is used to make a computer execute the method according to any one of claims 1 to 6.

10. An outbound calling system, characterized in that: include: A speech recognition module, used for recognizing the speech of the called user to convert it into first text data; a text analysis module, configured to analyze the first text data to generate second text data of a response; and A speech generating device is used to obtain the text to be synthesized from the second text data generated by the text analyzing module, and generate a speech reply to the called user according to the speech generating method according to any one of claims 1 to 6.

Images