Team Communication Processing & Audio Enhancement

Built a research pipeline to evaluate team communication datasets and enhance noisy meeting audio using spectral gating, bandpass filtering, and voice activity detection for improved downstream speech analysis.

Overview

Developed an end-to-end audio processing workflow around the AMI Meeting Corpus to evaluate real-world multi-speaker conversations and improve recording quality. The project combines exploratory dataset analysis, signal processing techniques, and quantitative SNR evaluation to produce cleaner audio for Automatic Speech Recognition (ASR) and speaker diarization systems.

Features

Evaluated the AMI Meeting Corpus for multi-speaker team communication research
Performed exploratory audio analysis using waveforms, spectrograms, and baseline SNR estimation
Implemented spectral gating for stationary background noise reduction
Applied 300–3400 Hz bandpass filtering to isolate speech frequencies
Integrated Voice Activity Detection (VAD) to remove silent audio segments
Measured SNR improvements across each enhancement stage for objective evaluation

Tech Stack

Python: Audio processing pipeline
Jupyter Notebook: Experimentation and analysis
Librosa: Audio loading, visualization, feature extraction, and VAD
Noisereduce: Spectral gating noise suppression
SciPy: Digital bandpass filtering
SoundFile: Audio I/O
NumPy + Pandas: Signal processing and metrics
Matplotlib: Waveform and spectrogram visualization

nemo os

Archyve