DSP Speech Enhancement (Restaurant Noise)
A DSP pipeline that mixes clean speech with real restaurant background noise, analyzes frequency content, applies FIR filtering, and evaluates improvement with spectrograms and SNR.
Project Summary
Goal
Reduce restaurant noise and improve speech clarity using classical DSP methods.
My contribution
End-to-end DSP pipeline, visualization, and evaluation.
Key techniques
FFT analysis, FIR low-pass filtering, spectrogram comparison, SNR measurement.
Evidence
All generated audio + plots are organized in GitHub for verification.
Problem
Restaurant environments contain overlapping noise sources (people, ambience, clatter). This makes speech harder to understand and challenges simple filtering.
Solution
Build a reproducible DSP workflow: create a noisy mixture, inspect frequency content, apply FIR filtering, then compare before/after using spectrograms and SNR.
Pipeline
- Load clean speech + restaurant background
- Mix signals to create a realistic noisy sample
- Analyze with FFT (find dominant noise bands)
- Apply FIR low-pass filter (reduce high-frequency noise)
- Compare with spectrograms + compute SNR improvement
Outcome
- Noise energy reduced in higher-frequency regions
- Spectrogram shows clearer speech structure after filtering
- Full outputs (audio + plots) are available on GitHub
Spectrogram Comparison
Clean vs Noisy vs Filtered — this is the main “proof” in the portfolio. (More plots + audio are in GitHub.)
