Speaker modules in an audio-visual real-time transcription system
AssigneeRTC Vision Ltd.
InventorsRon WEIN, Joseph RUBNER, Nir AVRAHAMI, Nir TSUK, Kfir GEDALYAHU
Life of this patent
Estimating the term…
Loading patent…
AssigneeRTC Vision Ltd.
InventorsRon WEIN, Joseph RUBNER, Nir AVRAHAMI, Nir TSUK, Kfir GEDALYAHU
Estimating the term…
The present invention provides a system and method for real-time target speaker recognition and transcription, enhancing accuracy in noisy and multi-speaker environments. Using audio and video inputs from a microphone and camera, the system processes features through neural networks, including cross-attention mechanisms, to isolate speaker-specific data and suppress noise. A Speaker Module tracks and identifies active speakers via face detection, recurrent neural networks, or decision mechanisms, ensuring transcription continuity even with temporary visual occlusions. Outputs generated by the invention include transcribed text and speaker labels, displayed on an augmented reality display or a mobile device as real-time subtitles. The system can suppress the display of the user's own speech for improved usability and supports multilingual transcription. This invention offers a robust solution for individuals with hearing impairments, enabling accurate comprehension in challenging acoustic settings.
A method of target speaker recognition and transcription of speech, comprising:
A computer system, comprising
Dependencies are inferred from the claim text.