Since the 1970s, scientists have been searching for ways to communicate with people in lucid dreams (LDs), during which it is possible to maintain consciousness. Previously, dreamers could hear sounds from reality and respond with some simple signals, but they could not speak back. In this study, facial surface electromyography (EMG) was tested as a proof of concept for unmuting people in LDs. Remmyo, an EMG distinctive constructed language, was used. The software was developed to translate facial EMG impulses into Remmyo sounds and letters, translate words into English, and digitally vocalize the final text in English. Four LD practitioners were trained to pronounce a short phrase or a word in Remmyo and were then asked to achieve the same task in LDs under polysomnographic observation. LDs were verified by preagreed eye movements in rapid eye movement (REM) sleep. Four volunteers tried to speak in Remmyo in 15 LDs. Due to software failures, mispronunciations, and missing sounds, the decoding efficiency in real time or in recordings ranged from 13% to 81%. The first phrase and word heard from sleeping people were "no war" and "freedom." The later was automatically translated and vocalized in English in real time for 11 times. Despite controversial results, the study shows that, with further development, people could possibly talk in LDs and could be heard in reality with the help of EMG sensors. To achieve this goal, a range of possible obstacles is discussed. This technology could provide opportunities for LD studies and their practical applications.