System

A system comprising:

• circuitry configured to:
• store personal-related data in an internal information processing environment, extract subject-specific speech segments from acoustic data and video data included in the personal-related data, convert the speech segments into acoustic feature data, and train a speech synthesis model based on the acoustic feature data;
• extract an utterance history from character string data included in the personal-related data, generate dialog-history-inclusive training data pairs from the utterance history, and train a language generation model to imitate a conversational style of a subject based on the training data pairs;
• receive, via a communication interface coupled to a packet-switched network, a prompt sentence from a terminal device, construct an input sequence incorporating dialog history data and the prompt sentence, execute the language generation model to generate utterance text, and generate audio waveform data from the utterance text using the speech synthesis model;
• and
• extract phoneme timing data and intensity data from the audio waveform data, generate display motion parameter sequences comprising mouth-shape data and facial expression data synchronized with the audio waveform data, and transmit response data comprising the audio waveform data and the display motion parameter sequences to the terminal device.

A method performed by circuitry, the method comprising:

• storing personal-related data in an internal information processing environment, encrypting and managing the personal-related data, and constructing training datasets for generative model learning based on the personal-related data;
• extracting subject-specific speech segments from acoustic data and video data included in the personal-related data, converting the speech segments into acoustic feature data, and training or adapting a speech synthesis model based on the acoustic feature data so that the speech synthesis model is associated with a subject;
• extracting an utterance history of the subject from character string data included in the personal-related data, converting the utterance history into token sequences, generating training data associating dialog-history-inclusive input sequences with subject response sequences, and training or adapting a language generation model to imitate a conversational style of the subject based on the training data;
• receiving, via a communication interface coupled to a packet-switched network, a prompt sentence from a terminal device, constructing an input sequence for the language generation model based on dialog history data including the prompt sentence and session identification data, and executing the language generation model to generate utterance text;
• and
• normalizing the utterance text into a speech synthesis input sequence, generating audio waveform data based on speech characteristics of the subject using the speech synthesis model, extracting phoneme timing data and intensity data from the audio waveform data, generating display motion parameter sequences comprising mouth-shape data and facial expression data synchronized with the audio waveform data, and transmitting response data comprising the audio waveform data and the display motion parameter sequences to the terminal device.