Cutoff frequency in one line
The cutoff frequency is the “border” where a filter starts reducing parts of the signal. Below the cutoff, we mostly keep the signal. Above it, we reduce it (for a low-pass filter).
Why it matters in speech
Speech is spread across frequencies. If your cutoff is too low, the voice becomes muffled because you removed important speech components. If your cutoff is too high, you keep too much noise.
The real trade-off
- Lower cutoff → stronger noise removal, but risk losing speech detail
- Higher cutoff → better speech detail, but more noise remains
How I chose the cutoff (practical)
- I inspected the noisy signal using FFT / spectrogram to see where noise dominates.
- I selected a cutoff that reduces the noisy band while keeping the core speech region.
- I validated the choice by comparing before/after spectrograms and SNR.
If a professor asks “Why 1.5 kHz?”
My answer: “I chose it to reduce higher-frequency restaurant noise while keeping the speech energy that matters most for clarity. Then I verified it with spectrogram comparison and SNR — not guesswork.”
One takeaway
Cutoff frequency is not a random number — it’s a decision that must be justified by frequency analysis and verified by results.