System

A system comprising:

• circuitry configured to:
• receive, via a communication interface operating over a packet-switched network, an audio signal representing a communication session, and preprocess the audio signal by performing noise reduction processing, reverberation suppression processing, and volume normalization processing;
• convert the preprocessed audio signal into text information by extracting acoustic features comprising log-mel filter bank features, inputting the acoustic features to a neural network comprising a convolutional front-end and transformer encoder layers, and decoding token sequences using beam search constrained by a language model;
• construct a first prompt sentence instructing a generative AI model to correct recognition errors and insert punctuation in the text information, input the first prompt sentence and the text information to the generative AI model, and obtain corrected text information;
• analyze the corrected text information by executing morphological analysis processing, syntactic analysis processing, sentiment analysis processing, and topic extraction processing to extract structured information comprising attribute information, and generate summary information on a basis of the structured information;
• determine a risk level by combining rule-based processing that evaluates the structured information and the summary information against predetermined determination conditions, and probability estimation processing that inputs a feature vector to a trained determination model to calculate a risk score, and mapping a combined result to a risk category;
• construct a second prompt sentence including the summary information, the attribute information, and the risk score, transmit the second prompt sentence to the generative AI model, and receive explanation information regarding the risk level;
• generate warning information comprising the explanation information and transmit the warning information to a user terminal via the packet-switched network, and transmit notification information comprising the summary information, the risk category, and recommended action information to a registered related-party terminal via the packet-switched network;
• and
• receive feedback information from the user terminal and update the trained determination model on a basis of the feedback information.

A method performed by circuitry of a system, the method comprising:

• receiving, via a communication interface coupled to a packet-switched network, an audio signal representing a communication session;
• converting the audio signal into text information;
• analyzing the text information to generate summary information of the communication session;
• determining a risk level by comparing the summary information with predetermined determination conditions;
• and
• generating warning information on a basis of the risk level and transmitting the warning information to at least one of a user terminal and a registered related-party terminal via the packet-switched network.