Researchers at the Saint Petersburg Federal Research Center of the Russian Academy of Sciences have unveiled a mobile application that can read a person’s speech by analyzing lip movements in loud environments. The development was reported by TASS, with confirmation from the center’s press service.
According to the press service, scientists have taught machines to interpret human speech from visual cues at the lips using advanced artificial intelligence algorithms and computer vision techniques. This approach blends facial motion information with audio signals to achieve more reliable recognition when background noise is high.
The spokesperson noted that the improvement could boost the accuracy of voice assistants in crowded settings or when heavy machinery or traffic masks spoken words. The dual data collection method mirrors human perception by capturing both the acoustic form of speech and the corresponding lip movements, then fusing them to enhance understanding.
Scientists explained that the neural network is designed to automatically select the data streams that yield the highest accuracy for any given situation. In essence, the system weighs audio and visual evidence to decide when lip-reading input will most improve comprehension, especially in noisy conditions.
During tests conducted with noisy truck drivers from a logistics company in Russia, recognition accuracy surpassed 90 percent, demonstrating the potential value of integrating visual speech cues into real-time recognition pipelines in demanding environments.
Earlier reports mentioned a related effort by the Main Radio Frequency Center, an entity connected with Roskomnadzor, to develop an AI-based service capable of analyzing video recordings of conversations for lies and manipulations. This line of work reflects a broader interest in AI-assisted media verification and speech interpretation across regulated sectors [Source: SPb FRC RAS press release].