WiMi Hologram Cloud Inc (WIMI)

[StockpickerCAPC]

WiMi Develops A Highly Reliable Silent Speech Recognition System.

Source
https://www.newstrail.com/wimi-develops-a-highly-reliable-silent-speech-recognition-system/

January 10, 2023

Language is the most common and direct way of human-to-human communication, and automatic speech recognition (ASR) is the most favored interaction method for future human-computer interaction. However, ASR also has limitations, such as degraded ground recognition in the presence of ambient noise, restricted privacy, and poor accessibility for patients with severe speech impairment, which prompt the need to use alternative non-acoustic modes of the human voice or silent speech recognition (SSR) technologies to assist. SSR converts electrical signals from facial and laryngeal vocal fold movements into speech signals and decodes what the user says through silent reading recognition without making a sound. Eye tracking, facial expression recognition, and brainwave control are all silent recognition technologies. SSR is a practical input in noisy environments, underwater or in space.

There are generally two ways of SSR. One is to install sensors to the terminal and judge the content of the user’s speech by sensing the airflow when the user speaks; the other is to analyze the facial sEMG (Surface Electromyography) signal characteristics when a human says by collecting the facial sEMG signal and correspond the sEMG signal to the text through the training of the neural network.

WiMi Hologram Cloud, Inc. (NASDAQ:WIMI) develops a highly reliable silent speech recognition system (WIMI SSR system) based on multi-feature signal sensing technology, which uses a multi-signal sensor and signal processing algorithm system to recognize silent words and phrases from surface sEMG signals involving facial and neck muscles. The WIMI SSR system recognizes silent words and phrases from sEMG by signal to recognize words, then recognizing word sequences using a grammar model, and finally recognizing untrained words using a phoneme model and transferring the untrained sEMG signal features to the deep learning module. The SSR algorithm is integrated with specially designed multi-featured signal sensors that can be arranged in flexible geometries to record high-fidelity sEMG signal measurements from the small articular muscles of the face and neck.

SEMG is one of many bioelectrical signals and is one of the relatively most accessible electrophysiological signals to acquire. By placing electrode pads on the skin surface of the human body, the weak potential difference generated by muscle contraction on the skin surface can be recorded, and this possible weak signal is amplified and converted by the EMG acquisition circuit to form a surface EMG signal that can be used for processing.

Pre-processing:
The acquired sEMG signal is processed by the software, denoised by the wavelet transform (WT) denoising method, resampled according to the processing performance, and the signal is analyzed and operated from the time domain, frequency domain, or time-frequency domain.

Feature extraction:
Used in combination with signal segmentation (window splitting), the moving window method is used to extract feature variables in a specific small window, which on the one hand can reduce the dimensionality, and on the other hand, the number of features can better reflect the characteristics of the segment relative to the original data, and the time-series signals often have context-dependent characteristics, so a certain overlap can better reflect the features. In the segmentation window to extract features, these features can be divided into time-domain segments, frequency-domain parts, and time-frequency-domain components, including deep learning methods CNN from a large amount of data to extract some “black box” features.

Feature classification:
The sEMG signal features are modeled with their corresponding categories by machine learning methods (including deep learning).

SSR technology ensures that users can express their intentions without making a sound, move their mouths in private without making an audible speech, and recognize the articulated information as visual content in noisy environments.

The successful application of WiMi’s multi-featured signal sensing technology’s highly reliable silent speech recognition system (WIMI SSR system) not only helps people with hearing and speech impairments but also applies to scenarios including disaster sites, extra-vehicular exploration, underwater operations, and factory floors. In the future, let the machine understand human language.
Relying only on sound information is not enough, “sound, light, light, heat, magnetic” and other physical perception technology must be integrated to use, so that the machine, beyond the five human senses, can see what humans can not see, hear what humans can not hear. Let the device hear far beyond human perception, which is not only the progress of the algorithm but also the need for the entire industry chain of standard technology upgrades, including more advanced sensors and more robust arithmetic chip.