Yingxi Chen
Pronouns: She/Her
Research Mentor(s): Alanson Sample
Research Mentor School/College/Department: Computer Science and Engineering / Engineering
Program:
Authors: Yingxi Chen, Yasha Iravantchi, Alanson Sample
Session: Session 1: 9:00 am – 9:50 am
Poster: 26
Abstract
In a world increasingly concerned with acoustic privacy and efficient interaction, our research aims to address the technological gap by enhancing sound localization and processing capabilities. To this end, We transitioned the PrivacyMic system from a single-channel to a multi-channel setup for precise sound localization. Functions, including filtering, callback, and clip, were rigorously tested across an 8-core processor using set affinity, identifying any memory constraints. We integrated the Python library setproctitle for better process management and employed a useful-transformers library to incorporate a fast Whisper model, focusing on accurate wake word detection and real-time transcription. Our refined system now adeptly identifies the “hey puffin” wake word, performs real-time transcription, and maintains heightened detection accuracy through VAD and appropriate prompting. This research bridges significant gaps in acoustic privacy and interaction, presenting a robust framework for future sound data collection and machine learning model development, ultimately enhancing both user privacy and interaction quality.
