US 12,444,406 B2
Inner speech iterative learning loop
Marcos Jimenez, Miami, FL (US); Meir Meshulam, Princeton, NJ (US); and Assif Ziv, Beit Yitzhak-Sha'ar Hefer (IL)
Assigned to Snap Inc., Santa Monica, CA (US)
Filed by Snap Inc., Santa Monica, CA (US)
Filed on Oct. 10, 2023, as Appl. No. 18/484,243.
Prior Publication US 2025/0118289 A1, Apr. 10, 2025
Int. Cl. G10L 15/06 (2013.01); G10L 15/02 (2006.01); G10L 15/16 (2006.01)
CPC G10L 15/063 (2013.01) [G10L 15/02 (2013.01); G10L 15/16 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A method comprising:
performing a first training iteration in which electromyograph (EMG) data corresponding to inner speech produced by a user for a set of specified phonemes, phoneme sounds, words or phrases is processed by a machine learning (ML) model to decode the EMG data into a set of predicted phonemes, phoneme sounds, words or phrases;
presenting the set of predicted phonemes, phoneme sounds, words or phrases to the user to conclude the first training iteration;
forming a first set of training data comprising the set of predicted phonemes, phoneme sounds, words or phrases, the EMG data, and the set of specified phonemes, phoneme sounds, words or phrases as ground truth information, the set of specified phonemes, phoneme sounds, words or phrases comprising a sequence of phonemes, phoneme sounds, words or phrases, forming the first set of training data comprising applying weights to portions of the EMG data corresponding to each of the set of specified phonemes, phoneme sounds, words or phrases, and a first portion of the EMG data corresponding to a later word or phrase in the sequence being associated with a higher weight than a second portion of the EMG data corresponding to an earlier word or phrase in the sequence; and
updating one or more parameters of the ML model based on the first set of training data prior to starting a second training iteration.