Granted Patent B2

US 12,592,218 · App. 18/178,563 · Granted Mar 31, 2026

Inventors: Hideo Tsuru (Yokohama, JP); Motomu Takada (Yokohama, JP); Hideya Tsujii (Yokohama, JP)

Assignee: JVCKENWOOD Corporation

G10L15/02G10L15/24G10L2015/025

Abstract

Claims (49)

1 . A communication device comprising:

a vibration transmitting unit configured to transmit an input vibration wave to a first portion of a subject;

a vibration receiving unit configured to receive, at a second portion of the subject, an output vibration wave generated based on the input vibration wave propagated through at least a part of the subject; and

a speech recognition device configured to recognize a phoneme uttered by the subject based on a difference wave between the input vibration wave and the output vibration wave, wherein

the first portion and the second portion are arranged on right and left of the subject at bilaterally symmetrical positions with respect to a center line of a body of the subject,

the first portion is designated to an articular disc of a temporomandibular joint of a right side, and

the second portion is designated to an articular disc of a temporomandibular joint of a left side,

wherein the speech recognition device includes:

a difference-wave generating unit configured to calculate the difference wave;

an orthogonal transformer configured to perform orthogonal transformation on the difference wave to calculate a frequency spectrum of the difference wave;

a feature-amount extracting unit configured to extract a feature-amount of the difference wave; and

a processor configured to identify a phoneme uttered by the subject based on the feature-amount extracted by the feature-amount extracting unit, and

further comprising a learned-model storage unit configured to:

acquire teacher data that indicates a relationship between the phoneme uttered by the subject and the frequency spectrum of the difference wave;

generate, by performing machine learning based on the teacher data, a learned model into which the frequency spectrum is input and from which a phoneme which is uttered by the subject is output; and

store the learned model, wherein

the processor is further configured to identify the phoneme by inputting the feature-amount extracted by the feature-amount extracting unit into the learned model.

2 . A communication method comprising:

transmitting an input vibration wave to a first portion of a subject;

receiving, at a second portion of the subject, an output vibration wave generated based on the input vibration wave propagated through at least a part of the subject; and

recognizing a phoneme uttered by the subject based on a difference wave between the input vibration wave and the output vibration wave, wherein

the first portion and the second portion are arranged on right and left of the subject at bilaterally symmetrical positions with respect to a center line of a body of the subject,

the first portion is designated to an articular disc of a temporomandibular joint of a right side, and

the second portion is designated to an articular disc of a temporomandibular joint of a left side,

and the communication method further comprising:

calculating the difference wave;

performing orthogonal transformation on the difference wave to calculate a frequency spectrum of the difference wave;

extracting a feature-amount of the difference wave;

identifying a phoneme uttered by the subject based on the feature-amount extracted;

acquiring teacher data that indicates a relationship between the phoneme uttered by the subject and the frequency spectrum of the difference wave;

generating, by performing machine learning based on the teacher data, a learned model into which the frequency spectrum is input and from which a phoneme which is uttered by the subject is output;

storing the learned model; and

identifying the phoneme by inputting the feature-amount extracted into the learned model.

3 . A non-transitory storage medium that stores a computer-program for causing a computer to execute a communication method including:

transmitting an input vibration wave to a first portion of a subject;

receiving, at a second portion of the subject, an output vibration wave generated based on the input vibration wave propagated through at least a part of the subject; and

recognizing a phoneme uttered by the subject based on a difference wave between the input vibration wave and the output vibration wave, wherein

the first portion and the second portion are arranged on a right side and a left side of the subject at bilaterally symmetrical positions with respect to a center line of a body of the subject,

the first portion is designated to an articular disc of a temporomandibular joint of a right side, and

the second portion is designated to an articular disc of a temporomandibular joint of a left side,

and the communication method further including:

calculating the difference wave;

performing orthogonal transformation on the difference wave to calculate a frequency spectrum of the difference wave;

extracting a feature-amount of the difference wave;

identifying a phoneme uttered by the subject based on the feature-amount extracted;

acquiring teacher data that indicates a relationship between the phoneme uttered by the subject and the frequency spectrum of the difference wave;

generating, by performing machine learning based on the teacher data, a learned model into which the frequency spectrum is input and from which a phoneme which is uttered by the subject is output;

storing the learned model; and

identifying the phoneme by inputting the feature-amount extracted into the learned model.

Assignments (2)

Priority Claims (1)

Continuity (2)

References Cited (73)