TWI833678B - Generative chatbot system for real multiplayer conversational and method thereof - Google Patents

Abstract

A generative chatbot system for real multiplayer conversational and method thereof is disclosed. By detecting multiple voices signals and convert them into corresponding feature vectors and text messages, and embedding a chronological tag and a classification tag in the text messages to store as contextual information, so as to allow the server host to determine the sequential logic of multiplayer conversations. Then, send the contextual information and the sequential logic to the artificial intelligence device to help it determine the current conversation stage, evolving topics, and predict conversation development. Subsequently, proactively generate at least one response message and store on the server host. The server host filters appropriate response message to be sent to a portable device for output. The mechanism is help to improve the conversational initiative and response efficiency.

Description

Generative chat robot system and method in real multi-person response situation

The present invention relates to a chat robot system and a method thereof, in particular to a generative chat robot system and a method thereof in a real multi-person response situation.

In recent years, with the popularization and vigorous development of artificial intelligence, various artificial intelligence applications have mushroomed. Among them, chatbots attract the most attention.

Generally speaking, traditional chatbots usually have one-on-one conversations with users. That is to say, when users send questions, the chatbot responds based on the questions. However, there is currently no chatbot that can proactively give appropriate response suggestions or prompts in a real multi-person response situation. For example, in a multi-person conversation environment, traditional chatbots cannot proactively and quickly provide users with appropriate response suggestions or tips. Suitable conversation suggestions. Therefore, they have the initiative to chat and answer questions with poor efficiency.

To sum up, it can be seen that the previous technology has had problems of poor chat initiative and poor response efficiency for a long time. Therefore, it is necessary to propose improved technical means to solve this problem.

The invention discloses a system and method of a generative chat robot in a real multi-person response situation.

First, the present invention discloses a system for generating chat robots in a real multi-person response situation. This system includes: an artificial intelligence device, a portable device and a server host. Among them, the artificial intelligence device is used to receive context information and its corresponding timing logic through an application programming interface (Application Programming Interface, API), and input it into a large language model (Large Language Model, LLM) to generate a response message, and then through This API sends the response message. The portable device includes: a sensor, a speaker, a storage device and a voice processor. Among them, the sensor is used to continuously sense multiple voice signals; the speaker is used to output feedback voice; the storage device is used to store multiple feature vectors corresponding to the voice signals and multiple corresponding text messages. Each text message includes a timing mark. and a classification mark; and a voice processor electrically connected to the sensor, speaker and storage device, the voice processor is configured to: use Mel-Frequency Cepstral Coefficients (MFCC) to convert the sensed The voice signal is converted into the corresponding feature vector to classify the voice signal; speech-to-text processing is performed to convert the voice signal into corresponding text messages respectively; based on the timing relationship and classification results, embedding timing marks and classification marks in the text messages corresponding to the voice signals and storing them in the storage device as the context messages; and when receiving the on-demand dialogue message, executing text-to-speech (Text-to-Speech) Processing, converting this on-demand dialogue message into feedback speech for output through the speaker. Then, the server host is connected to the artificial intelligence device and the portable device. The server host includes: a non-transitory computer-readable storage medium and a hardware processor. Wherein, the non-transitory computer-readable storage medium is used to store a plurality of computer-readable instructions; and the hardware processor is electrically connected to the non-transitory computer-readable storage medium and is used to execute a plurality of computer-readable instructions. , causing the server host to execute: continuously load context information from the storage device of the portable device, and determine the timing logic of multi-person conversations based on the timing marks and classification marks embedded in the text message. This timing logic includes the number of people in the conversation, the timing and topic; transmit context information and timing logic to the artificial intelligence device, and receive corresponding response messages from the artificial intelligence device to store in the response list; and automatically select at least one of the response messages from the response list as a subsequent Select a conversation message and send the on-demand conversation message to the portable device.

In addition, the present invention also discloses a method for generating a chat robot in a real multi-person response situation. The steps include: connecting the server host to an artificial intelligence device and a portable device respectively, wherein the artificial intelligence device passes an application program The interface receives context messages and their corresponding timing logic, and transmits response messages; the portable device continuously senses multiple voice signals through the sensor, and converts the sensed voice signals into corresponding The feature vector is used to classify the speech signal; the portable device performs speech-to-text processing to convert the speech signals into corresponding text messages; the portable device converts the corresponding speech signal based on the timing relationship and classification results. The timing mark and classification mark are embedded in the text message and stored in the storage device of the portable device as context information; the server host continues to load the context information from the storage device of the portable device, and based on the timing mark and classification embedded therein Tags determine the timing logic of multi-person conversations. The timing logic includes the number of people, timing and topics of the conversation; the server host transmits contextual information and timing logic to the artificial intelligence device for input to a large language model of the artificial intelligence device to generate The corresponding response message is then sent to the server host through the application program interface; the server host stores the response message in the response list, and automatically selects at least one of the response messages from the response list as on-demand conversation message, and then transmit the on-demand conversation message to the portable device; and when the portable device receives the on-demand conversation message, it performs text-to-speech processing to convert the on-demand conversation message into feedback voice for use through the speaker output.

The system and method disclosed by the present invention are as above. The difference from the prior art is that the present invention detects multiple speech signals to convert them into corresponding feature vectors and text messages, and embeds timing marks and classification marks in the text messages for storage. It is contextual information that allows the server host to determine the timing logic of multi-person conversations, and then transmits the contextual information and timing logic to the artificial intelligence device to help it determine the current conversation stage, topic evolution, and predict the development of the conversation, and then actively generate and store response messages. to the server host, and the server host filters out appropriate response messages and sends them to the portable device for output.

Through the above technical means, the present invention can achieve the technical effect of improving chat initiative and response efficiency.

The embodiments of the present invention will be described in detail below with reference to the drawings and examples, so that the implementation process of how to apply technical means to solve technical problems and achieve technical effects of the present invention can be fully understood and implemented accordingly.

First, please refer to "Figure 1". "Figure 1" is a system block diagram of a generative chat robot system in a real multi-person response situation of the present invention. This system includes: an artificial intelligence device 110, a portable device 120 and server host 130. Among them, the artificial intelligence device 110 is used to receive the context information and its corresponding timing logic through the application program interface, and input it into the large language model to generate a response message, and then transmit the response message through the application program interface. In actual implementation, the artificial intelligence device 110 is a chat robot using a large language model, such as: Generative Pre-trained Transformer (GPT), PaLM, Galactica, LLaMA, LaMDA or its analogues, and can determine the current dialogue stage, topic evolution, and predict the development of the dialogue based on contextual information and its corresponding temporal logic, and then use the predicted dialogue as a response message.

The portable device 120 includes: a sensor 121, a speaker 122, a storage device 123 and a voice processor 124. Among them, the sensor 121 is used to continuously sense multiple voice signals. In actual implementation, the sensor 121 can also sense at least one of the user's physiological state, facial expression and body movements to generate user behavior information, and transmit the user behavior information to the server host 130, which is then The host 130 determines the user's personality to set personality parameters. For example, physiological characteristics such as blood pressure, heartbeat, pulse, and blood sugar can be sensed to determine physiological states, such as happiness, excitement, depression, etc.; or facial expressions and moods can be determined by sensing faces, irises, etc. etc., in order to judge the user's personality through physiological state, facial expression and mood, such as: extroversion, introversion, enthusiasm, indifference, etc.

The speaker 122 is used to output feedback voice. In actual implementation, the speakers may include headphones, speakers, or the like. In addition, the portable device 120 may also include a display element for synchronously displaying on-demand dialogue messages on the display element when the speaker 122 outputs the feedback voice. In actual implementation, the display element may include: a display, a dot matrix light emitting diode or the like.

The storage device 123 is used to store a plurality of feature vectors corresponding to the speech signal and a plurality of corresponding text messages. Each text message includes a timing mark and a classification mark. In actual implementation, the storage device 123 may include a hard disk, an optical disk, a flash memory or the like. In addition, the storage device 123 will also use all text messages embedded with timing marks and classification marks as context messages.

The speech processor 124 is electrically connected to the sensor 121, the speaker 122 and the storage device 123. The speech processor 124 is configured to: convert the sensed speech signal into a corresponding feature vector through the Mel frequency cepstrum coefficient, using to classify the voice signals; perform speech-to-text processing to convert the voice signals into corresponding text messages; based on the timing relationship and classification results, embed timing marks and classification marks in the text messages corresponding to the voice signals and store them in the storage device 123 ; and when receiving the on-demand dialogue message, perform text-to-speech processing, convert the on-demand dialogue message into feedback voice for output through the speaker, for example: output through a wired or wireless (Bluetooth) headset, speaker or the like. feedback voice. In actual implementation, the speech processor 124 can be implemented using a processor dedicated to processing speech signals, such as a digital signal processor (Digital Signal Processing, DSP). In addition, the portable device 120 further converts the user's voice signal into a feature vector through Mel frequency cepstral coefficients and transmits it to the server host 130. The server host 130 combines the preset multiple personality characteristics with The vectors are compared to determine the user's personality, and personality parameters are set based on the judgment results.

Next, the server host 130 is connected to the artificial intelligence device 110 and the portable device 120. The server host 130 includes: a non-transitory computer-readable storage medium 131 and a hardware processor 132. Among them, the non-transitory computer-readable storage medium is used to store multiple computer-readable instructions. In actual implementation, the computer-readable instructions are executed by the server host 130, and the computer-readable instructions for performing the operations of the present invention can be combined language instructions, instruction set architecture instructions, machine instructions, machine-related instructions, micro-instructions, Firmware instructions, or source code or object code (Object Code) written in any combination of one or more programming languages, including object-oriented programming languages, such as: Common Lisp, Python, C++, Objective-C, Smalltalk, Delphi, Java, Swift, C#, Perl, Ruby and PHP, etc., as well as conventional procedural (Procedural) programming languages, such as C language or similar programming languages.

The hardware processor 133 is electrically connected to the non-transitory computer-readable storage medium 131 for executing the plurality of computer-readable instructions, causing the server host 130 to execute: continuously loading from the storage device 123 of the portable device 120 Context information, and determine the timing logic of multi-person conversations based on the timing tags and classification tags embedded in it. This timing logic includes the number of people, timing, and topics of the conversation; the context information and timing logic are transmitted to the artificial intelligence device 110, and are generated from the artificial intelligence device 110. The device 110 receives the corresponding response message and stores it in the response list; and automatically selects at least one of the response messages from the response list as an on-demand conversation message, and sends the on-demand conversation message to the portable device 120 . In actual implementation, the hardware processor 133 may be a central processing unit, a microprocessor or the like. In addition, taking the temporal logic of multi-person conversations as an example, the number of people can be judged from the number of classification tags, the order of the conversation can be judged from the timing tags, and the topic can be judged from the content of contextual messages, such as matching time and keywords. For example, Assuming that the time is noon and the keyword is "what to eat", the topic can be judged as "lunch discussion". In addition, the on-demand dialogue message can randomly select response messages that meet the personalized parameters from the response list as the on-demand dialogue message. The personalized parameters allow the portable device 120 to connect to the server host 130 for setting.

It should be noted that in actual implementation, the present invention can be implemented partially or completely based on hardware. For example, one or more components in the system can be implemented through an integrated circuit chip or a system on chip (System on Chip, SoC), Complex Programmable Logic Device (CPLD), Field Programmable Gate Array (FPGA) and other hardware processors (Hardware Processor). The non-transitory computer-readable storage medium of the present invention carries computer-readable instructions (or computer program instructions) for causing the processor to implement various aspects of the present invention. The non-transitory computer-readable storage medium can A tangible device that can hold and store instructions for use by an instruction execution device. The non-transitory computer-readable storage medium may be, but is not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: hard disks, random access memory, read-only memory, flash memory, optical disks, floppy disks, and any suitable combination of the foregoing. As used herein, non-transitory computer-readable storage media is not to be construed as transient signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., optical signals through fiber optic cables), Or electrical signals transmitted through wires. In addition, the computer-readable instructions described herein can be downloaded from a non-transitory computer-readable storage medium to various computing/processing devices, or through a network, such as the Internet, a local area network, a wide area network and/or a wireless network Download to an external computer device or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, hubs and/or gateways. A network card or network interface in each computing/processing device receives computer-readable instructions from the network and forwards the computer-readable instructions for non-transitory computer-readable storage in each computing/processing device in the media.

Please refer to "Figure 2A" and "Figure 2B". "Figure 2A" and "Figure 2B" are method flow charts of the method of the generative chat robot in a real multi-person response situation of the present invention. The steps include: The server host 130 is connected to the artificial intelligence device 110 and the portable device 120 respectively. The artificial intelligence device 110 receives the context message and its corresponding timing logic through the application program interface, and transmits the response message (step 210); The portable device 120 continuously senses multiple voice signals through the sensor, and converts the sensed voice signals into corresponding feature vectors through the Mel frequency cepstral coefficient to classify the voice signals (step 220 ); the portable device 120 performs speech-to-text processing and converts the voice signals into corresponding text messages (step 230); the portable device 120 embeds time series in the text messages corresponding to the voice signals based on the time series relationship and classification results. Mark and classify the tags and store them in the storage device 123 of the portable device 120 as context information (step 240); the server host 130 continues to load the context information from the storage device 123 of the portable device 120, and according to the timing embedded therein The tags and classification tags determine the timing logic of the multi-person conversation, which includes the number of people, timing, and topics of the conversation (step 250); the server host 130 transmits the context information and timing logic to the artificial intelligence device 110 for input to The large language model of the artificial intelligence device 110 generates a corresponding response message, and then sends the generated response message to the server host 130 through the application program interface (step 260); the server host 130 stores the response message in the response list, and automatically Select at least one of the response messages from the response list as an on-demand dialogue message, and then transmit the on-demand dialogue message to the portable device 120 (step 270); and the portable device 120 receives the on-demand dialogue message. When receiving a dialogue message, text-to-speech processing is performed to convert the on-demand dialogue message into feedback voice for output through the speaker 122 (step 280). Through the above steps, multiple voice signals can be detected and converted into corresponding feature vectors and text messages, and timing tags and classification tags can be embedded in the text messages to be stored as contextual information, so that the server host 130 can determine the multi-person conversation. sequential logic, and then transmit the contextual information and sequential logic to the artificial intelligence device 110 to help it determine the current dialogue stage, topic evolution and predict the dialogue development, and then actively generate a response message and store it in the server host 130, and by the server host 130 selects appropriate response messages and sends them to the portable device 120 for output.

The following description is provided in the form of embodiments with reference to "Fig. 3" to "Fig. 5". As shown in "Fig. 3", "Fig. 3" is a schematic diagram of a portable device applying the present invention. In actual implementation, the portable device 120 can be a smart phone 300, a voice recorder, a personal digital assistant (PDA), or other portable device with a sound collection function, which uses a sensor capable of sound collection, such as : The microphone continuously senses the voice signal of the human voice, and converts the sensed voice signal into the corresponding feature vector through MFCC technology to classify the voice signal. Taking the smartphone 300 as an example, assuming that multiple voice signals are obtained through the microphone 310 and there are three feature vectors after conversion, it means that three people may be having a conversation. Then, the smartphone 300 will perform STT processing to convert each voice signal into a corresponding text message, and embed corresponding timing tags and classification tags based on timing and classification. The timing tags can include time, date, etc.; classification The mark may contain at least one of words, numbers, and symbols to identify different persons. For example, "A" represents the first person, "B" represents the second person, and so on, or "U01" represents For the first person, "U02" represents the second person, and so on. In particular, if the microphone 310 continues to collect sound, the smartphone 300 will continue to convert it into corresponding feature vectors and text messages, and embed timing marks and classification marks for each text message, and store them in the storage All text messages in the device 123 or within a specified period (eg, within 30 minutes) are regarded as contextual messages. In this way, the server host 130 can continuously load context information from the storage device 123 and determine the timing logic of the multi-person conversation based on it, which includes the number of people, timing and topic of the conversation. Among them, the method of judging the number of people can be judged based on the number of types of classification marks. Suppose there are three classification marks representing three people; judging the timing can be based on the timing marks to judge the order of the conversation; judging the topic can be based on the content of the contextual message, targeting keywords or Judgment is made based on the frequency or time of word occurrence. For example, if multiple food or catering words are mentioned at noon, it can be judged that the topic is discussing lunch. In actual implementation, the contextual information can be displayed on the display element 301 of the smart phone 300 in a time sequence as shown in "Figure 3". In addition, the smart phone 300 can output feedback voice through the Bluetooth headset 320 .

As shown in "Figure 4", "Figure 4" is a schematic diagram of the context information and timing logic of the present invention. In actual implementation, the context information 410 may include timing marks, classification marks and text messages. Based on the context message 410, the server host 130 can determine the order of the dialogue (ie, the dialogue sequence) based on the timing mark, and can assign a unique sequence number as a partition, for example, it can be recorded as "01 -> 02 -＞03" represents the order of each text message; determine the number of people talking according to the classification mark. For example: there are three categories "A", "B" and "C", so it can be judged that the number of people talking is three; according to the keyword "lunch" ” and time point (such as noon time), the conversation topic is judged to be “discussing lunch”. At this time, the server host 130 can generate corresponding sequential logic 420 based on the above determination results. It should be noted that in actual implementation, in addition to using the above example to present the context information 410 and the timing logic 420, the two can also be integrated together, such as the context information and timing logic 430 shown in "Figure 4". In addition, when transmitting to the artificial intelligence device 110 to obtain the corresponding response message, the response message applicable to the person of this classification can be generated through the classification designation of the classification mark. For example, if you want to obtain a response message applicable to "A", you can add the requirement "Please generate a message for A to respond" when sending context information and timing logic. In this way, the artificial intelligence device 110 can return at least one corresponding response message to the server host 130 according to the context information and timing logic to store it in the response list. Even in the case of the above requirements, it can also only return Response messages that meet the above requirements, even when the conversation topic changes, use instructions to request the specified conversation topic, thereby achieving cross-topic responses. For example, when sending contextual information and timing logic, add "in the conversation topic" at the same time. "Generate a message for A to respond under the premise of M", where M represents different conversation topics, such as discussing lunch, discussing drinks, etc., so as to allow the user to specify a certain conversation topic for prompts and responses. In actual implementation, the above requirements can be input or set through the portable device 120, such as through voice input or typing in text, numbers, symbols, etc.

As shown in "Figure 5", "Figure 5" is a schematic diagram of applying the present invention to actively filter out response messages from the response list. Assuming that there are multiple response messages in the response list 500, the server host 130 can filter out the response messages that meet the personality parameters as on-demand conversation messages. For example, assuming that the personality parameter is set to "cold", then the server host 130 When selecting a response message, response messages with extended dialogue or guided dialogue (such as containing question marks) will be excluded. For example, "I want to eat chicken chop rice" will be selected as the on-demand dialogue message, and It is transmitted to the portable device 120 to be converted into feedback voice, and then output through the speaker 122 of the portable device 120. As shown in "Figure 3", it is output through the Bluetooth headset 320 connected to the smartphone 300. In actual implementation, the personality parameters can be set by the user. The server host 130 determines the user's personality and sets it based on the user behavior information sensed by the portable device 120. The server host 130 determines the user's personality according to the default setting. After comparing multiple personality feature vectors with the user's feature vector, the user's personality is determined and set accordingly. For example, the feature vector of a deep voice can be regarded as a personality feature vector representing "coldness", and the feature vector of a high-pitched voice can be regarded as a personality feature vector representing "enthusiasm". When the user of the portable device 120, its When the feature vector of the voice matches the personality feature vector representing "coldness", the server host 130 can set the personality parameter to "coldness".

To sum up, it can be seen that the difference between the present invention and the prior art is to detect multiple voice signals to convert them into corresponding feature vectors and text messages, and to embed timing marks and classification marks in the text messages to store them as contextual information. , so that the server host can judge the timing logic of multi-person conversations, and then transmit the contextual information and timing logic to the artificial intelligence device to help it determine the current conversation stage, topic evolution and predict the development of the conversation, and then actively generate response messages and store them in the server The host, and the server host selects appropriate response messages and transmits them to the portable device for output. This technical means can solve the problems existing in the previous technology, thereby achieving the technical effect of improving chat initiative and response efficiency.

Although the present invention has been disclosed in the foregoing embodiments, they are not intended to limit the present invention. Anyone skilled in the similar art can make some modifications and modifications without departing from the spirit and scope of the present invention. Therefore, the present invention is The scope of patent protection shall be determined by the scope of the patent application attached to this specification.

110:Artificial intelligence device 120: Portable device 121: Sensor 122: Speaker 123:Storage device 124: Voice processor 130:Server host 131: Non-transitory computer-readable storage media 132:Hardware processor 300:Smartphone 301:Display component 310:Microphone 320: Bluetooth headset 410:Context message 420: Sequential logic 430:Context information and temporal logic 500:Response list Step 210: Connect a server host to an artificial intelligence device and a portable device respectively, wherein the artificial intelligence device receives a context message and its corresponding time through an application programming interface (Application Programming Interface, API). sequence logic, and transmit at least one response message Step 220: The portable device continuously senses multiple voice signals through at least one sensor, and converts the sensed voice signals into The corresponding feature vector is used to classify the speech signal Step 230: The portable device performs speech-to-text processing to convert the voice signals into corresponding text messages. Step 240: Based on the timing relationship and the classification result, the portable device embeds the timing mark and the classification mark in the text message corresponding to the voice signal and stores it in a storage device of the portable device as the contextual message Step 250: The server host continues to load the context information from the storage device of the portable device, and determines a timing logic of the multi-person conversation based on the timing mark and the classification mark embedded therein. Logic includes the number of people, timing and topics of the conversation Step 260: The server host transmits the context information and the timing logic to the artificial intelligence device for input into the large language model (Large Language Model, LLM) of the artificial intelligence device to generate the corresponding response message. , and then transmit the generated response message to the server host through the application programming interface Step 270: The server host stores the response messages in a response list, and automatically selects at least one of the response messages from the response list as an on-demand dialogue message, and then sends the on-demand dialogue message. Send to the portable device Step 280: When the portable device receives the on-demand dialogue message, it performs text-to-speech processing to convert the on-demand dialogue message into a feedback voice for output through the speaker.

Figure 1 is a system block diagram of a generative chat robot system in a real multi-person response situation of the present invention. Figure 2A and Figure 2B are method flow charts of the method of the present invention's generative chat robot in a real multi-person response situation. Figure 3 is a schematic diagram of a portable device using the present invention. Figure 4 is a schematic diagram of context information and timing logic of the present invention. Figure 5 is a schematic diagram of applying the present invention to actively filter out response messages from the response list.

110:Artificial intelligence device

120: Portable device

121: Sensor

122: Speaker

123:Storage device

124: Voice processor

130:Server host

131: Non-transitory computer-readable storage media

132:Hardware processor

Claims (10)

A system of generative chat robots in a real multi-person response situation. The system includes: the artificial intelligence device is used to receive a contextual message and its corresponding temporal logic through an application programming interface (Application Programming Interface, API). and is input to a Large Language Model (LLM) to generate at least one response message, and then transmits the response message through the application programming interface; the portable device includes: at least one sensor for continuously Sensing multiple voice signals; a speaker for outputting a feedback voice; a storage device for storing multiple feature vectors corresponding to the voice signals and corresponding multiple text messages, each of the text messages including a timing mark and a classification mark; and a speech processor electrically connected to the sensor, the speaker and the storage device, the speech processor is configured to: pass the Mel frequency cepstral coefficient (Mel -Frequency Cepstral Coefficients (MFCC) converts the sensed speech signal into the corresponding feature vector to classify the speech signal; performs speech-to-text (Speech-to-Text) processing to convert all The voice signals are respectively converted into the corresponding text messages; Based on the timing relationship and the classification result, the timing mark and the classification mark are embedded in the text message corresponding to the voice signal and stored in the storage device as the context message; and when an on-demand dialogue message is received When performing text-to-speech processing, convert the on-demand dialogue message into the feedback voice for output through the speaker; and the server host connects the artificial intelligence device and the portable device, The server host includes: a non-transitory computer-readable storage medium for storing a plurality of computer-readable instructions; and a hardware processor electrically connected to the non-transitory computer-readable storage medium for executing The plurality of computer-readable instructions cause the server host to execute: continue to load the context information from the storage device of the portable device, and determine multiple context information based on the timing mark and the classification mark embedded therein. A temporal logic of human dialogue, the temporal logic includes the number of people, timing and topics of the conversation; transmit the contextual information and the temporal logic to the artificial intelligence device, and receive the corresponding response message from the artificial intelligence device to store to a response list; and Automatically select at least one of the response messages from the response list as the on-demand conversation message, and send the on-demand conversation message to the portable device. A system for a generative chat robot in a real multi-person response situation as in request item 1, wherein the on-demand dialogue message is randomly selected from the response list and the response message that meets a parameter is used as the on-demand dialogue message. Dialog message, the personalized parameters allow the portable device to connect to the server host for setting. As claimed in Claim 1, the system of a generative chat robot in a real multi-person response situation, wherein the portable device further includes a display element for synchronously displaying the feedback voice on the display element when the speaker outputs the feedback voice. On-demand chat messages. A system for a generative chat robot in a real multi-person response situation as claimed in claim 2, wherein the sensor further includes sensing at least one of the user's physiological state, facial expressions and body movements to generate a user behavior message , and the portable device transmits the user behavior information to the server host, and the server host determines the user's personality to set the personality parameters. A system for a generative chat robot in a real multi-person response situation as claimed in claim 2, wherein the portable device further includes converting the user's voice signal into the feature vector through a mel frequency cepstrum coefficient for transmission To the server host, the server host compares the user with a plurality of preset personality feature vectors to determine the user's personality, and sets the personality parameters according to the judgment results. A method of generating chatbots in a real multi-person response situation, the steps include: A server host is connected to an artificial intelligence device and a portable device respectively, wherein the artificial intelligence device receives a context message and a corresponding timing logic through an application programming interface (Application Programming Interface, API), and transmit at least one response message; the portable device continuously senses multiple voice signals through at least one sensor, and uses Mel-Frequency Cepstral Coefficients (MFCC) to The voice signal is converted into corresponding feature vectors for classifying the voice signal; the portable device performs speech-to-text processing to convert the voice signal into a corresponding text respectively. Message; based on the timing relationship and the classification result, the portable device embeds the timing mark and the classification mark in the text message corresponding to the voice signal and stores it in a storage device of the portable device as the The context information; the server host continues to load the context information from the storage device of the portable device, and determines a timing logic of the multi-person conversation based on the timing mark and the classification mark embedded therein. Sequential logic includes the number of people, timing, and topics of the conversation; the server host transmits the context information and the temporal logic to the artificial intelligence device for input into the large language model (Large Language Model, LLM) of the artificial intelligence device. To generate the corresponding response message, and then transmit the generated response message to the server host through the application programming interface; The server host stores the response messages in a response list, and automatically selects at least one of the response messages from the response list as an on-demand dialogue message, and then sends the on-demand dialogue message to the A portable device; and when the portable device receives the on-demand dialogue message, it performs text-to-speech processing to convert the on-demand dialogue message into a feedback voice for output through a speaker. The method of request 6 for a generative chat robot in a real multi-person response situation, wherein the on-demand dialogue message is randomly selected from the response list and the response message that conforms to a parameter is used as the on-demand dialogue message. Dialog message, the personalized parameters allow the portable device to connect to the server host for setting. The method of claim 6 for a generative chat robot in a real multi-person response situation, wherein the portable device further includes a display element for synchronously displaying the feedback voice on the display element when the speaker outputs the feedback voice. On-demand chat messages. The method of claim 7 for a generative chat robot in a real multi-person response situation, wherein the sensor further includes sensing at least one of the user's physiological state, facial expression and body movements to generate a user behavior message , and the portable device transmits the user behavior information to the server host, and the server host determines the user's personality to set the personality parameters. The method of request 7 for a generative chat robot in a real multi-person response situation, wherein the portable device further includes converting the user's voice signal into the feature vector through a mel frequency cepstrum coefficient for transmission To the server host, The server host compares the user's personality with a plurality of preset personality feature vectors to determine the user's personality, and sets the personality parameters according to the determination results.

Images