Acoustic signal identification in an audio track

Seng, Wei Xiang (2025) Acoustic signal identification in an audio track. Final Year Project, UTAR.

PDF Download (2126Kb)

Abstract

This project focuses on the identification and transcription of acoustic signals within audio tracks, specifically targeting multi-speaker English audio files without background noise or overlapping speech. The primary objective is to develop a standalone, local software program using Python that can reliably identify different speakers and produce transcriptions that are credited to each one without the need for an internet connection. The system employs techniques for audio signal processing, speaker diarization for segmenting the audio stream based on speaker identity, and automatic speech recognition for transcribing the spoken content, all implemented using local models and libraries. The methodology involves processing the input audio locally, applying speaker diarization to detect speaker changes and segment the audio, and subsequently transcribing each segment while associating it with the corresponding speaker, ensuring full offline operation. This project contributes to the field of audio analysis by creating a self-contained, offline-capable tool for speaker-aware acoustic signal processing and transcription in controlled environments, demonstrating the practical application of Python-based audio processing and machine learning tools that function independently of cloud services. The final output is a functional offline Python application capable of identifying speakers and generating speaker-labelled transcriptions for the specified audio constraints.

Actions (login required)