System

A system comprising:

• circuitry configured to:
• acquire, via a communication interface coupled to a packet-switched network, an audio signal from an acoustic sensor, convert the audio signal into a digital signal, segment the digital signal into frames, apply a window function to each frame, and compute a short-time Fourier transform to obtain a complex spectrum;
• estimate a noise power spectrum from the complex spectrum, suppress a noise component by attenuating the noise power spectrum in each frequency bin, and generate a processed speech signal in which the noise component is suppressed and a speech component is emphasized;
• compute, for each frame, mel-frequency cepstral coefficients, log-mel filterbank energies, frame energy, spectral centroid, and fundamental frequency, and concatenate per-frame feature vectors into a time-series feature array;
• perform speech recognition processing on the processed speech signal using a neural acoustic model to compute a probability distribution over phonetic units and a beam search decoder to obtain a text sequence, extract a keyword based on a stored keyword group, and identify a speech section including the keyword;
• convert the identified speech section into a vibration signal and output the vibration signal to a bone-conduction actuator configured to produce mechanical vibrations that propagate through bone to an inner ear of a user;
• calculate a position of the user based on at least one of a satellite positioning signal, a wireless communication signal, or an output of an inertial sensor, determine a position state, and acquire environment data and service data from an external information source via the packet-switched network based on the position state;
• calculate a relevance value for each acquired data item by computing a proximity score, an urgency score, an emotion-adjusted relevance score, and a preference match score and combining the scores to produce a presentation priority value, and select presentation target data based on the presentation priority value;
• input the time-series feature array into an emotion estimation neural network comprising at least one recurrent or attention-based layer, compute scores for a plurality of predefined emotion categories, and select an emotion label;
• generate combined audio data by mixing audio output data derived from the presentation target data with the processed speech signal for output via the bone-conduction actuator, and adjust at least one of a volume level, a speaking speed, or a notification frequency of the audio output data based on the emotion label;
• and
• construct an input prompt comprising the position state, the extracted keyword, the emotion label, and the acquired data, transmit the input prompt to a generative language model, receive a response, and update at least one of the audio output data or a ranking condition based on the response.

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

• acquiring, via a communication interface coupled to a packet-switched network, an audio signal from an acoustic sensor, converting the audio signal into a digital signal, and storing the digital signal in a memory;
• segmenting the digital signal into time-domain and frequency-domain frames, performing short-time spectral analysis, and generating a processed speech signal in which a noise component is suppressed and a speech component is emphasized;
• detecting a speech section in the processed speech signal, performing speech recognition processing, extracting a keyword based on a stored keyword group, and identifying the speech section including the keyword;
• converting the identified speech section into a vibration signal and outputting the vibration signal to an actuator;
• calculating a position of a user, generating position history data, determining a position state, and acquiring environment data and service data from an external information source via the packet-switched network based on the position state;
• calculating a relevance value for each acquired data item, ranking and selecting presentation target data, and generating combined audio data by mixing audio output data with the processed speech signal;
• and
• constructing an input prompt comprising the position state, the extracted keyword, and the acquired data, transmitting the input prompt to a generative language model, receiving a response, and updating at least one of the audio output data or a ranking condition based on the response.