Always-on microphones in commodity devices—MEMS capsules in phones, laptops, and smart speakers—expose room conversations to covert capture. Because confidentiality is lost once speech is recorded, this thesis investigates active acoustic protection: degrading intelligibility at capture time while keeping in-room speech comfortable. Most ultrasonic jammers radiate broadband noise above 20 kHz and rely on front-end nonlinearities to fold interference into the audible band. We therefore design PhonemJam- mer, a speech-aware jammer that emits phoneme-guided pseudo-speech instead of stochastic noise, aiming to interfere with the linguistic cues that enhancers and ASR systems try to preserve. PhonemJammer is inspired by phoneme-based ultrasonic jammers such as InfoMasker and its journal extension, but explores a different design point: a compact three-transducer cluster (three co-located emitters driven in phase), synthetic TTS logatoms (no recorded speech segments), and no cooperative content-recovery path.