As the utilization of voice assistants becomes more widespread in daily activities, the demand for an interface capable of accurately recognizing speech at low volume levels within noisy environments is becoming increasingly important. In this study, we developed a custom device that incorporates a bone conduction microphone (BCM) by integrating a piezo-electric transducer with a noise-isolating impedance matching layer, alongside a microelectromechanical systems (MEMS) air conduction microphone (ACM). Our study aims to assess the BCM’s effectiveness in facilitating soft speech communication with voice assistants in diverse noise environments and to validate its advantage over the ACM in noise reduction. We conducted experiments with participants using both the ACM and BCM to record audio samples for normal speech, soft speech, and whisper in three different noise level environments: ambient (quiet office setting), music, and loud noise. Spectrogram and signal-to-noise ratio (SNR) analysis assessed each audio file’s signal quality. Additionally, we utilized an automatic speech recognition (ASR) model to estimate word error rate (WER) as a benchmark for performance comparison. In all tested scenarios, the BCM consistently demonstrated a significantly higher SNR performance compared to the ACM, as evident in the BCM’s more distinguishable spectrograms for speech patterns in noisy environments. While the ACM’s WER scores in the ambient environment were lower in all speech modes, the BCM outperformed the ACM in noisy conditions. Our findings confirm the custom BCM’s ability to reduce background noise without requiring pre-processing or complex hardware while effectively capturing soft speech, emphasizing the potential benefits of integrating BCMs into interfaces designed for communication with voice assistants.