Abstract
By a News Reporter-Staff News Editor at Robotics & Machine Learning Daily News Daily News-Investigators publish new report on ro botics. According to news originating from Kyoto, Japan, by NewsRx correspondent s, research stated, "This study explores the vulnerability of robot dialogue sys tems' automatic speech recognition (ASR) module to adversarial music attacks." The news correspondents obtained a quote from the research from National Institu te of Information and Communications Technology: "Specifically, we explore music as a natural camouflage for such attacks. We propose a novel method to hide gho st speech commands in a music clip by slightly perturbing its raw waveform. We a pply our attack on an industry-popular ASR model, namely the time-delay neural n etwork (TDNN), widely used for speech and speaker recognition. Our experiment de monstrates that adversarial music crafted by our attack can easily mislead indus try-level TDNN models into picking up ghost commands with high success rates. Ho wever, it sounds no different from the original music to the human ear."
NETL
NSTL